Ghci #11

roiser · 2025-11-23T19:29:38Z

No description provided.

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-14T11:07:15Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.
See attached patch for details:

github-actions · 2026-01-14T14:00:42Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.
See attached patch for details:
--- src/mgOnGpuConfig.h 2026-01-14 14:00:34.435664702 +0000
+++ - 2026-01-14 14:00:36.427511728 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT

typedef float fptype; // single precision (4 bytes, fp32)

typedef float fptype; // single precision (4 bytes, fp32)
#endif

// Floating point type (for color algebra alone Single precision color algebra madgraph5/madgraph4gpu#537): fptype2

--- src/mgOnGpuVectors.h 2026-01-14 14:00:34.357664656 +0000
+++ - 2026-01-14 14:00:36.620091862 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
typedef long int bool_v attribute( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT

typedef int bool_v attribute( ( ext_vector_type( neppV ) ) ); // bbbb

typedef int bool_v attribute( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
typedef unsigned int uint_v attribute( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc 2026-01-14 14:00:34.435664702 +0000
+++ - 2026-01-14 14:00:37.369229554 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
void

color_sum_gpu( fptype* ghelAllMEs, // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)

            const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities

            fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities

            gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle

            gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)

            const int nGoodHel,               // input: number of good helicities

            const int gpublocks,              // input: cuda gpublocks

            const int gputhreads )            // input: cuda gputhreads

color_sum_gpu( fptype* ghelAllMEs, // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)

            const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities

            fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities

            gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle

            gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)

            const int nGoodHel,           // input: number of good helicities

            const int gpublocks,          // input: cuda gpublocks

            const int gputhreads )        // input: cuda gputhreads

{
const int nevt = gpublocks * gputhreads;
// CASE 1: KERNEL
if( !pBlasHandle )
{

 assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set

 assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
 // Loop over helicities
 for( int ighel = 0; ighel < nGoodHel; ighel++ )
 {

@@ -409,13 +409,13 @@
assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed

```
 // Reset the tmp buffer
```

                                     // Reset the tmp buffer

#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif

 // Delegate the color sum to BLAS for

 // Delegate the color sum to BLAS for
 color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );

#endif
}

--- SubProcesses/P1_uux_ttx/CPPProcess.cc 2026-01-14 14:00:34.434664701 +0000
+++ - 2026-01-14 14:00:37.444504349 +0000
@@ -735,11 +735,11 @@
const fptype* allcouplings, // input: couplings[nevtndcoup2]
fptype* allMEs, // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL

                  fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities

                  fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities

                  fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities

                  fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities

#endif

                  bool* isGoodHel,            // output: isGoodHel[ncomb] - host array

                  const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)

                  bool* isGoodHel, // output: isGoodHel[ncomb] - host array

```
                  const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
```
{
//assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
//assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
}
constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering

   //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;

                                            //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;

#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else

   cxtype_sv jamp_sv[ncolor] = {};  // all zeros

   cxtype_sv jamp_sv[ncolor] = {}; // all zeros

#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off /
constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
assert( npagV % 2 == 0 ); // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2neppV-page
const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else

const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time

const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
// OMP multithreading Retry OMP multithreading in cudacpp (and prototype custom multithreading, and compare to MP) - suboptimal results in ggttgg (Dec 2022) madgraph5/madgraph4gpu#575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
// Running sum of partial amplitudes squared for event by event color selection (Randomly pick one color for each event written to file (jamp2 handling) madgraph5/madgraph4gpu#402)
// (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
fptype_sv jamp2_sv[nParity * ncolor] = {};

 fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)

 fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)

#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc 2026-01-14 14:00:34.434664701 +0000
+++ - 2026-01-14 14:00:37.768768570 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else

RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU

RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
// Bridge emulation mode (NB Bridge implies RamboHost!)
bool bridge = false;
@@ -918,7 +918,7 @@
wrkflwtxt += "/sse4";
#endif
#else

wrkflwtxt += "/????"; // no path to this statement

wrkflwtxt += "/????"; // no path to this statement
#endif
// -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
<< ""STD::COMPLEX"," << std::endl
#else

        << "\"???\"," << std::endl                           // no path to this statement...

        << "\"???\"," << std::endl // no path to this statement...

#endif
<< ""RanNumb memory layout": "
<< ""AOSOA[" << neppR << "]""

--- SubProcesses/P1_gg_ttx/color_sum.cc 2026-01-14 14:00:34.387664674 +0000
+++ - 2026-01-14 14:00:38.172421262 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
void

color_sum_gpu( fptype* ghelAllMEs, // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)

            const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities

            fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities

            gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle

            gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)

            const int nGoodHel,               // input: number of good helicities

            const int gpublocks,              // input: cuda gpublocks

            const int gputhreads )            // input: cuda gputhreads

color_sum_gpu( fptype* ghelAllMEs, // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)

            const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities

            fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities

            gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle

            gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)

            const int nGoodHel,           // input: number of good helicities

            const int gpublocks,          // input: cuda gpublocks

            const int gputhreads )        // input: cuda gputhreads

{
const int nevt = gpublocks * gputhreads;
// CASE 1: KERNEL
if( !pBlasHandle )
{

 assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set

 assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
 // Loop over helicities
 for( int ighel = 0; ighel < nGoodHel; ighel++ )
 {

@@ -409,13 +409,13 @@
assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed

```
 // Reset the tmp buffer
```

                                     // Reset the tmp buffer

#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif

 // Delegate the color sum to BLAS for

 // Delegate the color sum to BLAS for
 color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );

#endif
}

--- SubProcesses/P1_gg_ttx/CPPProcess.cc 2026-01-14 14:00:34.386664673 +0000
+++ - 2026-01-14 14:00:38.249067813 +0000
@@ -758,11 +758,11 @@
const fptype* allcouplings, // input: couplings[nevtndcoup2]
fptype* allMEs, // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL

                  fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities

                  fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities

                  fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities

                  fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities

#endif

                  bool* isGoodHel,            // output: isGoodHel[ncomb] - host array

                  const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)

                  bool* isGoodHel, // output: isGoodHel[ncomb] - host array

```
                  const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
```
{
//assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
//assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
}
constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering

   //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;

                                            //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;

#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else

   cxtype_sv jamp_sv[ncolor] = {};  // all zeros

   cxtype_sv jamp_sv[ncolor] = {}; // all zeros

#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off /
constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
assert( npagV % 2 == 0 ); // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2neppV-page
const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else

const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time

const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
// OMP multithreading Retry OMP multithreading in cudacpp (and prototype custom multithreading, and compare to MP) - suboptimal results in ggttgg (Dec 2022) madgraph5/madgraph4gpu#575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
// Running sum of partial amplitudes squared for event by event color selection (Randomly pick one color for each event written to file (jamp2 handling) madgraph5/madgraph4gpu#402)
// (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
fptype_sv jamp2_sv[nParity * ncolor] = {};

 fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)

 fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)

#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc 2026-01-14 14:00:34.387664674 +0000
+++ - 2026-01-14 14:00:38.573087912 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else

RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU

RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
// Bridge emulation mode (NB Bridge implies RamboHost!)
bool bridge = false;
@@ -918,7 +918,7 @@
wrkflwtxt += "/sse4";
#endif
#else

wrkflwtxt += "/????"; // no path to this statement

wrkflwtxt += "/????"; // no path to this statement
#endif
// -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
<< ""STD::COMPLEX"," << std::endl
#else

        << "\"???\"," << std::endl                           // no path to this statement...

        << "\"???\"," << std::endl // no path to this statement...

#endif
<< ""RanNumb memory layout": "
<< ""AOSOA[" << neppR << "]""

github-actions · 2026-01-14T14:07:32Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.
See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:07:20.545692722 +0000
+++ -	2026-01-14 14:07:24.334005860 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:07:20.468692884 +0000
+++ -	2026-01-14 14:07:24.527893709 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:07:20.545692722 +0000
+++ -	2026-01-14 14:07:25.256162896 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:07:20.544692724 +0000
+++ -	2026-01-14 14:07:25.335112782 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:07:20.544692724 +0000
+++ -	2026-01-14 14:07:25.660834524 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:07:20.498692821 +0000
+++ -	2026-01-14 14:07:26.063950125 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:07:20.497692823 +0000
+++ -	2026-01-14 14:07:26.141225849 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:07:20.498692821 +0000
+++ -	2026-01-14 14:07:26.465730473 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-14T14:41:53Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend avx2.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:45.916186263 +0000
+++ -	2026-01-14 14:41:47.564286723 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:45.841187085 +0000
+++ -	2026-01-14 14:41:47.750103864 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:41:45.916186263 +0000
+++ -	2026-01-14 14:41:48.387612967 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:41:45.915186274 +0000
+++ -	2026-01-14 14:41:48.458200431 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:41:45.916186263 +0000
+++ -	2026-01-14 14:41:48.774384174 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:41:45.870186767 +0000
+++ -	2026-01-14 14:41:49.157288966 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:41:45.869186778 +0000
+++ -	2026-01-14 14:41:49.229550576 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:41:45.869186778 +0000
+++ -	2026-01-14 14:41:49.545385979 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:54Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend avx512.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:46.457628633 +0000
+++ -	2026-01-14 14:41:48.681624263 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:46.383628628 +0000
+++ -	2026-01-14 14:41:48.869405649 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:41:46.458628634 +0000
+++ -	2026-01-14 14:41:49.557764425 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:41:46.456628633 +0000
+++ -	2026-01-14 14:41:49.632977531 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:41:46.457628633 +0000
+++ -	2026-01-14 14:41:49.952676473 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:41:46.412628630 +0000
+++ -	2026-01-14 14:41:50.343310435 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:41:46.411628630 +0000
+++ -	2026-01-14 14:41:50.417472321 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:41:46.411628630 +0000
+++ -	2026-01-14 14:41:50.740263303 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:54Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
14 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttxj with backend avx2.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:45.641644678 +0000
+++ -	2026-01-14 14:41:47.716162589 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:45.390640573 +0000
+++ -	2026-01-14 14:41:47.909269564 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_gu_ttxu/color_sum.cc	2026-01-14 14:41:45.539643010 +0000
+++ -	2026-01-14 14:41:48.461525595 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gu_ttxu/CPPProcess.cc	2026-01-14 14:41:45.538642994 +0000
+++ -	2026-01-14 14:41:48.539176086 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gu_ttxu/check_sa.cc	2026-01-14 14:41:45.538642994 +0000
+++ -	2026-01-14 14:41:48.863849626 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_uux_ttxg/color_sum.cc	2026-01-14 14:41:45.642644694 +0000
+++ -	2026-01-14 14:41:49.285530406 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttxg/CPPProcess.cc	2026-01-14 14:41:45.641644678 +0000
+++ -	2026-01-14 14:41:49.365265783 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttxg/check_sa.cc	2026-01-14 14:41:45.641644678 +0000
+++ -	2026-01-14 14:41:49.687296390 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gux_ttxux/color_sum.cc	2026-01-14 14:41:45.591643860 +0000
+++ -	2026-01-14 14:41:52.745951450 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gux_ttxux/CPPProcess.cc	2026-01-14 14:41:45.590643844 +0000
+++ -	2026-01-14 14:41:52.823911749 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gux_ttxux/check_sa.cc	2026-01-14 14:41:45.590643844 +0000
+++ -	2026-01-14 14:41:53.148642413 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttxg/color_sum.cc	2026-01-14 14:41:45.472641914 +0000
+++ -	2026-01-14 14:41:53.603699141 +0000
@@ -383,20 +383,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -413,13 +413,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttxg/CPPProcess.cc	2026-01-14 14:41:45.471641898 +0000
+++ -	2026-01-14 14:41:53.689442459 +0000
@@ -975,11 +975,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -1020,11 +1020,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1361,7 +1361,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1424,7 +1424,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttxg/check_sa.cc	2026-01-14 14:41:45.471641898 +0000
+++ -	2026-01-14 14:41:54.022920505 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:55Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend cuda.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:46.275649385 +0000
+++ -	2026-01-14 14:41:49.806868900 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:46.199649129 +0000
+++ -	2026-01-14 14:41:49.999500392 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:41:46.275649385 +0000
+++ -	2026-01-14 14:41:50.716907218 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:41:46.275649385 +0000
+++ -	2026-01-14 14:41:50.792411347 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:41:46.275649385 +0000
+++ -	2026-01-14 14:41:51.115680348 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:41:46.229649230 +0000
+++ -	2026-01-14 14:41:51.518864936 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:41:46.228649226 +0000
+++ -	2026-01-14 14:41:51.596814709 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:41:46.228649226 +0000
+++ -	2026-01-14 14:41:51.923222061 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:55Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
14 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttxj with backend avx512.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:45.457122020 +0000
+++ -	2026-01-14 14:41:48.867643356 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:45.205123714 +0000
+++ -	2026-01-14 14:41:49.053673132 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_gu_ttxu/color_sum.cc	2026-01-14 14:41:45.355122706 +0000
+++ -	2026-01-14 14:41:49.595211719 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gu_ttxu/CPPProcess.cc	2026-01-14 14:41:45.354122712 +0000
+++ -	2026-01-14 14:41:49.671154815 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gu_ttxu/check_sa.cc	2026-01-14 14:41:45.354122712 +0000
+++ -	2026-01-14 14:41:49.993347181 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_uux_ttxg/color_sum.cc	2026-01-14 14:41:45.457122020 +0000
+++ -	2026-01-14 14:41:50.410946597 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttxg/CPPProcess.cc	2026-01-14 14:41:45.456122027 +0000
+++ -	2026-01-14 14:41:50.487291823 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttxg/check_sa.cc	2026-01-14 14:41:45.456122027 +0000
+++ -	2026-01-14 14:41:50.807240488 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gux_ttxux/color_sum.cc	2026-01-14 14:41:45.406122363 +0000
+++ -	2026-01-14 14:41:53.836555139 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gux_ttxux/CPPProcess.cc	2026-01-14 14:41:45.405122370 +0000
+++ -	2026-01-14 14:41:53.914077071 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gux_ttxux/check_sa.cc	2026-01-14 14:41:45.406122363 +0000
+++ -	2026-01-14 14:41:54.233630187 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttxg/color_sum.cc	2026-01-14 14:41:45.288123156 +0000
+++ -	2026-01-14 14:41:54.684000977 +0000
@@ -383,20 +383,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -413,13 +413,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttxg/CPPProcess.cc	2026-01-14 14:41:45.287123163 +0000
+++ -	2026-01-14 14:41:54.769152262 +0000
@@ -975,11 +975,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -1020,11 +1020,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1361,7 +1361,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1424,7 +1424,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttxg/check_sa.cc	2026-01-14 14:41:45.287123163 +0000
+++ -	2026-01-14 14:41:55.097979057 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:56Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:48.436766918 +0000
+++ -	2026-01-14 14:41:50.217650410 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:48.351766523 +0000
+++ -	2026-01-14 14:41:50.409747751 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:41:48.436766918 +0000
+++ -	2026-01-14 14:41:51.094459859 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:41:48.435766913 +0000
+++ -	2026-01-14 14:41:51.168911526 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:41:48.435766913 +0000
+++ -	2026-01-14 14:41:51.487867143 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:41:48.381766662 +0000
+++ -	2026-01-14 14:41:51.879278498 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:41:48.380766658 +0000
+++ -	2026-01-14 14:41:51.953198464 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:41:48.381766662 +0000
+++ -	2026-01-14 14:41:52.275690747 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-14T14:41:56Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend hip.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-14 14:41:48.546753444 +0000
+++ -	2026-01-14 14:41:50.537749405 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-14 14:41:48.470752483 +0000
+++ -	2026-01-14 14:41:50.717310388 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-14 14:41:48.546753444 +0000
+++ -	2026-01-14 14:41:51.362661886 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-14 14:41:48.544753419 +0000
+++ -	2026-01-14 14:41:51.433884888 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-14 14:41:48.545753432 +0000
+++ -	2026-01-14 14:41:51.748913363 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-14 14:41:48.500752863 +0000
+++ -	2026-01-14 14:41:52.134106287 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-14 14:41:48.499752850 +0000
+++ -	2026-01-14 14:41:52.206247236 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-14 14:41:48.499752850 +0000
+++ -	2026-01-14 14:41:52.526936543 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-16T07:40:20Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
14 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttxj with backend avx2.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 07:40:11.042691828 +0000
+++ -	2026-01-16 07:40:13.230073713 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 07:40:10.792694367 +0000
+++ -	2026-01-16 07:40:13.412043219 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_gu_ttxu/color_sum.cc	2026-01-16 07:40:10.941692853 +0000
+++ -	2026-01-16 07:40:13.940714700 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gu_ttxu/CPPProcess.cc	2026-01-16 07:40:10.940692863 +0000
+++ -	2026-01-16 07:40:14.016634357 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gu_ttxu/check_sa.cc	2026-01-16 07:40:10.940692863 +0000
+++ -	2026-01-16 07:40:14.336672835 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_uux_ttxg/color_sum.cc	2026-01-16 07:40:11.042691828 +0000
+++ -	2026-01-16 07:40:14.756554852 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttxg/CPPProcess.cc	2026-01-16 07:40:11.041691838 +0000
+++ -	2026-01-16 07:40:14.831973207 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttxg/check_sa.cc	2026-01-16 07:40:11.042691828 +0000
+++ -	2026-01-16 07:40:15.149994953 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gux_ttxux/color_sum.cc	2026-01-16 07:40:10.992692335 +0000
+++ -	2026-01-16 07:40:18.151655352 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gux_ttxux/CPPProcess.cc	2026-01-16 07:40:10.991692345 +0000
+++ -	2026-01-16 07:40:18.227686414 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gux_ttxux/check_sa.cc	2026-01-16 07:40:10.991692345 +0000
+++ -	2026-01-16 07:40:18.549201393 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttxg/color_sum.cc	2026-01-16 07:40:10.874693534 +0000
+++ -	2026-01-16 07:40:18.994676426 +0000
@@ -383,20 +383,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -413,13 +413,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttxg/CPPProcess.cc	2026-01-16 07:40:10.874693534 +0000
+++ -	2026-01-16 07:40:19.079614155 +0000
@@ -975,11 +975,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -1020,11 +1020,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1361,7 +1361,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1424,7 +1424,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttxg/check_sa.cc	2026-01-16 07:40:10.874693534 +0000
+++ -	2026-01-16 07:40:19.406590541 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-16T10:04:10Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 10:03:59.599484267 +0000
+++ -	2026-01-16 10:04:01.433261574 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 10:03:59.524483540 +0000
+++ -	2026-01-16 10:04:01.637104098 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 10:03:59.599484267 +0000
+++ -	2026-01-16 10:04:02.387552635 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 10:03:59.598484258 +0000
+++ -	2026-01-16 10:04:02.467179013 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 10:03:59.598484258 +0000
+++ -	2026-01-16 10:04:02.803298297 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 10:03:59.553483821 +0000
+++ -	2026-01-16 10:04:03.213216907 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 10:03:59.552483812 +0000
+++ -	2026-01-16 10:04:03.294556310 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 10:03:59.552483812 +0000
+++ -	2026-01-16 10:04:03.631829275 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

change branch pwd Enhance hello world step with directory listing new file

fix process move ci file update submodule

pwd pwd pwd mg5 fetch submodules disable all processes, generate first process fix remote repo fix typo generate one process

move to PAT empty add pwd add git status spell out repo fix repo name more debug go one level up copy files over

remove persist credentials change pat name add ls remove if, single line commit simplify ci pat in local repo use PAT as a token

generate access token run only one config fix typo indentation error simplify token gen remove installation-id add repo name test github app

github-actions · 2026-01-16T10:28:29Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 10:28:17.540172151 +0000
+++ -	2026-01-16 10:28:20.789418786 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 10:28:17.466171972 +0000
+++ -	2026-01-16 10:28:20.980450658 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 10:28:17.540172151 +0000
+++ -	2026-01-16 10:28:21.685213159 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 10:28:17.539172149 +0000
+++ -	2026-01-16 10:28:21.759729666 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 10:28:17.540172151 +0000
+++ -	2026-01-16 10:28:22.080799084 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 10:28:17.496172045 +0000
+++ -	2026-01-16 10:28:22.489688747 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 10:28:17.495172042 +0000
+++ -	2026-01-16 10:28:22.568377254 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 10:28:17.495172042 +0000
+++ -	2026-01-16 10:28:22.890636365 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-16T12:35:48Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 12:35:38.839789976 +0000
+++ -	2026-01-16 12:35:41.974092126 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 12:35:38.778790250 +0000
+++ -	2026-01-16 12:35:42.146377848 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 12:35:38.839789976 +0000
+++ -	2026-01-16 12:35:42.747892692 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 12:35:38.838789980 +0000
+++ -	2026-01-16 12:35:42.813633905 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 12:35:38.838789980 +0000
+++ -	2026-01-16 12:35:43.122161615 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 12:35:38.802790142 +0000
+++ -	2026-01-16 12:35:43.492144652 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 12:35:38.801790147 +0000
+++ -	2026-01-16 12:35:43.559318714 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 12:35:38.801790147 +0000
+++ -	2026-01-16 12:35:43.870046936 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-16T14:56:18Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 14:56:04.449059974 +0000
+++ -	2026-01-16 14:56:08.371480458 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 14:56:04.376059718 +0000
+++ -	2026-01-16 14:56:08.558863598 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 14:56:04.450059978 +0000
+++ -	2026-01-16 14:56:09.219962529 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 14:56:04.449059974 +0000
+++ -	2026-01-16 14:56:09.292064852 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 14:56:04.449059974 +0000
+++ -	2026-01-16 14:56:09.613278863 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 14:56:04.405059819 +0000
+++ -	2026-01-16 14:56:10.002535531 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 14:56:04.404059816 +0000
+++ -	2026-01-16 14:56:10.077907912 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 14:56:04.404059816 +0000
+++ -	2026-01-16 14:56:10.402999294 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-16T15:19:35Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend scalar.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 15:19:21.155942275 +0000
+++ -	2026-01-16 15:19:25.520377352 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 15:19:21.079942265 +0000
+++ -	2026-01-16 15:19:25.702449783 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 15:19:21.155942275 +0000
+++ -	2026-01-16 15:19:26.356949865 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 15:19:21.154942275 +0000
+++ -	2026-01-16 15:19:26.427282526 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 15:19:21.154942275 +0000
+++ -	2026-01-16 15:19:26.743107522 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 15:19:21.109942269 +0000
+++ -	2026-01-16 15:19:27.127974583 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 15:19:21.108942269 +0000
+++ -	2026-01-16 15:19:27.199789315 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 15:19:21.109942269 +0000
+++ -	2026-01-16 15:19:27.514850239 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-16T16:19:07Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
14 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttxj with backend undefined.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 16:18:58.016254504 +0000
+++ -	2026-01-16 16:18:59.639549528 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 16:18:57.767256874 +0000
+++ -	2026-01-16 16:18:59.832539421 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_gu_ttxu/color_sum.cc	2026-01-16 16:18:57.915255396 +0000
+++ -	2026-01-16 16:19:00.399652979 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gu_ttxu/CPPProcess.cc	2026-01-16 16:18:57.914255406 +0000
+++ -	2026-01-16 16:19:00.479959367 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gu_ttxu/check_sa.cc	2026-01-16 16:18:57.915255396 +0000
+++ -	2026-01-16 16:19:00.803519251 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_uux_ttxg/color_sum.cc	2026-01-16 16:18:58.016254504 +0000
+++ -	2026-01-16 16:19:01.224554339 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttxg/CPPProcess.cc	2026-01-16 16:18:58.015254512 +0000
+++ -	2026-01-16 16:19:01.305121864 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttxg/check_sa.cc	2026-01-16 16:18:58.016254504 +0000
+++ -	2026-01-16 16:19:01.629830063 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gux_ttxux/color_sum.cc	2026-01-16 16:18:57.966254935 +0000
+++ -	2026-01-16 16:19:04.636468242 +0000
@@ -381,20 +381,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -411,13 +411,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gux_ttxux/CPPProcess.cc	2026-01-16 16:18:57.965254944 +0000
+++ -	2026-01-16 16:19:04.712019664 +0000
@@ -813,11 +813,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -858,11 +858,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1199,7 +1199,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1262,7 +1262,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gux_ttxux/check_sa.cc	2026-01-16 16:18:57.965254944 +0000
+++ -	2026-01-16 16:19:05.030109298 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttxg/color_sum.cc	2026-01-16 16:18:57.849256055 +0000
+++ -	2026-01-16 16:19:05.478365503 +0000
@@ -383,20 +383,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -413,13 +413,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttxg/CPPProcess.cc	2026-01-16 16:18:57.848256065 +0000
+++ -	2026-01-16 16:19:05.562128437 +0000
@@ -975,11 +975,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -1020,11 +1020,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1361,7 +1361,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1424,7 +1424,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttxg/check_sa.cc	2026-01-16 16:18:57.848256065 +0000
+++ -	2026-01-16 16:19:05.889087306 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

github-actions · 2026-01-16T16:19:09Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend undefined.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 16:18:57.321709930 +0000
+++ -	2026-01-16 16:18:59.985197355 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 16:18:57.243709278 +0000
+++ -	2026-01-16 16:19:00.171356848 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 16:18:57.321709930 +0000
+++ -	2026-01-16 16:19:00.839196357 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 16:18:57.320709921 +0000
+++ -	2026-01-16 16:19:00.912205806 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 16:18:57.320709921 +0000
+++ -	2026-01-16 16:19:01.232734557 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 16:18:57.275709546 +0000
+++ -	2026-01-16 16:19:01.625255136 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 16:18:57.274709537 +0000
+++ -	2026-01-16 16:19:01.699041532 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 16:18:57.275709546 +0000
+++ -	2026-01-16 16:19:02.022618975 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-16T16:27:15Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend undefined.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 16:27:03.404345463 +0000
+++ -	2026-01-16 16:27:06.570030805 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 16:27:03.324345508 +0000
+++ -	2026-01-16 16:27:06.762097824 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 16:27:03.404345463 +0000
+++ -	2026-01-16 16:27:07.444786479 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 16:27:03.403345464 +0000
+++ -	2026-01-16 16:27:07.520655456 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 16:27:03.403345464 +0000
+++ -	2026-01-16 16:27:07.849375035 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 16:27:03.356345490 +0000
+++ -	2026-01-16 16:27:08.264139453 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 16:27:03.355345491 +0000
+++ -	2026-01-16 16:27:08.345094682 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 16:27:03.355345491 +0000
+++ -	2026-01-16 16:27:08.689472366 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

Source PR: roiser/madgraph4gpu#11

update remote readme gather data for readme

fix python script always use home dir

print python version add 8 use python setup

fix typo empty empty re-order table, add branch name

move to 0, check success invert use multiple processes

github-actions · 2026-01-16T16:33:16Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend undefined.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 16:33:07.988583084 +0000
+++ -	2026-01-16 16:33:09.961899282 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 16:33:07.908582388 +0000
+++ -	2026-01-16 16:33:10.150219083 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 16:33:07.988583084 +0000
+++ -	2026-01-16 16:33:10.849912422 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 16:33:07.988583084 +0000
+++ -	2026-01-16 16:33:10.923657727 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 16:33:07.988583084 +0000
+++ -	2026-01-16 16:33:11.245644119 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 16:33:07.936582631 +0000
+++ -	2026-01-16 16:33:11.636797746 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 16:33:07.935582622 +0000
+++ -	2026-01-16 16:33:11.711957042 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 16:33:07.936582631 +0000
+++ -	2026-01-16 16:33:12.034587658 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

github-actions · 2026-01-16T16:45:25Z

❌ Code Format Check Failed

The generated code does not conform to clang-format rules.

Please update your code generator to produce properly formatted code.
8 files need formatting.

Note: The CI is setup so it fails early, if formatting issues are detected for any of the processes.
The report below is for process PROC_pp_ttx with backend undefined.

See attached patch for details:

--- src/mgOnGpuConfig.h	2026-01-16 16:45:18.166641813 +0000
+++ -	2026-01-16 16:45:20.159209862 +0000
@@ -180,7 +180,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef double fptype; // double precision (8 bytes, fp64)
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef float fptype;  // single precision (4 bytes, fp32)
+  typedef float fptype; // single precision (4 bytes, fp32)
#endif

  // Floating point type (for color algebra alone #537): fptype2

--- src/mgOnGpuVectors.h	2026-01-16 16:45:18.084642616 +0000
+++ -	2026-01-16 16:45:20.339800460 +0000
@@ -123,7 +123,7 @@
#if defined MGONGPU_FPTYPE_DOUBLE
  typedef long int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#elif defined MGONGPU_FPTYPE_FLOAT
-  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) );                         // bbbb
+  typedef int bool_v __attribute__( ( ext_vector_type( neppV ) ) ); // bbbb
#endif
#else // gcc
  typedef unsigned int uint_v __attribute__( ( vector_size( neppV * sizeof( unsigned int ) ) ) );

--- SubProcesses/P1_uux_ttx/color_sum.cc	2026-01-16 16:45:18.167641804 +0000
+++ -	2026-01-16 16:45:20.979991471 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_uux_ttx/CPPProcess.cc	2026-01-16 16:45:18.166641813 +0000
+++ -	2026-01-16 16:45:21.051299423 +0000
@@ -735,11 +735,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -780,11 +780,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1121,7 +1121,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1184,7 +1184,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_uux_ttx/check_sa.cc	2026-01-16 16:45:18.166641813 +0000
+++ -	2026-01-16 16:45:21.365809576 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

--- SubProcesses/P1_gg_ttx/color_sum.cc	2026-01-16 16:45:18.122642244 +0000
+++ -	2026-01-16 16:45:21.746712717 +0000
@@ -379,20 +379,20 @@

#ifdef MGONGPUCPP_GPUIMPL
  void
-  color_sum_gpu( fptype* ghelAllMEs,               // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
-                 const fptype* ghelAllJamps,       // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
-                 fptype2* ghelAllBlasTmp,          // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
-                 gpuBlasHandle_t* pBlasHandle,     // input: cuBLAS/hipBLAS handle
-                 gpuStream_t* ghelStreams,         // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
-                 const int nGoodHel,               // input: number of good helicities
-                 const int gpublocks,              // input: cuda gpublocks
-                 const int gputhreads )            // input: cuda gputhreads
+  color_sum_gpu( fptype* ghelAllMEs,           // output: allMEs super-buffer for nGoodHel <= ncomb individual helicities (index is ighel)
+                 const fptype* ghelAllJamps,   // input: allJamps super-buffer[2][ncol][nGoodHel][nevt] for nGoodHel <= ncomb individual helicities
+                 fptype2* ghelAllBlasTmp,      // tmp: allBlasTmp super-buffer for nGoodHel <= ncomb individual helicities
+                 gpuBlasHandle_t* pBlasHandle, // input: cuBLAS/hipBLAS handle
+                 gpuStream_t* ghelStreams,     // input: cuda streams (index is ighel: only the first nGoodHel <= ncomb are non-null)
+                 const int nGoodHel,           // input: number of good helicities
+                 const int gpublocks,          // input: cuda gpublocks
+                 const int gputhreads )        // input: cuda gputhreads
  {
    const int nevt = gpublocks * gputhreads;
    // CASE 1: KERNEL
    if( !pBlasHandle )
    {
-      assert( ghelAllBlasTmp == nullptr );  // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
+      assert( ghelAllBlasTmp == nullptr ); // sanity check for HASBLAS=hasNoBlas or CUDACPP_RUNTIME_BLASCOLORSUM not set
      // Loop over helicities
      for( int ighel = 0; ighel < nGoodHel; ighel++ )
      {
@@ -409,13 +409,13 @@
      assert( false ); // sanity check: no path to this statement for HASBLAS=hasNoBlas
#else
      checkGpu( gpuDeviceSynchronize() ); // do not start the BLAS color sum for all helicities until the loop over helicities has completed
-      // Reset the tmp buffer
+                                          // Reset the tmp buffer
#if defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( 2 * ncolor * mgOnGpu::nx2 + 1 ) * sizeof( fptype2 ) );
#else
      gpuMemset( ghelAllBlasTmp, 0, nGoodHel * nevt * ( ncolor * mgOnGpu::nx2 ) * sizeof( fptype2 ) );
#endif
-      // Delegate the color sum to BLAS for 
+      // Delegate the color sum to BLAS for
      color_sum_blas( ghelAllMEs, ghelAllJamps, ghelAllBlasTmp, pBlasHandle, ghelStreams, nGoodHel, gpublocks, gputhreads );
#endif
    }

--- SubProcesses/P1_gg_ttx/CPPProcess.cc	2026-01-16 16:45:18.121642254 +0000
+++ -	2026-01-16 16:45:21.819008155 +0000
@@ -758,11 +758,11 @@
                       const fptype* allcouplings, // input: couplings[nevt*ndcoup*2]
                       fptype* allMEs,             // output: allMEs[nevt], |M|^2 final_avg_over_helicities
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                       fptype* allNumerators,      // output: multichannel numerators[nevt], running_sum_over_helicities
-                       fptype* allDenominators,    // output: multichannel denominators[nevt], running_sum_over_helicities
+                       fptype* allNumerators,   // output: multichannel numerators[nevt], running_sum_over_helicities
+                       fptype* allDenominators, // output: multichannel denominators[nevt], running_sum_over_helicities
#endif
-                       bool* isGoodHel,            // output: isGoodHel[ncomb] - host array
-                       const int nevt )            // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
+                       bool* isGoodHel, // output: isGoodHel[ncomb] - host array
+                       const int nevt ) // input: #events (for cuda: nevt == ndim == gpublocks*gputhreads)
  {
    //assert( (size_t)(allmomenta) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
    //assert( (size_t)(allMEs) % mgOnGpu::cppAlign == 0 ); // SANITY CHECK: require SIMD-friendly alignment [COMMENT OUT TO TEST MISALIGNED ACCESS]
@@ -803,11 +803,11 @@
#endif
        }
        constexpr fptype_sv* jamp2_sv = nullptr; // no need for color selection during helicity filtering
-        //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
+                                                 //std::cout << "sigmaKin_getGoodHel ihel=" << ihel << ( isGoodHel[ihel] ? " true" : " false" ) << std::endl;
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
        cxtype_sv jamp_sv[2 * ncolor] = {}; // all zeros
#else
-        cxtype_sv jamp_sv[ncolor] = {};  // all zeros
+        cxtype_sv jamp_sv[ncolor] = {}; // all zeros
#endif
#ifdef MGONGPU_SUPPORTS_MULTICHANNEL /* clang-format off */
        constexpr unsigned int channelId = 0; // disable multichannel single-diagram enhancement
@@ -1144,7 +1144,7 @@
    assert( npagV % 2 == 0 );     // SANITY CHECK for mixed fptypes: two neppV-pages are merged to one 2*neppV-page
    const int npagV2 = npagV / 2; // loop on two SIMD pages (neppV events) at a time
#else
-    const int npagV2 = npagV;            // loop on one SIMD page (neppV events) at a time
+    const int npagV2 = npagV; // loop on one SIMD page (neppV events) at a time
#endif
#ifdef _OPENMP
    // OMP multithreading #575 (NB: tested only with gcc11 so far)
@@ -1207,7 +1207,7 @@
      // Running sum of partial amplitudes squared for event by event color selection (#402)
      // (jamp2[nParity][ncolor][neppV] for the SIMD vector - or the two SIMD vectors - of events processed in calculate_jamps)
      fptype_sv jamp2_sv[nParity * ncolor] = {};
-      fptype_sv MEs_ighel[ncomb] = {};  // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
+      fptype_sv MEs_ighel[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the first - and/or only - neppV page)
#if defined MGONGPU_CPPSIMD and defined MGONGPU_FPTYPE_DOUBLE and defined MGONGPU_FPTYPE2_FLOAT
      fptype_sv MEs_ighel2[ncomb] = {}; // sum of MEs for all good helicities up to ighel (for the second neppV page)
#endif

--- SubProcesses/P1_gg_ttx/check_sa.cc	2026-01-16 16:45:18.122642244 +0000
+++ -	2026-01-16 16:45:22.136845237 +0000
@@ -135,7 +135,7 @@
#ifdef MGONGPUCPP_GPUIMPL
  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboDevice; // default on GPU
#else
-  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost;   // default on CPU
+  RamboSamplingMode rmbsmp = RamboSamplingMode::RamboHost; // default on CPU
#endif
  // Bridge emulation mode (NB Bridge implies RamboHost!)
  bool bridge = false;
@@ -918,7 +918,7 @@
  wrkflwtxt += "/sse4";
#endif
#else
-  wrkflwtxt += "/????";                                           // no path to this statement
+  wrkflwtxt += "/????"; // no path to this statement
#endif
  // -- Has cxtype_v::operator[] bracket with non-const reference?
#if defined MGONGPU_CPPSIMD
@@ -1143,7 +1143,7 @@
#elif defined MGONGPU_CUCXTYPE_STDCOMPLEX
             << "\"STD::COMPLEX\"," << std::endl
#else
-             << "\"???\"," << std::endl                           // no path to this statement...
+             << "\"???\"," << std::endl // no path to this statement...
#endif
             << "\"RanNumb memory layout\": "
             << "\"AOSOA[" << neppR << "]\""

roiser closed this Nov 23, 2025

roiser reopened this Nov 26, 2025

roiser force-pushed the ghci branch 2 times, most recently from 361fe98 to 4cdb1bc Compare January 12, 2026 07:31

roiser pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 12, 2026

Code generated from PR #11

98e966c

Source PR: roiser/madgraph4gpu#11

roiser pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 14, 2026

Code generated from PR #11

f56d092

Source PR: roiser/madgraph4gpu#11

roiser pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 14, 2026

Code generated from PR #11

7e0b525

Source PR: roiser/madgraph4gpu#11

roiser pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 14, 2026

Code generated from PR #11

910e629

Source PR: roiser/madgraph4gpu#11

roiser pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 14, 2026

Code generated from PR #11

4232865

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

2e94452

Source PR: roiser/madgraph4gpu#11

roiser added 8 commits January 16, 2026 11:22

new file

ac0d741

change branch pwd Enhance hello world step with directory listing new file

update submodule

339c58c

fix process move ci file update submodule

add py.py

c196787

gen one process

dc1e05b

pwd pwd pwd mg5 fetch submodules disable all processes, generate first process fix remote repo fix typo generate one process

copy files over

f47c8ff

move to PAT empty add pwd add git status spell out repo fix repo name more debug go one level up copy files over

move to new token

17c9d4d

remove persist credentials change pat name add ls remove if, single line commit simplify ci pat in local repo use PAT as a token

fix clang format

753368d

fix commit to remote repo

cb08b11

roiser added 2 commits January 16, 2026 11:26

test github app

b8846e1

generate access token run only one config fix typo indentation error simplify token gen remove installation-id add repo name test github app

empty

12aac65

roiser force-pushed the ghci branch from 505cc7e to 12aac65 Compare January 16, 2026 10:27

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

1a44103

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

4b0e052

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

ea658d0

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

4a7fdeb

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

cc6a3ae

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

09096ff

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

c4c63a6

Source PR: roiser/madgraph4gpu#11

ci-push bot pushed a commit to roiser/madgraph4gpu-generated-processes that referenced this pull request Jan 16, 2026

Code generated from PR #11

7579f16

Source PR: roiser/madgraph4gpu#11

roiser added 6 commits January 16, 2026 17:30

gather data for readme

b8a92a6

update remote readme gather data for readme

always use home dir

fc2d67c

fix python script always use home dir

use python setup

4a776ef

print python version add 8 use python setup

re-order table, add branch name

b7124cf

fix typo empty empty re-order table, add branch name

multiple processes

f183d8b

move to 0, check success invert use multiple processes

move to 0

bf37539

roiser force-pushed the ghci branch from 11bfa58 to bf37539 Compare January 16, 2026 16:32

roiser added 3 commits January 16, 2026 17:38

make ci fail immediately when code check fails

a476711

remove last step, not needed anymore

e4e2fb3

preprend all steps with if statement

9c9238d

Ghci #11

Are you sure you want to change the base?

Ghci #11

Uh oh!

Conversation

roiser commented Nov 23, 2025

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 14, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

github-actions bot commented Jan 16, 2026

❌ Code Format Check Failed

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants