GPU: Add HIP per-kernel compilation mode

Author: davidrohr

GPU/GPUTracking/Base/cuda/CMakeLists.txt

@@ -48,7 +48,8 @@ if(NOT ALIGPU_BUILD_TYPE STREQUAL "ALIROOT")
   else()
       set(RTC_CUDA_ARCH "750")
   endif()
-  separate_arguments(GPU_RTC_FLAGS)
+  set(GPU_RTC_FLAGS_SEPARATED "${GPU_RTC_FLAGS}")
+  separate_arguments(GPU_RTC_FLAGS_SEPARATED)
 
   # convenience variables
   if(ALIGPU_BUILD_TYPE STREQUAL "Standalone")
@@ -74,7 +75,7 @@ if(NOT ALIGPU_BUILD_TYPE STREQUAL "ALIROOT")
 
   add_custom_command(
       OUTPUT ${GPU_RTC_BIN}.command
-      COMMAND echo -n "${CMAKE_CUDA_COMPILER} ${GPU_RTC_FLAGS} ${GPU_RTC_DEFINES} -fatbin" > ${GPU_RTC_BIN}.command
+      COMMAND echo -n "${CMAKE_CUDA_COMPILER} ${GPU_RTC_FLAGS_SEPARATED} ${GPU_RTC_DEFINES} -fatbin" > ${GPU_RTC_BIN}.command
       COMMAND_EXPAND_LISTS
       VERBATIM
       COMMENT "Preparing CUDA RTC command file ${GPU_RTC_BIN}.command"

GPU/GPUTracking/Base/cuda/GPUReconstructionCUDA.cu

@@ -26,7 +26,12 @@
 
 #if defined(GPUCA_KERNEL_COMPILE_MODE) && GPUCA_KERNEL_COMPILE_MODE == 1
 #include "utils/qGetLdBinarySymbols.h"
-#define GPUCA_KRNL(x_class, ...) QGET_LD_BINARY_SYMBOLS(GPUCA_M_CAT3(cuda_kernel_module_fatbin_krnl_, GPUCA_M_KRNL_NAME(x_class), _fatbin))
+#ifndef __HIPCC__ // CUDA
+#define PER_KERNEL_OBJECT_EXT _fatbin
+#else // HIP
+#define PER_KERNEL_OBJECT_EXT _hip_cxx_o
+#endif
+#define GPUCA_KRNL(x_class, ...) QGET_LD_BINARY_SYMBOLS(GPUCA_M_CAT3(cuda_kernel_module_fatbin_krnl_, GPUCA_M_KRNL_NAME(x_class), PER_KERNEL_OBJECT_EXT))
 #include "GPUReconstructionKernelList.h"
 #undef GPUCA_KRNL
 #endif
@@ -368,7 +373,7 @@ int GPUReconstructionCUDA::InitDevice_Runtime()
     else {
 #define GPUCA_KRNL(x_class, ...)                                        \
   mInternals->kernelModules.emplace_back(std::make_unique<CUmodule>()); \
-  GPUFailedMsg(cuModuleLoadData(mInternals->kernelModules.back().get(), GPUCA_M_CAT3(_binary_cuda_kernel_module_fatbin_krnl_, GPUCA_M_KRNL_NAME(x_class), _fatbin_start)));
+  GPUFailedMsg(cuModuleLoadData(mInternals->kernelModules.back().get(), GPUCA_M_CAT3(_binary_cuda_kernel_module_fatbin_krnl_, GPUCA_M_KRNL_NAME(x_class), GPUCA_M_CAT(PER_KERNEL_OBJECT_EXT, _start))));
 #include "GPUReconstructionKernelList.h"
 #undef GPUCA_KRNL
       loadKernelModules(true, false);

GPU/GPUTracking/Base/hip/CMakeLists.txt

@@ -80,7 +80,8 @@ if(NOT ALIGPU_BUILD_TYPE STREQUAL "ALIROOT")
 
   # build flags to use for RTC
   set(GPU_RTC_FLAGS "${CMAKE_HIP_FLAGS} ${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -std=c++${CMAKE_HIP_STANDARD}")
-  separate_arguments(GPU_RTC_FLAGS)
+  set(GPU_RTC_FLAGS_SEPARATED "${GPU_RTC_FLAGS}")
+  separate_arguments(GPU_RTC_FLAGS_SEPARATED)
 
   # convenience variables
   if(ALIGPU_BUILD_TYPE STREQUAL "Standalone")
@@ -106,7 +107,7 @@ if(NOT ALIGPU_BUILD_TYPE STREQUAL "ALIROOT")
 
   add_custom_command(
       OUTPUT ${GPU_RTC_BIN}.command
-      COMMAND echo -n "${hip_HIPCC_EXECUTABLE} ${GPU_RTC_FLAGS} ${GPU_RTC_DEFINES} --genco" > ${GPU_RTC_BIN}.command
+      COMMAND echo -n "${hip_HIPCC_EXECUTABLE} ${GPU_RTC_FLAGS_SEPARATED} ${GPU_RTC_DEFINES} --genco" > ${GPU_RTC_BIN}.command
       COMMAND_EXPAND_LISTS
       VERBATIM
       COMMENT "Preparing HIP RTC command file ${GPU_RTC_BIN}.command"
@@ -206,19 +207,23 @@ if(NOT DEFINED GPUCA_HIP_COMPILE_MODE)
   if(ALIGPU_BUILD_TYPE STREQUAL "ALIROOT")
     set(GPUCA_HIP_COMPILE_MODE "onefile")
   else()
-    set(GPUCA_HIP_COMPILE_MODE "onefile")
+    set(GPUCA_HIP_COMPILE_MODE "perkernel")
   endif()
 endif()
 if(GPUCA_HIP_COMPILE_MODE STREQUAL "onefile")
   target_compile_definitions(${targetName} PRIVATE GPUCA_KERNEL_COMPILE_MODE=0)
 elseif(GPUCA_HIP_COMPILE_MODE STREQUAL "perkernel")
-  add_library(GPUTrackingHIPKernels OBJECT $<JOIN:$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<TARGET_PROPERTY:O2_GPU_KERNELS,O2_GPU_KERNEL_NAMES>,REPLACE,[^A-Za-z0-9]+,_>,PREPEND,${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_>,APPEND,.cu>, >)
-  set_property(TARGET GPUTrackingHIPKernels PROPERTY HIP_FATBIN_COMPILATION ON)
+  #add_library(GPUTrackingHIPKernels OBJECT $<JOIN:$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<TARGET_PROPERTY:O2_GPU_KERNELS,O2_GPU_KERNEL_NAMES>,REPLACE,[^A-Za-z0-9]+,_>,PREPEND,${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_>,APPEND,.cu>, >)
+  #set_property(TARGET GPUTrackingHIPKernels PROPERTY HIP_FATBIN_COMPILATION ON)
+  add_subdirectory(per_kernel)
   target_compile_options(GPUTrackingHIPKernels PRIVATE $<$<COMPILE_LANGUAGE:HIP>:-fno-gpu-rdc>)
   target_link_options(GPUTrackingHIPKernels PRIVATE $<$<COMPILE_LANGUAGE:HIP>:-fno-gpu-rdc>)
   target_compile_definitions(${targetName} PRIVATE GPUCA_KERNEL_COMPILE_MODE=1)
   target_compile_definitions(GPUTrackingHIPKernels PRIVATE $<TARGET_PROPERTY:${targetName},COMPILE_DEFINITIONS>)
   target_include_directories(GPUTrackingHIPKernels PRIVATE $<TARGET_PROPERTY:${targetName},INCLUDE_DIRECTORIES>)
+  if(NOT DEFINED GPUCA_HIP_HIPIFY_FROM_CUDA OR "${GPUCA_HIP_HIPIFY_FROM_CUDA}")
+    add_dependencies(GPUTrackingHIPKernels ${MODULE}_HIPIFIED)
+  endif()
   file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/hip_kernel_module_fatbin)
   add_custom_command(
     OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/GPUTrackingHIPKernelModules.o

GPU/GPUTracking/Base/hip/per_kernel/CMakeLists.txt

@@ -0,0 +1,15 @@
+# Copyright 2019-2020 CERN and copyright holders of ALICE O2.
+# See https://alice-o2.web.cern.ch/copyright for details of the copyright holders.
+# All rights not expressly granted are reserved.
+#
+# This software is distributed under the terms of the GNU General Public
+# License v3 (GPL Version 3), copied verbatim in the file "COPYING".
+#
+# In applying this license CERN does not waive the privileges and immunities
+# granted to it by virtue of its status as an Intergovernmental Organization
+# or submit itself to any jurisdiction.
+
+add_library(GPUTrackingHIPKernels OBJECT $<JOIN:$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<LIST:TRANSFORM,$<TARGET_PROPERTY:O2_GPU_KERNELS,O2_GPU_KERNEL_NAMES>,REPLACE,[^A-Za-z0-9]+,_>,PREPEND,${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_>,APPEND,.hip.cxx>, >)
+set(CMAKE_CXX_COMPILER ${hip_HIPCC_EXECUTABLE})
+set(CMAKE_CXX_FLAGS "${GPU_RTC_FLAGS} --genco")
+unset(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER})

GPU/GPUTracking/cmake/kernel_helpers.cmake

@@ -95,7 +95,7 @@ function(o2_gpu_add_kernel kernel_name kernel_files kernel_bounds kernel_type)
   endif()
 
   if(HIP_ENABLED)
-    set(TMP_FILENAMEA "${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_${TMP_FILENAME}.hip")
+    set(TMP_FILENAMEA "${O2_GPU_KERNEL_WRAPPER_FOLDER}/krnl_${TMP_FILENAME}.hip.cxx")
     set(O2_GPU_KERNEL_TEMPLATE_REPLACE "${TMP_KERNEL}")
     configure_file(${O2_GPU_BASE_DIR}/Base/hip/GPUReconstructionHIPkernel.template.hip ${TMP_FILENAMEA})
   endif()