Skip to content

GPU Cmake: Fix passing clang x86 denormal command to clang device compilation #14116

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
davidrohr merged 2 commits into from
Mar 27, 2025
Merged

GPU Cmake: Fix passing clang x86 denormal command to clang device compilation #14116

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7913bcd
GPU CMake: Cleanup and do not use Clang x86 denormals flag for device…
davidrohr Mar 27, 2025
5459352
GPU HIP CMake: Remove deprecated option
davidrohr Mar 27, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions GPU/GPUTracking/Base/opencl/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,9 +23,9 @@ endif()
set(CL_SRC ${GPUDIR}/Base/opencl/GPUReconstructionOCL.cl)
set(CL_BIN ${CMAKE_CURRENT_BINARY_DIR}/GPUReconstructionOCLCode)

set(OCL_FLAGS -Dcl_clang_storage_class_specifiers -cl-std=CLC++2021)
set(OCL_FLAGS -Dcl_clang_storage_class_specifiers -cl-std=CLC++2021 ${GPUCA_OCL_DENORMALS_FLAGS})
if(NOT GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
set(OCL_FLAGS ${OCL_FLAGS} -cl-denorms-are-zero -cl-mad-enable -cl-no-signed-zeros -cl-fast-relaxed-math)
set(OCL_FLAGS ${OCL_FLAGS} -cl-mad-enable -cl-no-signed-zeros -cl-fast-relaxed-math)
else()
set(OCL_FLAGS ${OCL_FLAGS} ${GPUCA_OCL_NO_FAST_MATH_FLAGS})
endif()
Expand Down
5 changes: 3 additions & 2 deletions GPU/GPUTracking/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,11 +17,12 @@ set(MODULE GPUTracking)
if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_OPTO2})
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O2 ${GPUCA_CLANG_FTZ}")
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O2")
endif()
elseif(NOT CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O3 -ffast-math ${GPUCA_CLANG_FTZ}")
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O3 -ffast-math")
endif()
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_DENORMALS_FLAGS}")

include(cmake/helpers.cmake)

Expand Down
56 changes: 27 additions & 29 deletions dependencies/FindO2GPU.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,17 +85,20 @@ elseif(NOT GPUCA_DETERMINISTIC_MODE MATCHES "^[0-9]+$")
message(STATUS "Set to ${GPUCA_DETERMINISTIC_MODE}")
endif()
if (CMAKE_SYSTEM_NAME MATCHES Darwin OR NOT CMAKE_SYSTEM_PROCESSOR MATCHES "(x86)|(X86)|(amd64)|(AMD64)")
set(GPUCA_CLANG_FTZ "")
set(GPUCA_CXX_DENORMALS_FLAGS "")
else()
set(GPUCA_CLANG_FTZ "-mdaz-ftz")
set(GPUCA_CXX_DENORMALS_FLAGS "-mdaz-ftz")
endif()
set(GPUCA_CXX_NO_FAST_MATH_FLAGS "-fno-fast-math -ffp-contract=off ${GPUCA_CLANG_FTZ}")
set(GPUCA_CUDA_NO_FAST_MATH_FLAGS "--ftz=true --prec-div=true --prec-sqrt=true --fmad false")
set(GPUCA_OCL_NO_FAST_MATH_FLAGS -cl-fp32-correctly-rounded-divide-sqrt -cl-denorms-are-zero)
set(GPUCA_CUDA_DENORMALS_FLAGS "--ftz=true")
set(GPUCA_OCL_DENORMALS_FLAGS "-cl-denorms-are-zero")
set(GPUCA_HIP_DENORMALS_FLAGS "-fgpu-flush-denormals-to-zero")
set(GPUCA_CXX_NO_FAST_MATH_FLAGS "-fno-fast-math -ffp-contract=off")
set(GPUCA_CUDA_NO_FAST_MATH_FLAGS "--prec-div=true --prec-sqrt=true --fmad false")
set(GPUCA_OCL_NO_FAST_MATH_FLAGS -cl-fp32-correctly-rounded-divide-sqrt )
if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_WHOLEO2})
add_definitions(-DGPUCA_DETERMINISTIC_MODE)
set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
set(CMAKE_C_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_C_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
string(APPEND CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
string(APPEND CMAKE_C_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
endif()


Expand Down Expand Up @@ -143,27 +146,27 @@ if(ENABLE_CUDA)
endif()
endif()
if(CMAKE_CUDA_COMPILER)
set(CMAKE_CUDA_FLAGS "-Xcompiler \"${O2_GPU_CMAKE_CXX_FLAGS_NOSTD}\" ${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr --extended-lambda -Xcompiler -Wno-attributes")
set(CMAKE_CUDA_FLAGS "-Xcompiler \"${O2_GPU_CMAKE_CXX_FLAGS_NOSTD}\" ${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr --extended-lambda -Xcompiler -Wno-attributes ${GPUCA_CUDA_DENORMALS_FLAGS}")
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "-Xcompiler \"${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}\" ${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
if(GPUCA_KERNEL_RESOURCE_USAGE_VERBOSE)
string(APPEND CMAKE_CUDA_FLAGS " -Xptxas -v")
endif()
string(APPEND CMAKE_CUDA_FLAGS " -Xcudafe --diag_suppress=114")
if (NOT ENABLE_CUDA STREQUAL "AUTO")
string(APPEND CMAKE_CUDA_FLAGS " --allow-unsupported-compiler")
endif()
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "-Xcompiler \"${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}\" ${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
if(CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -lineinfo -Xptxas -O0 -Xcompiler -O0")
string(APPEND CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " -lineinfo -Xptxas -O0")
else()
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -Xptxas -O4 -Xcompiler -O4")
string(APPEND CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " -Xptxas -O4 -Xcompiler -O4")
endif()
if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CUDA_NO_FAST_MATH_FLAGS}")
string(APPEND CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " ${GPUCA_CUDA_NO_FAST_MATH_FLAGS}")
elseif(NOT CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
set(CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -use_fast_math --ftz=true")#
string(APPEND CMAKE_CUDA_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " -use_fast_math --ftz=true")
endif()
if(CMAKE_CXX_FLAGS MATCHES "(^| )-Werror( |$)")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Werror=cross-execution-space-call")
string(APPEND CMAKE_CUDA_FLAGS " -Werror=cross-execution-space-call")
endif()
if(GPUCA_CUDA_GCCBIN)
list(FILTER CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES EXCLUDE REGEX "^/usr/lib.*/gcc/") # Workaround, since CMake adds old GCC lib paths implicitly if we request that gcc for CUDA
Expand Down Expand Up @@ -285,27 +288,22 @@ if(ENABLE_HIP)
set(HIP_ENABLED ON)
set_target_properties(roc::rocthrust PROPERTIES IMPORTED_GLOBAL TRUE)
message(STATUS "HIP Found (${hip_HIPCC_EXECUTABLE} version ${hip_VERSION})")
set(O2_HIP_CMAKE_CXX_FLAGS "-fgpu-defer-diag -mllvm -amdgpu-enable-lower-module-lds=false -mllvm -amdgpu-function-calls=true -Wno-invalid-command-line-argument -Wno-unused-command-line-argument -Wno-invalid-constexpr -Wno-ignored-optimization-argument -Wno-unused-private-field -Wno-pass-failed")
if(hip_VERSION VERSION_GREATER_EQUAL "6.0" AND NOT hip_VERSION VERSION_GREATER_EQUAL "6.2")
string(APPEND O2_HIP_CMAKE_CXX_FLAGS " -mllvm -amdgpu-legacy-sgpr-spill-lowering=true") # TODO: Cleanup
endif()
set(CMAKE_HIP_FLAGS "${O2_GPU_CMAKE_CXX_FLAGS_NOSTD} ${CMAKE_HIP_FLAGS} ${GPUCA_HIP_DENORMALS_FLAGS}")
set(CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
string(APPEND CMAKE_HIP_FLAGS " -fgpu-defer-diag -mllvm -amdgpu-enable-lower-module-lds=false -mllvm -amdgpu-function-calls=true -Wno-invalid-command-line-argument -Wno-unused-command-line-argument -Wno-invalid-constexpr -Wno-ignored-optimization-argument -Wno-unused-private-field -Wno-pass-failed ")
if(GPUCA_KERNEL_RESOURCE_USAGE_VERBOSE)
string(APPEND O2_HIP_CMAKE_CXX_FLAGS " -Rpass-analysis=kernel-resource-usage")
string(APPEND CMAKE_HIP_FLAGS " -Rpass-analysis=kernel-resource-usage")
endif()
if(NOT GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
string(APPEND CMAKE_HIP_FLAGS " -ffast-math")
endif()
if(NOT CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
string(APPEND CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} " -O3")
endif()
string(REGEX REPLACE "(gfx1[0-9]+;?)" "" CMAKE_HIP_ARCHITECTURES "${CMAKE_HIP_ARCHITECTURES}") # ROCm currently doesn’t support integrated graphics
if(HIP_AMDGPUTARGET)
set(CMAKE_HIP_ARCHITECTURES "${HIP_AMDGPUTARGET}") # If GPU build is enforced we override autodetection
endif()
if(NOT GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
string(APPEND O2_HIP_CMAKE_CXX_FLAGS " -fgpu-flush-denormals-to-zero -ffast-math")
endif()
set(CMAKE_HIP_FLAGS "${O2_GPU_CMAKE_CXX_FLAGS_NOSTD} ${CMAKE_HIP_FLAGS} ${O2_HIP_CMAKE_CXX_FLAGS}")
set(CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}}")
if(CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
set(CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O0 -ggdb")
else()
set(CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_HIP_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O3")
endif()
else()
set(HIP_ENABLED OFF)
endif()
Expand Down