-
Notifications
You must be signed in to change notification settings - Fork 12k
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.
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
[libc] Only add '-fno-builtin-*' on the entrypoints that use them #100481
Conversation
Summary: The GPU build needs to be able to inline stuff in LTO. Builtin transformations cause problems on the functions that the optimizer does heavy libcall recognition on. Previously we moved to using `-fno-builtin-*` to allow us to only disable the problematic ones. However, this still didn't allow inlining because each function had the attribute that told the inliner not to inlining a nobuiltin function into a non-nobuiltin function This patch fixes that by only applying these attributes to the entrypoints that define them. That is enough to prevent recursive calls within the definitoins themselves.
@llvm/pr-subscribers-libc Author: Joseph Huber (jhuber6) ChangesSummary: This patch fixes that by only applying these attributes to the Full diff: https://github.com/llvm/llvm-project/pull/100481.diff 2 Files Affected:
diff --git a/libc/cmake/modules/LLVMLibCCompileOptionRules.cmake b/libc/cmake/modules/LLVMLibCCompileOptionRules.cmake
index 0f1ef6a575277..9fc10375a1d37 100644
--- a/libc/cmake/modules/LLVMLibCCompileOptionRules.cmake
+++ b/libc/cmake/modules/LLVMLibCCompileOptionRules.cmake
@@ -115,16 +115,7 @@ function(_get_common_compile_options output_var flags)
list(APPEND compile_options "-ffixed-point")
endif()
- # Builtin recognition causes issues when trying to implement the builtin
- # functions themselves. The GPU backends do not use libcalls so we disable
- # the known problematic ones. This allows inlining during LTO linking.
- if(LIBC_TARGET_OS_IS_GPU)
- set(libc_builtins bcmp strlen memmem bzero memcmp memcpy memmem memmove
- memset strcmp strstr)
- foreach(builtin ${libc_builtins})
- list(APPEND compile_options "-fno-builtin-${builtin}")
- endforeach()
- else()
+ if(NOT LIBC_TARGET_OS_IS_GPU)
list(APPEND compile_options "-fno-builtin")
endif()
diff --git a/libc/cmake/modules/LLVMLibCObjectRules.cmake b/libc/cmake/modules/LLVMLibCObjectRules.cmake
index 2d3db38ecd8a3..68b5ed1ed51c0 100644
--- a/libc/cmake/modules/LLVMLibCObjectRules.cmake
+++ b/libc/cmake/modules/LLVMLibCObjectRules.cmake
@@ -279,6 +279,17 @@ function(create_entrypoint_object fq_target_name)
add_dependencies(${fq_target_name} ${full_deps_list})
target_link_libraries(${fq_target_name} ${full_deps_list})
+ # Builtin recognition causes issues when trying to implement the builtin
+ # functions themselves. The GPU backends do not use libcalls so we disable the
+ # known problematic ones on the entrypoints that implement them.
+ if(LIBC_TARGET_OS_IS_GPU)
+ set(libc_builtins bcmp strlen memmem bzero memcmp memcpy memmem memmove
+ memset strcmp strstr)
+ if(${ADD_ENTRYPOINT_OBJ_NAME} IN_LIST libc_builtins)
+ target_compile_options(${fq_target_name} PRIVATE -fno-builtin-${ADD_ENTRYPOINT_OBJ_NAME})
+ endif()
+ endif()
+
set_target_properties(
${fq_target_name}
PROPERTIES
|
…vm#100481) Summary: The GPU build needs to be able to inline stuff in LTO. Builtin transformations cause problems on the functions that the optimizer does heavy libcall recognition on. Previously we moved to using `-fno-builtin-*` to allow us to only disable the problematic ones. However, this still didn't allow inlining because each function had the attribute that told the inliner not to inlining a nobuiltin function into a non-nobuiltin function This patch fixes that by only applying these attributes to the entrypoints that define them. That is enough to prevent recursive calls within the definitoins themselves. (cherry picked from commit 8e43acb)
…00481) Summary: The GPU build needs to be able to inline stuff in LTO. Builtin transformations cause problems on the functions that the optimizer does heavy libcall recognition on. Previously we moved to using `-fno-builtin-*` to allow us to only disable the problematic ones. However, this still didn't allow inlining because each function had the attribute that told the inliner not to inlining a nobuiltin function into a non-nobuiltin function This patch fixes that by only applying these attributes to the entrypoints that define them. That is enough to prevent recursive calls within the definitoins themselves. Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60250630
…vm#100481) Summary: The GPU build needs to be able to inline stuff in LTO. Builtin transformations cause problems on the functions that the optimizer does heavy libcall recognition on. Previously we moved to using `-fno-builtin-*` to allow us to only disable the problematic ones. However, this still didn't allow inlining because each function had the attribute that told the inliner not to inlining a nobuiltin function into a non-nobuiltin function This patch fixes that by only applying these attributes to the entrypoints that define them. That is enough to prevent recursive calls within the definitoins themselves. (cherry picked from commit 8e43acb)
Seems like we can probably do this for non-GPU targets, too? I'm very interested in supporting LTO for CPU targets as well. |
Might be a little more broken because the GPU targets intentionally don't emit any libcalls in their backends (except a few atomic ones I think?). I'm not an expert on those targets however, so we could probably find some amount that worked. |
Summary:
The GPU build needs to be able to inline stuff in LTO. Builtin
transformations cause problems on the functions that the optimizer does
heavy libcall recognition on. Previously we moved to using
-fno-builtin-*
to allow us to only disable the problematic ones.However, this still didn't allow inlining because each function had the
attribute that told the inliner not to inlining a nobuiltin function
into a non-nobuiltin function
This patch fixes that by only applying these attributes to the
entrypoints that define them. That is enough to prevent recursive calls
within the definitoins themselves.