[libclc] Remove (vload|vstore)_half helpers #137181
Merged
Conversation
This file contains hidden or 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
These were only being used when compiling with versions of clang older than clang 6. As such they were essentially unsuppored and untested. This somewhat simplifies the codebase, producing fewer helper functions in the final builtins library. There's no change to any of the targets' bytecode other than removing these helper functions.
|
@llvm/pr-subscribers-backend-amdgpu Author: Fraser Cormack (frasercrmck) ChangesThese were only being used when compiling with versions of clang older than clang 6. As such they were essentially unsuppored and untested. This somewhat simplifies the codebase, producing fewer helper functions in the final builtins library. There's no change to any of the targets' bytecode other than removing these helper functions. Full diff: https://github.com/llvm/llvm-project/pull/137181.diff 13 Files Affected:
diff --git a/libclc/amdgpu/lib/SOURCES_3.9 b/libclc/amdgpu/lib/SOURCES_3.9
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/amdgpu/lib/SOURCES_3.9
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/amdgpu/lib/SOURCES_4.0 b/libclc/amdgpu/lib/SOURCES_4.0
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/amdgpu/lib/SOURCES_4.0
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/amdgpu/lib/SOURCES_5.0 b/libclc/amdgpu/lib/SOURCES_5.0
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/amdgpu/lib/SOURCES_5.0
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/amdgpu/lib/shared/vload_half_helpers.ll b/libclc/amdgpu/lib/shared/vload_half_helpers.ll
deleted file mode 100644
index bf7d544afbf55..0000000000000
--- a/libclc/amdgpu/lib/shared/vload_half_helpers.ll
+++ /dev/null
@@ -1,31 +0,0 @@
-;;===----------------------------------------------------------------------===;;
-;
-; Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-; See https://llvm.org/LICENSE.txt for license information.
-; SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-;
-;;===----------------------------------------------------------------------===;;
-
-define float @__clc_vload_half_float_helper__private(half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(0)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__global(half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(1)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__local(half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(3)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__constant(half addrspace(2)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(2)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
diff --git a/libclc/amdgpu/lib/shared/vstore_half_helpers.ll b/libclc/amdgpu/lib/shared/vstore_half_helpers.ll
deleted file mode 100644
index 5ec193d6d40a0..0000000000000
--- a/libclc/amdgpu/lib/shared/vstore_half_helpers.ll
+++ /dev/null
@@ -1,43 +0,0 @@
-;;===----------------------------------------------------------------------===;;
-;
-; Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-; See https://llvm.org/LICENSE.txt for license information.
-; SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-;
-;;===----------------------------------------------------------------------===;;
-
-define void @__clc_vstore_half_float_helper__private(float %data, half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(0)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_float_helper__global(float %data, half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(1)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_float_helper__local(float %data, half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(3)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__private(double %data, half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(0)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__global(double %data, half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(1)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__local(double %data, half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(3)* %ptr
- ret void
-}
diff --git a/libclc/generic/lib/shared/vload.cl b/libclc/generic/lib/shared/vload.cl
index a0306c500d5cd..05737aaa88b71 100644
--- a/libclc/generic/lib/shared/vload.cl
+++ b/libclc/generic/lib/shared/vload.cl
@@ -81,18 +81,7 @@ VLOAD_ADDR_SPACES(half)
/* vload_half are legal even without cl_khr_fp16 */
/* no vload_half for double */
-#if __clang_major__ < 6
-float __clc_vload_half_float_helper__constant(const __constant half *);
-float __clc_vload_half_float_helper__global(const __global half *);
-float __clc_vload_half_float_helper__local(const __local half *);
-float __clc_vload_half_float_helper__private(const __private half *);
-
-#define VEC_LOAD1(val, AS) \
- val = __clc_vload_half_float_helper##AS(&mem[offset++]);
-#else
#define VEC_LOAD1(val, AS) val = __builtin_load_halff(&mem[offset++]);
-#endif
-
#define VEC_LOAD2(val, AS) \
VEC_LOAD1(val.lo, AS) \
VEC_LOAD1(val.hi, AS)
diff --git a/libclc/generic/lib/shared/vstore.cl b/libclc/generic/lib/shared/vstore.cl
index 525f3d08bf0d8..0c0513f08a778 100644
--- a/libclc/generic/lib/shared/vstore.cl
+++ b/libclc/generic/lib/shared/vstore.cl
@@ -75,57 +75,35 @@ VSTORE_ADDR_SPACES(double)
VSTORE_ADDR_SPACES(half)
#endif
-/* vstore_half are legal even without cl_khr_fp16 */
-#if __clang_major__ < 6
-#define DECLARE_HELPER(STYPE, AS, builtin) \
- void __clc_vstore_half_##STYPE##_helper##AS(STYPE, AS half *);
-#else
-#define DECLARE_HELPER(STYPE, AS, __builtin) \
- _CLC_DEF void __clc_vstore_half_##STYPE##_helper##AS(STYPE s, AS half *d) { \
- __builtin(s, d); \
- }
-#endif
-
-DECLARE_HELPER(float, __private, __builtin_store_halff);
-DECLARE_HELPER(float, __global, __builtin_store_halff);
-DECLARE_HELPER(float, __local, __builtin_store_halff);
-
-#ifdef cl_khr_fp64
-DECLARE_HELPER(double, __private, __builtin_store_half);
-DECLARE_HELPER(double, __global, __builtin_store_half);
-DECLARE_HELPER(double, __local, __builtin_store_half);
-#endif
-
-#define VEC_STORE1(STYPE, AS, val, ROUNDF) \
- __clc_vstore_half_##STYPE##_helper##AS(ROUNDF(val), &mem[offset++]);
-
-#define VEC_STORE2(STYPE, AS, val, ROUNDF) \
- VEC_STORE1(STYPE, AS, val.lo, ROUNDF) \
- VEC_STORE1(STYPE, AS, val.hi, ROUNDF)
-#define VEC_STORE3(STYPE, AS, val, ROUNDF) \
- VEC_STORE1(STYPE, AS, val.s0, ROUNDF) \
- VEC_STORE1(STYPE, AS, val.s1, ROUNDF) \
- VEC_STORE1(STYPE, AS, val.s2, ROUNDF)
-#define VEC_STORE4(STYPE, AS, val, ROUNDF) \
- VEC_STORE2(STYPE, AS, val.lo, ROUNDF) \
- VEC_STORE2(STYPE, AS, val.hi, ROUNDF)
-#define VEC_STORE8(STYPE, AS, val, ROUNDF) \
- VEC_STORE4(STYPE, AS, val.lo, ROUNDF) \
- VEC_STORE4(STYPE, AS, val.hi, ROUNDF)
-#define VEC_STORE16(STYPE, AS, val, ROUNDF) \
- VEC_STORE8(STYPE, AS, val.lo, ROUNDF) \
- VEC_STORE8(STYPE, AS, val.hi, ROUNDF)
-
-#define __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS, ROUNDF) \
+#define VEC_STORE1(val, ROUNDF, BUILTIN) BUILTIN(ROUNDF(val), &mem[offset++]);
+
+#define VEC_STORE2(val, ROUNDF, BUILTIN) \
+ VEC_STORE1(val.lo, ROUNDF, BUILTIN) \
+ VEC_STORE1(val.hi, ROUNDF, BUILTIN)
+#define VEC_STORE3(val, ROUNDF, BUILTIN) \
+ VEC_STORE1(val.s0, ROUNDF, BUILTIN) \
+ VEC_STORE1(val.s1, ROUNDF, BUILTIN) \
+ VEC_STORE1(val.s2, ROUNDF, BUILTIN)
+#define VEC_STORE4(val, ROUNDF, BUILTIN) \
+ VEC_STORE2(val.lo, ROUNDF, BUILTIN) \
+ VEC_STORE2(val.hi, ROUNDF, BUILTIN)
+#define VEC_STORE8(val, ROUNDF, BUILTIN) \
+ VEC_STORE4(val.lo, ROUNDF, BUILTIN) \
+ VEC_STORE4(val.hi, ROUNDF, BUILTIN)
+#define VEC_STORE16(val, ROUNDF, BUILTIN) \
+ VEC_STORE8(val.lo, ROUNDF, BUILTIN) \
+ VEC_STORE8(val.hi, ROUNDF, BUILTIN)
+
+#define __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, AS, ROUNDF, BUILTIN) \
_CLC_OVERLOAD _CLC_DEF void vstore_half##SUFFIX(TYPE vec, size_t offset, \
AS half *mem) { \
offset *= VEC_SIZE; \
- VEC_STORE##VEC_SIZE(STYPE, AS, vec, ROUNDF) \
+ VEC_STORE##VEC_SIZE(vec, ROUNDF, BUILTIN) \
} \
_CLC_OVERLOAD _CLC_DEF void vstorea_half##SUFFIX(TYPE vec, size_t offset, \
AS half *mem) { \
offset *= OFFSET; \
- VEC_STORE##VEC_SIZE(STYPE, AS, vec, ROUNDF) \
+ VEC_STORE##VEC_SIZE(vec, ROUNDF, BUILTIN) \
}
_CLC_DEF _CLC_OVERLOAD float __clc_noop(float x) { return x; }
@@ -246,15 +224,15 @@ _CLC_DEF _CLC_OVERLOAD double __clc_rte(double x) {
}
#endif
-#define __XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) \
- __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_noop) \
- __FUNC(SUFFIX##_rtz, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtz) \
- __FUNC(SUFFIX##_rtn, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtn) \
- __FUNC(SUFFIX##_rtp, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtp) \
- __FUNC(SUFFIX##_rte, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rte)
+#define __XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, AS, BUILTIN) \
+ __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, AS, __clc_noop, BUILTIN) \
+ __FUNC(SUFFIX##_rtz, VEC_SIZE, OFFSET, TYPE, AS, __clc_rtz, BUILTIN) \
+ __FUNC(SUFFIX##_rtn, VEC_SIZE, OFFSET, TYPE, AS, __clc_rtn, BUILTIN) \
+ __FUNC(SUFFIX##_rtp, VEC_SIZE, OFFSET, TYPE, AS, __clc_rtp, BUILTIN) \
+ __FUNC(SUFFIX##_rte, VEC_SIZE, OFFSET, TYPE, AS, __clc_rte, BUILTIN)
-#define FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) \
- __XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS)
+#define FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, AS, BUILTIN) \
+ __XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, AS, BUILTIN)
#define __CLC_BODY "vstore_half.inc"
#include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/shared/vstore_half.inc b/libclc/generic/lib/shared/vstore_half.inc
index 138c19ae78b3f..78f137f76e832 100644
--- a/libclc/generic/lib/shared/vstore_half.inc
+++ b/libclc/generic/lib/shared/vstore_half.inc
@@ -8,6 +8,15 @@
// This does not exist for fp16
#if __CLC_FPSIZE > 16
+
+#if __CLC_FPSIZE == 32
+#define STORE_HALF_BUILTIN __builtin_store_halff
+#elif __CLC_FPSIZE == 64
+#define STORE_HALF_BUILTIN __builtin_store_half
+#else
+#error "Invalid FP size"
+#endif
+
#ifndef __CLC_SCALAR
#if __CLC_VECSIZE == 3
@@ -16,17 +25,20 @@
#define __CLC_OFFSET __CLC_VECSIZE
#endif
-FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE,
- __CLC_SCALAR_GENTYPE, __private);
-FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE,
- __CLC_SCALAR_GENTYPE, __local);
-FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE,
- __CLC_SCALAR_GENTYPE, __global);
+FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __private,
+ STORE_HALF_BUILTIN);
+FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __local,
+ STORE_HALF_BUILTIN);
+FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __global,
+ STORE_HALF_BUILTIN);
#undef __CLC_OFFSET
#else
-FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __private);
-FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __local);
-FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __global);
+FUNC(, 1, 1, __CLC_GENTYPE, __private, STORE_HALF_BUILTIN);
+FUNC(, 1, 1, __CLC_GENTYPE, __local, STORE_HALF_BUILTIN);
+FUNC(, 1, 1, __CLC_GENTYPE, __global, STORE_HALF_BUILTIN);
#endif
+
+#undef STORE_HALF_BUILTIN
+
#endif
diff --git a/libclc/ptx/lib/SOURCES_3.9 b/libclc/ptx/lib/SOURCES_3.9
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/ptx/lib/SOURCES_3.9
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/ptx/lib/SOURCES_4.0 b/libclc/ptx/lib/SOURCES_4.0
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/ptx/lib/SOURCES_4.0
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/ptx/lib/SOURCES_5.0 b/libclc/ptx/lib/SOURCES_5.0
deleted file mode 100644
index 69c5e5ce9fbac..0000000000000
--- a/libclc/ptx/lib/SOURCES_5.0
+++ /dev/null
@@ -1,2 +0,0 @@
-shared/vload_half_helpers.ll
-shared/vstore_half_helpers.ll
diff --git a/libclc/ptx/lib/shared/vload_half_helpers.ll b/libclc/ptx/lib/shared/vload_half_helpers.ll
deleted file mode 100644
index 5fbae19fde287..0000000000000
--- a/libclc/ptx/lib/shared/vload_half_helpers.ll
+++ /dev/null
@@ -1,31 +0,0 @@
-;;===----------------------------------------------------------------------===;;
-;
-; Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-; See https://llvm.org/LICENSE.txt for license information.
-; SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-;
-;;===----------------------------------------------------------------------===;;
-
-define float @__clc_vload_half_float_helper__private(half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(0)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__global(half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(1)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__local(half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(3)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
-
-define float @__clc_vload_half_float_helper__constant(half addrspace(4)* nocapture %ptr) nounwind alwaysinline {
- %data = load half, half addrspace(4)* %ptr
- %res = fpext half %data to float
- ret float %res
-}
diff --git a/libclc/ptx/lib/shared/vstore_half_helpers.ll b/libclc/ptx/lib/shared/vstore_half_helpers.ll
deleted file mode 100644
index 5ec193d6d40a0..0000000000000
--- a/libclc/ptx/lib/shared/vstore_half_helpers.ll
+++ /dev/null
@@ -1,43 +0,0 @@
-;;===----------------------------------------------------------------------===;;
-;
-; Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-; See https://llvm.org/LICENSE.txt for license information.
-; SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-;
-;;===----------------------------------------------------------------------===;;
-
-define void @__clc_vstore_half_float_helper__private(float %data, half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(0)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_float_helper__global(float %data, half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(1)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_float_helper__local(float %data, half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc float %data to half
- store half %res, half addrspace(3)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__private(double %data, half addrspace(0)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(0)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__global(double %data, half addrspace(1)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(1)* %ptr
- ret void
-}
-
-define void @__clc_vstore_half_double_helper__local(double %data, half addrspace(3)* nocapture %ptr) nounwind alwaysinline {
- %res = fptrunc double %data to half
- store half %res, half addrspace(3)* %ptr
- ret void
-}
|
arsenm
approved these changes
Apr 24, 2025
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.
These were only being used when compiling with versions of clang older than clang 6. As such they were essentially unsuppored and untested.
This somewhat simplifies the codebase, producing fewer helper functions in the final builtins library. There's no change to any of the targets' bytecode other than removing these helper functions.