GPU: Fix typo in variable name, fix comments, fix debug messages

Author: davidrohr

GPU/GPUTracking/Base/GPUReconstructionCPU.cxx

@@ -68,7 +68,7 @@ inline void GPUReconstructionCPUBackend::runKernelBackendInternal(const krnlSetu
   int32_t nThreads = getNKernelHostThreads(false);
   if (nThreads > 1) {
     if (mProcessingSettings.debugLevel >= 5) {
-      printf("Running %d Threads\n", nThreads);
+      printf("Running %d Threads\n", mThreading->activeThreads->max_concurrency());
     }
     tbb::this_task_arena::isolate([&] {
       mThreading->activeThreads->execute([&] {
@@ -91,10 +91,10 @@ inline void GPUReconstructionCPUBackend::runKernelBackendInternal(const krnlSetu
 template <>
 inline void GPUReconstructionCPUBackend::runKernelBackendInternal<GPUMemClean16, 0>(const krnlSetupTime& _xyz, void* const& ptr, uint64_t const& size)
 {
-  int32_t nnThreads = std::max<int32_t>(1, std::min<int32_t>(size / (16 * 1024 * 1024), getNKernelHostThreads(true)));
-  if (nnThreads > 1) {
-    tbb::parallel_for(0, nnThreads, [&](int iThread) {
-      size_t threadSize = size / nnThreads;
+  int32_t nThreads = std::max<int32_t>(1, std::min<int32_t>(size / (16 * 1024 * 1024), getNKernelHostThreads(true)));
+  if (nThreads > 1) {
+    tbb::parallel_for(0, nThreads, [&](int iThread) {
+      size_t threadSize = size / nThreads;
       if (threadSize % 4096) {
         threadSize += 4096 - threadSize % 4096;
       }

GPU/GPUTracking/Base/cuda/CMakeLists.txt

@@ -77,8 +77,8 @@ add_custom_command(
 create_binary_resource(${GPU_RTC_BIN}.src ${GPU_RTC_BIN}.src.o)
 
 add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_CUDA_SRC_CHK.done
-                    COMMAND ! grep "# [0-9]* \"\\(/usr/\\|.*GCC-Toolchain\\)" ${GPU_RTC_BIN}.src > ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_CUDA_SRC_CHK.done || bash -c "echo ERROR: CUDA RTC sources contain standard headers 1>&2 && exit 1"
-                    COMMENT "Checking CUDA RTC File ${GPU_RTC_BIN}.src"
+                    COMMAND ! grep "# [0-9]* \"\\(/usr/\\|.*GCC-Toolchain\\)" ${GPU_RTC_BIN}.src > ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_CUDA_SRC_CHK.done || bash -c "echo ERROR: CUDA RTC sources contain system headers 1>&2 && exit 1"
+                    COMMENT "Checking CUDA RTC File ${GPU_RTC_BIN}.src for system headers"
                     DEPENDS ${GPU_RTC_BIN}.src VERBATIM)
 add_custom_target(${MODULE}_CUDA_SRC_CHK ALL DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_CUDA_SRC_CHK.done)
 

GPU/GPUTracking/Base/hip/CMakeLists.txt

@@ -115,8 +115,8 @@ add_custom_command(
 create_binary_resource(${GPU_RTC_BIN}.src ${GPU_RTC_BIN}.src.o)
 
 add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_HIP_SRC_CHK.done
-                    COMMAND ! grep "# [0-9]* \"\\(/usr/\\|.*GCC-Toolchain\\)" ${GPU_RTC_BIN}.src > ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_HIP_SRC_CHK.done || bash -c "echo ERROR: HIP RTC sources contain standard headers 1>&2 && exit 1"
-                    COMMENT "Checking HIP RTC File ${GPU_RTC_BIN}.src"
+                    COMMAND ! grep "# [0-9]* \"\\(/usr/\\|.*GCC-Toolchain\\)" ${GPU_RTC_BIN}.src > ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_HIP_SRC_CHK.done || bash -c "echo ERROR: HIP RTC sources contain system headers 1>&2 && exit 1"
+                    COMMENT "Checking HIP RTC File ${GPU_RTC_BIN}.src for system headers"
                     DEPENDS ${GPU_RTC_BIN}.src VERBATIM)
 add_custom_target(${MODULE}_HIP_SRC_CHK ALL DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${MODULE}_HIP_SRC_CHK.done)
 

GPU/GPUTracking/Global/GPUChainTracking.cxx

@@ -267,7 +267,7 @@ bool GPUChainTracking::ValidateSettings()
     return false;
   }
   if ((GetRecoStepsGPU() & RecoStep::TPCClusterFinding) && std::max(GetProcessingSettings().nTPCClustererLanes + 1, GetProcessingSettings().nTPCClustererLanes * 2) + (GetProcessingSettings().doublePipeline ? 1 : 0) > (int32_t)mRec->NStreams()) {
-    GPUError("NStreams (%d) must be > nTPCClustererLanes (%d)", mRec->NStreams(), (int32_t)GetProcessingSettings().nTPCClustererLanes);
+    GPUError("NStreams of %d insufficient for %d nTPCClustererLanes", mRec->NStreams(), (int32_t)GetProcessingSettings().nTPCClustererLanes);
     return false;
   }
   if (GetProcessingSettings().noGPUMemoryRegistration && GetProcessingSettings().tpcCompressionGatherMode != 3) {

GPU/GPUTracking/cmake/kernel_helpers.cmake

@@ -76,7 +76,6 @@ function(o2_gpu_add_kernel kernel_name kernel_files)
   set_property(TARGET O2_GPU_KERNELS APPEND PROPERTY O2_GPU_KERNEL_NAMES "${kernel_name}")
   set_property(TARGET O2_GPU_KERNELS APPEND PROPERTY O2_GPU_KERNEL_INCLUDES "${TMP_KERNEL_CLASS_FILE}")
   set_property(TARGET O2_GPU_KERNELS APPEND PROPERTY O2_GPU_KERNEL_FILES "${TMP_KERNEL_CLASS_FILE}.cxx")
-  # add_custom_command OUTPUT option does not support target-dependend generator expressions, thus this workaround
 
   set(O2_GPU_KERNEL_TEMPLATE_FILES "GPUConstantMem.h")
   if (GPUCA_BUILD_DEBUG)
@@ -102,6 +101,7 @@ function(o2_gpu_add_kernel kernel_name kernel_files)
   list(TRANSFORM O2_GPU_KERNEL_TEMPLATE_FILES PREPEND "#include \"")
   list(JOIN O2_GPU_KERNEL_TEMPLATE_FILES "\n" O2_GPU_KERNEL_TEMPLATE_FILES)
 
+  # add_custom_command OUTPUT option does not support target-dependend generator expressions, thus this workaround to create CUDA and HIP files
   string(REPLACE ", " "_" TMP_FILENAME "${kernel_name}")
   if(CUDA_ENABLED)
     set(TMP_FILENAMEA "${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_${TMP_FILENAME}.cu")