[june24] in gg_tt.mad Bridge*/runTest/check, add useChannelIds argume…

…nt in computeMatrixElements in madgraph5#892 - bldall ok but runTest segfaults

epochX/cudacpp/gg_tt.mad/SubProcesses/Bridge.h

@@ -335,18 +335,19 @@ namespace mg5amcCpu
       std::copy( rndhel, rndhel + m_nevt, m_hstRndHel.data() );
       std::copy( rndcol, rndcol + m_nevt, m_hstRndCol.data() );
     }
-    if( channelIds ) memcpy( m_hstChannelIds.data(), channelIds, m_nevt * sizeof( unsigned int ) );
+    const bool useChannelIds = ( channelIds != nullptr ) && ( !goodHelOnly );
+    if( useChannelIds ) memcpy( m_hstChannelIds.data(), channelIds, m_nevt * sizeof( unsigned int ) );
     copyDeviceFromHost( m_devGs, m_hstGs );
     copyDeviceFromHost( m_devRndHel, m_hstRndHel );
     copyDeviceFromHost( m_devRndCol, m_hstRndCol );
+    if( useChannelIds ) copyDeviceFromHost( m_devChannelIds, m_hstChannelIds );
     if( m_nGoodHel < 0 )
     {
       m_nGoodHel = m_pmek->computeGoodHelicities();
       if( m_nGoodHel < 0 ) throw std::runtime_error( "Bridge gpu_sequence: computeGoodHelicities returned nGoodHel<0" );
     }
     if( goodHelOnly ) return;
-    copyDeviceFromHost( m_devChannelIds, m_hstChannelIds );
-    m_pmek->computeMatrixElements();
+    m_pmek->computeMatrixElements( useChannelIds );
     copyHostFromDevice( m_hstMEs, m_devMEs );
     flagAbnormalMEs( m_hstMEs.data(), m_nevt );
     copyHostFromDevice( m_hstSelHel, m_devSelHel );
@@ -391,14 +392,15 @@ namespace mg5amcCpu
       std::copy( rndhel, rndhel + m_nevt, m_hstRndHel.data() );
       std::copy( rndcol, rndcol + m_nevt, m_hstRndCol.data() );
     }
+    const bool useChannelIds = ( channelIds != nullptr ) && ( !goodHelOnly );
+    if( useChannelIds ) memcpy( m_hstChannelIds.data(), channelIds, m_nevt * sizeof( unsigned int ) );
     if( m_nGoodHel < 0 )
     {
       m_nGoodHel = m_pmek->computeGoodHelicities();
       if( m_nGoodHel < 0 ) throw std::runtime_error( "Bridge cpu_sequence: computeGoodHelicities returned nGoodHel<0" );
     }
     if( goodHelOnly ) return;
-    if( channelIds ) memcpy( m_hstChannelIds.data(), channelIds, m_nevt * sizeof( unsigned int ) );
-    m_pmek->computeMatrixElements();
+    m_pmek->computeMatrixElements( useChannelIds );
     flagAbnormalMEs( m_hstMEs.data(), m_nevt );
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {

epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.cc

@@ -79,17 +79,18 @@ namespace mg5amcCpu
   int BridgeKernelHost::computeGoodHelicities()
   {
     constexpr bool goodHelOnly = true;
-    constexpr unsigned int* channelId = nullptr; // disable multi-channel for helicity filtering
-    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), channelId, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
+    constexpr unsigned int* pChannelIds = nullptr; // disable multi-channel for helicity filtering
+    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), pChannelIds, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
     return m_bridge.nGoodHel();
   }
 
   //--------------------------------------------------------------------------
 
-  void BridgeKernelHost::computeMatrixElements()
+  void BridgeKernelHost::computeMatrixElements( const bool useChannelIds )
   {
     constexpr bool goodHelOnly = false;
-    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), m_channelIds.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
+    const unsigned int* pChannelIds = ( useChannelIds ? m_channelIds.data() : nullptr );
+    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), pChannelIds, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
   }
 
   //--------------------------------------------------------------------------
@@ -137,17 +138,18 @@ namespace mg5amcGpu
   int BridgeKernelDevice::computeGoodHelicities()
   {
     constexpr bool goodHelOnly = true;
-    constexpr unsigned int* channelId = nullptr;
-    m_bridge.gpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), channelId, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
+    constexpr unsigned int* pChannelIds = nullptr; // disable multi-channel for helicity filtering
+    m_bridge.gpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), pChannelIds, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
     return m_bridge.nGoodHel();
   }
 
   //--------------------------------------------------------------------------
 
-  void BridgeKernelDevice::computeMatrixElements()
+  void BridgeKernelDevice::computeMatrixElements( const bool useChannelIds )
   {
     constexpr bool goodHelOnly = false;
-    m_bridge.gpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), m_channelIds.data(), m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
+    const unsigned int* pChannelIds = ( useChannelIds ? m_channelIds.data() : nullptr );
+    m_bridge.gpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_rndhel.data(), m_rndcol.data(), pChannelIds, m_matrixElements.data(), m_selhel.data(), m_selcol.data(), goodHelOnly );
   }
 
   //--------------------------------------------------------------------------

epochX/cudacpp/gg_tt.mad/SubProcesses/BridgeKernels.h

@@ -77,7 +77,7 @@ namespace mg5amcCpu
     int computeGoodHelicities() override final;
 
     // Compute matrix elements
-    void computeMatrixElements() override final;
+    void computeMatrixElements( const bool useChannelIds ) override final;
 
     // Is this a host or device kernel?
     bool isOnDevice() const override final { return false; }
@@ -119,7 +119,7 @@ namespace mg5amcCpu
     int computeGoodHelicities() override final;
 
     // Compute matrix elements
-    void computeMatrixElements() override final;
+    void computeMatrixElements( const bool useChannelIds ) override final;
 
     // Is this a host or device kernel?
     bool isOnDevice() const override final { return true; }

epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/CPPProcess.cc

@@ -189,7 +189,7 @@ namespace mg5amcCpu
                            const fptype* allcouplings,      // input: couplings[nevt*ndcoup*2]
                            fptype* allMEs,                  // output: allMEs[nevt], |M|^2 running_sum_over_helicities
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-                           const unsigned int* channelIds,  // input: multichannel channel id (1 to #diagrams); 0 to disable single-diagram enhancement
+                           const unsigned int* channelIds,  // input: multichannel channel id (1 to #diagrams); nullptr to disable single-diagram enhancement
                            fptype* allNumerators,           // output: multichannel numerators[nevt], running_sum_over_helicities
                            fptype* allDenominators,         // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif
@@ -919,7 +919,7 @@ namespace mg5amcCpu
             const fptype* allrndcol,         // input: random numbers[nevt] for color selection
             fptype* allMEs,                  // output: allMEs[nevt], |M|^2 final_avg_over_helicities
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-            const unsigned int* channelIds,  // input: multichannel channel id (1 to #diagrams); 0 to disable single-diagram enhancement
+            const unsigned int* channelIds,  // input: multichannel channel id (1 to #diagrams); nullptr to disable single-diagram enhancement
             fptype* allNumerators,           // output: multichannel numerators[nevt], running_sum_over_helicities
             fptype* allDenominators,         // output: multichannel denominators[nevt], running_sum_over_helicities
 #endif

epochX/cudacpp/gg_tt.mad/SubProcesses/P1_gg_ttx/check_sa.cc

@@ -369,7 +369,9 @@ main( int argc, char** argv )
   DeviceBufferGs devGs( nevt );
 #endif
 
-// Memory buffer for channelIDs
+  // Memory buffer for channelIDs
+  // [AV: channelId arrays are needed to keep a simpler signature for MatrixElementKernel constructors]
+  // [but they are not used internally (fix #892) as long as check.exe uses no-multichannel (see #896)]
 #ifndef MGONGPUCPP_GPUIMPL
   HostBufferChannelIds hstChannelIds( nevt );
 #else
@@ -379,12 +381,12 @@ main( int argc, char** argv )
 
   // Hardcode Gs for now (eventually they should come from Fortran MadEvent)
   // Hardcode channelID to 0
-  constexpr unsigned int channelId = 0; // TEMPORARY? disable multi-channel in check.exe and gcheck.exe #466
+  //constexpr unsigned int channelId = 0; // TEMPORARY? disable multi-channel in check.exe and gcheck.exe #466
   for( unsigned int i = 0; i < nevt; ++i )
   {
     constexpr fptype fixedG = 1.2177157847767195; // fixed G for aS=0.118 (hardcoded for now in check_sa.cc, fcheck_sa.f, runTest.cc)
     hstGs[i] = fixedG;
-    hstChannelIds[i] = channelId;
+    //hstChannelIds[i] = channelId; // AV ChannelId arrays are not needed in check.exe (fix #892) as long as check.exe uses no-multichannel (see #896)
     //if ( i > 0 ) hstGs[i] = 0; // try hardcoding G only for event 0
     //hstGs[i] = i;
   }
@@ -658,7 +660,8 @@ main( int argc, char** argv )
     // --- 3a. SigmaKin
     const std::string skinKey = "3a SigmaKin";
     timermap.start( skinKey );
-    pmek->computeMatrixElements();
+    constexpr bool useChannelIds = false; // TEMPORARY? disable multi-channel in check.exe and gcheck.exe #466
+    pmek->computeMatrixElements( useChannelIds );
 
     // *** STOP THE NEW OLD-STYLE TIMER FOR MATRIX ELEMENTS (WAVEFUNCTIONS) ***
     wv3atime += timermap.stop(); // calc only

epochX/cudacpp/gg_tt.mad/SubProcesses/runTest.cc

@@ -103,8 +103,10 @@ struct CPUTest : public CUDA_CPU_TestBase
   {
     constexpr fptype fixedG = 1.2177157847767195; // fixed G for aS=0.118 (hardcoded for now in check_sa.cc, fcheck_sa.f, runTest.cc)
     for( unsigned int i = 0; i < nevt; ++i ) hstGs[i] = fixedG;
+    // [AV: there is no need to fill channelId arrays if runTest.exe uses no-multichannel]
     if( iiter == 0 ) pmek->computeGoodHelicities();
-    pmek->computeMatrixElements();
+    constexpr bool useChannelIds = false; // TEMPORARY? disable multi-channel in runTest.exe #466
+    pmek->computeMatrixElements( useChannelIds );
   }
 
   fptype getMomentum( std::size_t ievt, unsigned int ipar, unsigned int ip4 ) const override
@@ -196,8 +198,6 @@ struct CUDATest : public CUDA_CPU_TestBase
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton?
     process.initProc( "../../Cards/param_card.dat" );
-    std::fill_n( hstChannelIds.data(), nevt, 0 );
-    copyDeviceFromHost( devChannelIds, hstChannelIds );
   }
 
   virtual ~CUDATest() {}
@@ -229,8 +229,10 @@ struct CUDATest : public CUDA_CPU_TestBase
     constexpr fptype fixedG = 1.2177157847767195; // fixed G for aS=0.118 (hardcoded for now in check_sa.cc, fcheck_sa.f, runTest.cc)
     for( unsigned int i = 0; i < nevt; ++i ) hstGs[i] = fixedG;
     copyDeviceFromHost( devGs, hstGs ); // BUG FIX #566
+    // [AV: there is no need to fill channelId arrays if runTest.exe uses no-multichannel]
     if( iiter == 0 ) pmek->computeGoodHelicities();
-    pmek->computeMatrixElements();
+    constexpr bool useChannelIds = false; // TEMPORARY? disable multi-channel in runTest.exe #466
+    pmek->computeMatrixElements( useChannelIds );
     copyHostFromDevice( hstMatrixElements, devMatrixElements );
   }