[libc] Build the GPU during the projects setup like libc-hdrgen #84667
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jhuber6
requested review from
jdoerfert,
lntue,
michaelrj-google,
nickdesaulniers,
petrhosek and
shiltian
|
@llvm/pr-subscribers-libc @llvm/pr-subscribers-backend-amdgpu Author: Joseph Huber (jhuber6) ChangesSummary: Depends on #84664 Full diff: https://github.com/llvm/llvm-project/pull/84667.diff 10 Files Affected:
diff --git a/libc/CMakeLists.txt b/libc/CMakeLists.txt
index b4a2523b778877..6edf5c656193db 100644
--- a/libc/CMakeLists.txt
+++ b/libc/CMakeLists.txt
@@ -60,6 +60,10 @@ if(LLVM_LIBC_FULL_BUILD OR LLVM_LIBC_GPU_BUILD)
message(STATUS "Will use ${LIBC_HDRGEN_EXE} for libc header generation.")
endif()
endif()
+# We will build the GPU utilities if we are not doing a runtimes build.
+if(LLVM_LIBC_GPU_BUILD AND NOT LLVM_RUNTIMES_BUILD)
+ add_subdirectory(utils/gpu)
+endif()
set(NEED_LIBC_HDRGEN FALSE)
if(NOT LLVM_RUNTIMES_BUILD)
@@ -79,11 +83,6 @@ if(LIBC_HDRGEN_ONLY OR NEED_LIBC_HDRGEN)
# When libc is build as part of the runtimes/bootstrap build's CMake run, we
# only need to build the host tools to build the libc. So, we just do enough
# to build libc-hdrgen and return.
-
- # Always make the RPC server availible to other projects for GPU mode.
- if(LLVM_LIBC_GPU_BUILD)
- add_subdirectory(utils/gpu/server)
- endif()
return()
endif()
unset(NEED_LIBC_HDRGEN)
diff --git a/libc/cmake/modules/prepare_libc_gpu_build.cmake b/libc/cmake/modules/prepare_libc_gpu_build.cmake
index 2de4cb8d82b28b..bea6bb016491b6 100644
--- a/libc/cmake/modules/prepare_libc_gpu_build.cmake
+++ b/libc/cmake/modules/prepare_libc_gpu_build.cmake
@@ -93,6 +93,41 @@ else()
endif()
set(LIBC_GPU_TARGET_ARCHITECTURE "${gpu_test_architecture}")
+# Identify the GPU loader utility used to run tests.
+set(LIBC_GPU_LOADER_EXECUTABLE "" CACHE STRING "Executable for the GPU loader.")
+if(LIBC_GPU_LOADER_EXECUTABLE)
+ set(gpu_loader_executable ${LIBC_GPU_LOADER_EXECUTABLE})
+elseif(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
+ find_program(LIBC_AMDHSA_LOADER_EXECUTABLE
+ NAMES amdhsa-loader NO_DEFAULT_PATH
+ PATHS ${LLVM_BINARY_DIR}/bin ${compiler_path})
+ if(LIBC_AMDHSA_LOADER_EXECUTABLE)
+ set(gpu_loader_executable ${LIBC_AMDHSA_LOADER_EXECUTABLE})
+ endif()
+elseif(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
+ find_program(LIBC_NVPTX_LOADER_EXECUTABLE
+ NAMES nvptx-loader NO_DEFAULT_PATH
+ PATHS ${LLVM_BINARY_DIR}/bin ${compiler_path})
+ if(LIBC_NVPTX_LOADER_EXECUTABLE)
+ set(gpu_loader_executable ${LIBC_NVPTX_LOADER_EXECUTABLE})
+ endif()
+endif()
+if(NOT TARGET libc.utils.gpu.loader AND gpu_loader_executable)
+ add_custom_target(libc.utils.gpu.loader)
+ set_target_properties(
+ libc.utils.gpu.loader
+ PROPERTIES
+ EXECUTABLE "${gpu_loader_executable}"
+ )
+endif()
+
+if(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
+ # The AMDGPU environment uses different code objects to encode the ABI for
+ # kernel calls and intrinsic functions. We want to specify this manually to
+ # conform to whatever the test suite was built to handle.
+ set(LIBC_GPU_CODE_OBJECT_VERSION 5)
+endif()
+
if(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
# FIXME: This is a hack required to keep the CUDA package from trying to find
# pthreads. We only link the CUDA driver, so this is unneeded.
@@ -103,10 +138,3 @@ if(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
get_filename_component(LIBC_CUDA_ROOT "${CUDAToolkit_BIN_DIR}" DIRECTORY ABSOLUTE)
endif()
endif()
-
-if(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
- # The AMDGPU environment uses different code objects to encode the ABI for
- # kernel calls and intrinsic functions. We want to specify this manually to
- # conform to whatever the test suite was built to handle.
- set(LIBC_GPU_CODE_OBJECT_VERSION 5)
-endif()
diff --git a/libc/utils/CMakeLists.txt b/libc/utils/CMakeLists.txt
index 7bf02a4af7deae..11f25503cc13e2 100644
--- a/libc/utils/CMakeLists.txt
+++ b/libc/utils/CMakeLists.txt
@@ -1,6 +1,3 @@
if(LLVM_INCLUDE_TESTS)
add_subdirectory(MPFRWrapper)
endif()
-if(LIBC_TARGET_OS_IS_GPU)
- add_subdirectory(gpu)
-endif()
diff --git a/libc/utils/gpu/CMakeLists.txt b/libc/utils/gpu/CMakeLists.txt
index 4d1ebcfb9f8e65..7c15f36052cf3b 100644
--- a/libc/utils/gpu/CMakeLists.txt
+++ b/libc/utils/gpu/CMakeLists.txt
@@ -1,4 +1,2 @@
add_subdirectory(server)
-if(LIBC_TARGET_OS_IS_GPU)
- add_subdirectory(loader)
-endif()
+add_subdirectory(loader)
diff --git a/libc/utils/gpu/loader/CMakeLists.txt b/libc/utils/gpu/loader/CMakeLists.txt
index 189460bb02e6e5..c05b3f032f9644 100644
--- a/libc/utils/gpu/loader/CMakeLists.txt
+++ b/libc/utils/gpu/loader/CMakeLists.txt
@@ -6,37 +6,19 @@ target_include_directories(gpu_loader PUBLIC
${LIBC_SOURCE_DIR}
)
-# This utility needs to be compiled for the host system when cross compiling.
-if(LLVM_RUNTIMES_TARGET OR LIBC_TARGET_TRIPLE)
- target_compile_options(gpu_loader PUBLIC --target=${LLVM_HOST_TRIPLE})
- target_link_libraries(gpu_loader PUBLIC "--target=${LLVM_HOST_TRIPLE}")
-endif()
-
find_package(hsa-runtime64 QUIET 1.2.0 HINTS ${CMAKE_INSTALL_PREFIX} PATHS /opt/rocm)
-if(hsa-runtime64_FOUND AND LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
+if(hsa-runtime64_FOUND)
add_subdirectory(amdgpu)
-elseif(LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
- message(STATUS "Skipping HSA loader for gpu target, no HSA was detected")
endif()
# The CUDA loader requires LLVM to traverse the ELF image for symbols.
+find_package(CUDAToolkit 11.2 QUIET)
find_package(LLVM QUIET)
-if(CUDAToolkit_FOUND AND LLVM_FOUND AND LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
+if(CUDAToolkit_FOUND AND LLVM_FOUND)
add_subdirectory(nvptx)
-elseif(LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
- message(STATUS "Skipping CUDA loader for gpu target, no CUDA was detected")
endif()
-# Add a custom target to be used for testing.
-set(LIBC_GPU_LOADER_EXECUTABLE "" CACHE STRING "Overriding binary for the GPU loader.")
-if(LIBC_GPU_LOADER_EXECUTABLE)
- add_custom_target(libc.utils.gpu.loader)
- set_target_properties(
- libc.utils.gpu.loader
- PROPERTIES
- EXECUTABLE "${LIBC_GPU_LOADER_EXECUTABLE}"
- )
-elseif(TARGET amdhsa-loader AND LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
+if(TARGET amdhsa-loader AND LIBC_TARGET_ARCHITECTURE_IS_AMDGPU)
add_custom_target(libc.utils.gpu.loader)
add_dependencies(libc.utils.gpu.loader amdhsa-loader)
set_target_properties(
@@ -56,11 +38,10 @@ elseif(TARGET nvptx-loader AND LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
)
endif()
-if(TARGET libc.utils.gpu.loader)
- get_target_property(gpu_loader_tgt libc.utils.gpu.loader "TARGET")
+foreach(gpu_loader_tgt amdhsa-loader nvptx-loader)
if(gpu_loader_tgt)
install(TARGETS ${gpu_loader_tgt}
DESTINATION ${CMAKE_INSTALL_BINDIR}
COMPONENT libc)
endif()
-endif()
+endforeach()
diff --git a/libc/utils/gpu/loader/amdgpu/CMakeLists.txt b/libc/utils/gpu/loader/amdgpu/CMakeLists.txt
index b99319f5040112..97a2de9f8379ab 100644
--- a/libc/utils/gpu/loader/amdgpu/CMakeLists.txt
+++ b/libc/utils/gpu/loader/amdgpu/CMakeLists.txt
@@ -1,5 +1,4 @@
add_executable(amdhsa-loader Loader.cpp)
-add_dependencies(amdhsa-loader libc.src.__support.RPC.rpc)
target_link_libraries(amdhsa-loader
PRIVATE
diff --git a/libc/utils/gpu/loader/nvptx/CMakeLists.txt b/libc/utils/gpu/loader/nvptx/CMakeLists.txt
index e76362a1e8cca6..552edc996291bc 100644
--- a/libc/utils/gpu/loader/nvptx/CMakeLists.txt
+++ b/libc/utils/gpu/loader/nvptx/CMakeLists.txt
@@ -1,5 +1,4 @@
add_executable(nvptx-loader Loader.cpp)
-add_dependencies(nvptx-loader libc.src.__support.RPC.rpc)
if(NOT LLVM_ENABLE_RTTI)
target_compile_options(nvptx-loader PRIVATE -fno-rtti)
diff --git a/libc/utils/gpu/server/CMakeLists.txt b/libc/utils/gpu/server/CMakeLists.txt
index 8712f24de84f7b..6930e145e58e62 100644
--- a/libc/utils/gpu/server/CMakeLists.txt
+++ b/libc/utils/gpu/server/CMakeLists.txt
@@ -5,21 +5,12 @@ target_include_directories(llvmlibc_rpc_server PRIVATE ${LIBC_SOURCE_DIR})
target_include_directories(llvmlibc_rpc_server PUBLIC ${LIBC_SOURCE_DIR}/include)
target_include_directories(llvmlibc_rpc_server PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
-
# Ignore unsupported clang attributes if we're using GCC.
target_compile_options(llvmlibc_rpc_server PUBLIC
$<$<CXX_COMPILER_ID:GNU>:-Wno-attributes>)
target_compile_definitions(llvmlibc_rpc_server PUBLIC
LIBC_NAMESPACE=${LIBC_NAMESPACE})
-# This utility needs to be compiled for the host system when cross compiling.
-if(LLVM_RUNTIMES_TARGET OR LIBC_TARGET_TRIPLE)
- target_compile_options(llvmlibc_rpc_server PUBLIC
- --target=${LLVM_HOST_TRIPLE})
- target_link_libraries(llvmlibc_rpc_server PUBLIC
- "--target=${LLVM_HOST_TRIPLE}")
-endif()
-
# Install the server and associated header.
install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/llvmlibc_rpc_server.h
DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}
diff --git a/llvm/CMakeLists.txt b/llvm/CMakeLists.txt
index 111c8cfa15d828..d0e33c29be58f8 100644
--- a/llvm/CMakeLists.txt
+++ b/llvm/CMakeLists.txt
@@ -173,16 +173,15 @@ endforeach()
set(NEED_LIBC_HDRGEN FALSE)
if("libc" IN_LIST LLVM_ENABLE_RUNTIMES)
set(NEED_LIBC_HDRGEN TRUE)
-else()
- foreach(_name ${LLVM_RUNTIME_TARGETS})
- if("libc" IN_LIST RUNTIMES_${_name}_LLVM_ENABLE_RUNTIMES)
- set(NEED_LIBC_HDRGEN TRUE)
- if("${_name}" STREQUAL "amdgcn-amd-amdhsa" OR "${_name}" STREQUAL "nvptx64-nvidia-cuda")
- set(LLVM_LIBC_GPU_BUILD ON)
- endif()
- endif()
- endforeach()
endif()
+foreach(_name ${LLVM_RUNTIME_TARGETS})
+ if("libc" IN_LIST RUNTIMES_${_name}_LLVM_ENABLE_RUNTIMES)
+ set(NEED_LIBC_HDRGEN TRUE)
+ if("${_name}" STREQUAL "amdgcn-amd-amdhsa" OR "${_name}" STREQUAL "nvptx64-nvidia-cuda")
+ set(LLVM_LIBC_GPU_BUILD ON)
+ endif()
+ endif()
+endforeach()
if(NEED_LIBC_HDRGEN)
# To build the libc runtime, we need to be able to build few libc build
# tools from the "libc" project. So, we add it to the list of enabled
diff --git a/llvm/runtimes/CMakeLists.txt b/llvm/runtimes/CMakeLists.txt
index 9b5e758b6ede57..54ec8bd28d4ca6 100644
--- a/llvm/runtimes/CMakeLists.txt
+++ b/llvm/runtimes/CMakeLists.txt
@@ -358,6 +358,14 @@ function(runtime_register_target name)
endif()
endif()
endforeach()
+ foreach(variable_name ${${name}_extra_args})
+ string(FIND "${variable_name}" "-DRUNTIMES_${extra_name}_" out)
+ if("${out}" EQUAL 0)
+ string(REPLACE "-DRUNTIMES_${extra_name}_" "" new_name ${variable_name})
+ string(REPLACE ";" "|" new_value "${new_name}")
+ list(APPEND ${name}_extra_args "-D${new_value}")
+ endif()
+ endforeach()
endforeach()
set_enable_per_target_runtime_dir()
@@ -438,21 +446,37 @@ if(runtimes)
if(NOT hdrgen_exe)
message(FATAL_ERROR "libc-hdrgen executable missing")
endif()
- set(libc_cmake_args "-DLIBC_HDRGEN_EXE=${hdrgen_exe}"
- "-DLLVM_LIBC_FULL_BUILD=ON")
+ list(APPEND libc_cmake_args "-DLIBC_HDRGEN_EXE=${hdrgen_exe}")
list(APPEND extra_deps ${hdrgen_deps})
- if(LLVM_LIBC_GPU_BUILD)
- list(APPEND libc_cmake_args "-DLLVM_LIBC_GPU_BUILD=ON")
- # The `libc` project may require '-DCUDAToolkit_ROOT' in GPU mode.
- if(CUDAToolkit_ROOT)
- list(APPEND libc_cmake_args "-DCUDAToolkit_ROOT=${CUDAToolkit_ROOT}")
+ endif()
+ if(LLVM_LIBC_GPU_BUILD)
+ list(APPEND libc_cmake_args "-DLLVM_LIBC_GPU_BUILD=ON")
+ if("libc" IN_LIST RUNTIMES_amdgcn-amd-amdhsa_LLVM_ENABLE_RUNTIMES)
+ if(TARGET amdhsa-loader)
+ list(APPEND libc_cmake_args
+ "-DRUNTIMES_amdgcn-amd-amdhsa_LIBC_GPU_LOADER_EXECUTABLE=$<TARGET_FILE:amdhsa-loader>")
+ list(APPEND extra_deps amdhsa-loader amdgpu-arch)
endif()
- foreach(dep clang-offload-packager nvptx-arch amdgpu-arch)
- if(TARGET ${dep})
- list(APPEND extra_deps ${dep})
- endif()
- endforeach()
+ list(APPEND libc_cmake_args "-DRUNTIMES_amdgcn-amd-amdhsa_LLVM_LIBC_FULL_BUILD=ON")
endif()
+ if("libc" IN_LIST RUNTIMES_nvptx64-nvidia-cuda_LLVM_ENABLE_RUNTIMES)
+ if(TARGET nvptx-loader)
+ list(APPEND libc_cmake_args
+ "-DRUNTIMES_nvptx64-nvidia-cuda_LIBC_GPU_LOADER_EXECUTABLE=$<TARGET_FILE:nvptx-loader>")
+ list(APPEND extra_deps nvptx-loader nvptx-arch)
+ endif()
+ list(APPEND libc_cmake_args "-DRUNTIMES_nvptx64-nvidia-cuda_LLVM_LIBC_FULL_BUILD=ON")
+ endif()
+ # The `libc` project may require '-DCUDAToolkit_ROOT' in GPU mode.
+ if(CUDAToolkit_ROOT)
+ list(APPEND libc_cmake_args "-DCUDAToolkit_ROOT=${CUDAToolkit_ROOT}")
+ endif()
+ if(TARGET clang-offload-packager)
+ list(APPEND extra_deps clang-offload-packager)
+ endif()
+ endif()
+ if(LLVM_LIBC_FULL_BUILD)
+ list(APPEND libc_cmake_args "-DLLVM_LIBC_FULL_BUILD=ON")
endif()
if(NOT LLVM_RUNTIME_TARGETS)
runtime_default_target(
|
lntue
reviewed
Mar 10, 2024
libc/CMakeLists.txt
Outdated
| @@ -60,6 +60,10 @@ if(LLVM_LIBC_FULL_BUILD OR LLVM_LIBC_GPU_BUILD) | |||
| message(STATUS "Will use ${LIBC_HDRGEN_EXE} for libc header generation.") | |||
| endif() | |||
| endif() | |||
| # We will build the GPU utilities if we are not doing a runtimes build. | |||
| if((LIBC_TARGET_OS_IS_GPU OR LLVM_LIBC_GPU_BUILD) AND NOT LLVM_RUNTIMES_BUILD) | |||
There was a problem hiding this comment.
Why do you have to skip this when LLVM_RUNTIMES_BUILD is set?
There was a problem hiding this comment.
Because in a full cross-compiling build we don't necessarily have access to the LLVM and CUDA libraries that it requires. I already have around find_package(CUDA) and find_package(LLVM) but it wasn't a sound solution. This is pretty much the same reason we don't build libc-hdrgen during the runtimes build.
jhuber6
force-pushed
the
LoaderTools
branch
2 times, most recently
from
d820309
to
d6b09d3
Compare
lntue
reviewed
Mar 11, 2024
libc/utils/gpu/loader/CMakeLists.txt
Outdated
| @@ -56,11 +37,10 @@ elseif(TARGET nvptx-loader AND LIBC_TARGET_ARCHITECTURE_IS_NVPTX) | |||
| ) | |||
| endif() | |||
|
|
|||
| if(TARGET libc.utils.gpu.loader) | |||
| get_target_property(gpu_loader_tgt libc.utils.gpu.loader "TARGET") | |||
| foreach(gpu_loader_tgt amdhsa-loader nvptx-loader) | |||
| if(gpu_loader_tgt) | |||
There was a problem hiding this comment.
Are they always available to install now? If so remove the if statement.
There was a problem hiding this comment.
No, they can only be built if the user had HSA or CUDA installed on their machine in a location it could find.
There was a problem hiding this comment.
So, we have a lot of other tools that depend on CUDA / HSA that are installed unconditionally (see amdgpu-arch) but those use dlopen. I could do the same for my loader utilities but that's a lot of work for what is essentially a test tool.
There was a problem hiding this comment.
I mean ${gpu_loader_tgt} is either "amdhsa-loader" or "nvptx-loader", so the check if(gpu_loader_tgt) is always true, isn't it?
There was a problem hiding this comment.
You are right, I forgot to put TARGET there.
There was a problem hiding this comment.
Thanks for catching that, people who didn't have both CUDA and HSA on their system would've gotten install errors.
Summary: The libc build has a few utilties that need to be built before we can do everything in the full build. The one requirement currently is the `libc-hdrgen` binary. If we are doing a full build runtimes mode we first add `libc` to the projects list and then only use the `projects` portion to buld the `libc` portion. We also use utilities for the GPU build, namely the loader utilities. Previously we would build these tools on-demand inside of the cross-build, which tool some hacky workarounds for the dependency finding and target triple. This patch instead just builds them similarly to libc-hdrgen and then passses them in. We now either pass it manually it it was built, or just look it up like we do with the other `clang` tools. Depends on llvm#84664
lntue
approved these changes
Mar 11, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
The libc build has a few utilties that need to be built before we can do
everything in the full build. The one requirement currently is the
libc-hdrgenbinary. If we are doing a full build runtimes mode wefirst add
libcto the projects list and then only use theprojectsportion to buld the
libcportion. We also use utilities for the GPUbuild, namely the loader utilities. Previously we would build these
tools on-demand inside of the cross-build, which tool some hacky
workarounds for the dependency finding and target triple. This patch
instead just builds them similarly to libc-hdrgen and then passses them
in. We now either pass it manually it it was built, or just look it up
like we do with the other
clangtools.Depends on #84664