Skip to content

Commit

Permalink
[libc] Fix linking of AMDGPU device runtime control constants for math
Browse files Browse the repository at this point in the history 
Summary:
Currently, `libc` temporarily provides math by linking against existing
vendor implementations. To use the AMDGPU DeviceRTL we need to define a
handful of control constants that alter behaviour for architecture
specific things. Previously these were marked `extern const` because
they must be present when we link-in the vendor bitcode library.
However, this causes linker errors if more than one math function was
used.

This patch fixes the issue by marking these functions as used and inline
on top of being external. This means that they are linkable, but it
gives us `linkonce_odr` semantics. The downside is that these globals
won't be optimized out, but it allows us to perform constant propagation
on them unlike using `weak`.
  • Loading branch information
@jhuber6
jhuber6 committed Sep 14, 2023
1 parent 089b811 commit d11fae1
Showing 1 changed file with 53 additions and 41 deletions.
94 changes: 53 additions & 41 deletions libc/src/math/gpu/vendor/amdgpu/platform.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,8 @@
#ifndef LLVM_LIBC_SRC_MATH_GPU_AMDGPU_PLATFORM_H
#define LLVM_LIBC_SRC_MATH_GPU_AMDGPU_PLATFORM_H

#include "src/__support/macros/attributes.h"

#include <stdint.h>

namespace __llvm_libc {
Expand All @@ -19,96 +21,106 @@ namespace __llvm_libc {
extern "C" {

// Disable unsafe math optimizations in the implementation.
extern const uint8_t __oclc_unsafe_math_opt = 0;
extern const LIBC_INLINE uint8_t __oclc_unsafe_math_opt = 0;

// Disable denormalization at zero optimizations in the implementation.
extern const uint8_t __oclc_daz_opt = 0;
extern const LIBC_INLINE uint8_t __oclc_daz_opt = 0;

// Disable rounding optimizations for 32-bit square roots.
extern const uint8_t __oclc_correctly_rounded_sqrt32 = 1;
extern const LIBC_INLINE uint8_t __oclc_correctly_rounded_sqrt32 = 1;

// Disable finite math optimizations.
extern const uint8_t __oclc_finite_only_opt = 0;
extern const LIBC_INLINE uint8_t __oclc_finite_only_opt = 0;

#if defined(__gfx700__)
extern const uint32_t __oclc_ISA_version = 7000;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7000;
#elif defined(__gfx701__)
extern const uint32_t __oclc_ISA_version = 7001;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7001;
#elif defined(__gfx702__)
extern const uint32_t __oclc_ISA_version = 7002;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7002;
#elif defined(__gfx703__)
extern const uint32_t __oclc_ISA_version = 7003;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7003;
#elif defined(__gfx704__)
extern const uint32_t __oclc_ISA_version = 7004;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7004;
#elif defined(__gfx705__)
extern const uint32_t __oclc_ISA_version = 7005;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 7005;
#elif defined(__gfx801__)
extern const uint32_t __oclc_ISA_version = 8001;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 8001;
#elif defined(__gfx802__)
extern const uint32_t __oclc_ISA_version = 8002;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 8002;
#elif defined(__gfx803__)
extern const uint32_t __oclc_ISA_version = 8003;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 8003;
#elif defined(__gfx805__)
extern const uint32_t __oclc_ISA_version = 8005;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 8005;
#elif defined(__gfx810__)
extern const uint32_t __oclc_ISA_version = 8100;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 8100;
#elif defined(__gfx900__)
extern const uint32_t __oclc_ISA_version = 9000;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9000;
#elif defined(__gfx902__)
extern const uint32_t __oclc_ISA_version = 9002;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9002;
#elif defined(__gfx904__)
extern const uint32_t __oclc_ISA_version = 9004;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9004;
#elif defined(__gfx906__)
extern const uint32_t __oclc_ISA_version = 9006;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9006;
#elif defined(__gfx908__)
extern const uint32_t __oclc_ISA_version = 9008;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9008;
#elif defined(__gfx909__)
extern const uint32_t __oclc_ISA_version = 9009;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9009;
#elif defined(__gfx90a__)
extern const uint32_t __oclc_ISA_version = 9010;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9010;
#elif defined(__gfx90c__)
extern const uint32_t __oclc_ISA_version = 9012;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9012;
#elif defined(__gfx940__)
extern const uint32_t __oclc_ISA_version = 9400;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 9400;
#elif defined(__gfx1010__)
extern const uint32_t __oclc_ISA_version = 10100;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10100;
#elif defined(__gfx1011__)
extern const uint32_t __oclc_ISA_version = 10101;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10101;
#elif defined(__gfx1012__)
extern const uint32_t __oclc_ISA_version = 10102;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10102;
#elif defined(__gfx1013__)
extern const uint32_t __oclc_ISA_version = 10103;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10103;
#elif defined(__gfx1030__)
extern const uint32_t __oclc_ISA_version = 10300;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10300;
#elif defined(__gfx1031__)
extern const uint32_t __oclc_ISA_version = 10301;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10301;
#elif defined(__gfx1032__)
extern const uint32_t __oclc_ISA_version = 10302;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10302;
#elif defined(__gfx1033__)
extern const uint32_t __oclc_ISA_version = 10303;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10303;
#elif defined(__gfx1034__)
extern const uint32_t __oclc_ISA_version = 10304;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10304;
#elif defined(__gfx1035__)
extern const uint32_t __oclc_ISA_version = 10305;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10305;
#elif defined(__gfx1036__)
extern const uint32_t __oclc_ISA_version = 10306;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 10306;
#elif defined(__gfx1100__)
extern const uint32_t __oclc_ISA_version = 11000;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11000;
#elif defined(__gfx1101__)
extern const uint32_t __oclc_ISA_version = 11001;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11001;
#elif defined(__gfx1102__)
extern const uint32_t __oclc_ISA_version = 11002;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11002;
#elif defined(__gfx1103__)
extern const uint32_t __oclc_ISA_version = 11003;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11003;
#elif defined(__gfx1150__)
extern const uint32_t __oclc_ISA_version = 11500;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11500;
#elif defined(__gfx1151__)
extern const uint32_t __oclc_ISA_version = 11501;
extern const LIBC_INLINE uint32_t __oclc_ISA_version = 11501;
#else
#error "Unknown AMDGPU architecture"
#endif
}

// These aliases cause clang to emit the control constants with ODR linkage.
// This allows us to link against the symbols without preventing them from being
// optimized out or causing symbol collisions.
[[gnu::alias("__oclc_unsafe_math_opt")]] const uint8_t __oclc_unsafe_math_opt__;
[[gnu::alias("__oclc_daz_opt")]] const uint8_t __oclc_daz_opt__;
[[gnu::alias("__oclc_correctly_rounded_sqrt32")]] const uint8_t
__oclc_correctly_rounded_sqrt32__;
[[gnu::alias("__oclc_finite_only_opt")]] const uint8_t __oclc_finite_only_opt__;
[[gnu::alias("__oclc_ISA_version")]] const uint32_t __oclc_ISA_version__;

} // namespace __llvm_libc

#endif // LLVM_LIBC_SRC_MATH_GPU_AMDGPU_PLATFORM_H

0 comments on commit d11fae1

Please sign in to comment.