[AMDGPU] Modifies builtin def to take _Float16('x') for both HIP/C++ and for OpenCL #167652
+32
−17
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
llvmbot
added
clang
Member
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-backend-amdgpu Author: Rana Pratap Reddy (ranapratap55) ChangesFor extended imges insts amdgcn_image_sample_/gather4 builtins, using 'x' in the builtin def so that it will take _Float16 for both HIP/C++ and OpenCL. Full diff: https://github.com/llvm/llvm-project/pull/167652.diff 3 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 2b6fcb1fd479b..f4bd9e0bc1bd2 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -968,45 +968,45 @@ TARGET_BUILTIN(__builtin_amdgcn_image_sample_3d_v4f16_f32, "V4hifffQtV4ibii", "n
TARGET_BUILTIN(__builtin_amdgcn_image_sample_cube_v4f32_f32, "V4fifffQtV4ibii", "nc", "image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_cube_v4f16_f32, "V4hifffQtV4ibii", "nc", "image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1d_v4f32_f32, "V4fifQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1d_v4f16_f32, "V4eifQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1d_v4f16_f32, "V4xifQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1darray_v4f32_f32, "V4fiffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1darray_v4f16_f32, "V4eiffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_1darray_v4f16_f32, "V4xiffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2d_f32_f32, "fiffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2d_v4f32_f32, "V4fiffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2d_v4f16_f32, "V4eiffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2d_v4f16_f32, "V4xiffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2darray_f32_f32, "fifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2darray_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2darray_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_2darray_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_3d_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_3d_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_3d_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_cube_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_cube_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_lz_cube_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1d_v4f32_f32, "V4fiffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1d_v4f16_f32, "V4eiffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1d_v4f16_f32, "V4xiffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1darray_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1darray_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_1darray_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2d_f32_f32, "fifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2d_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2d_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2d_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2darray_f32_f32, "fiffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2darray_v4f32_f32, "V4fiffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2darray_v4f16_f32, "V4eiffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_2darray_v4f16_f32, "V4xiffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_3d_v4f32_f32, "V4fiffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_3d_v4f16_f32, "V4eiffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_3d_v4f16_f32, "V4xiffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_cube_v4f32_f32, "V4fiffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_cube_v4f16_f32, "V4eiffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_l_cube_v4f16_f32, "V4xiffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1d_v4f32_f32, "V4fifffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1d_v4f16_f32, "V4eifffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1d_v4f16_f32, "V4xifffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1darray_v4f32_f32, "V4fiffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1darray_v4f16_f32, "V4eiffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_1darray_v4f16_f32, "V4xiffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2d_f32_f32, "fiffffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2d_v4f32_f32, "V4fiffffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2d_v4f16_f32, "V4eiffffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2d_v4f16_f32, "V4xiffffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2darray_f32_f32, "fifffffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2darray_v4f32_f32, "V4fifffffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2darray_v4f16_f32, "V4eifffffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_2darray_v4f16_f32, "V4xifffffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_3d_v4f32_f32, "V4fifffffffffQtV4ibii", "nc", "extended-image-insts")
-TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_3d_v4f16_f32, "V4eifffffffffQtV4ibii", "nc", "extended-image-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_sample_d_3d_v4f16_f32, "V4xifffffffffQtV4ibii", "nc", "extended-image-insts")
TARGET_BUILTIN(__builtin_amdgcn_image_gather4_lz_2d_v4f32_f32, "V4fiffQtV4ibii", "nc", "extended-image-insts")
#undef BUILTIN
diff --git a/clang/test/SemaOpenCL/builtins-extended-image-param-gfx1100-err.cl b/clang/test/SemaOpenCL/builtins-extended-image-param-gfx1100-err.cl
index 47dbdd4e51782..4ed65e6e32f93 100644
--- a/clang/test/SemaOpenCL/builtins-extended-image-param-gfx1100-err.cl
+++ b/clang/test/SemaOpenCL/builtins-extended-image-param-gfx1100-err.cl
@@ -5,7 +5,7 @@
typedef int int4 __attribute__((ext_vector_type(4)));
typedef float float4 __attribute__((ext_vector_type(4)));
-typedef half half4 __attribute__((ext_vector_type(4)));
+typedef _Float16 half4 __attribute__((ext_vector_type(4)));
float4 test_amdgcn_image_gather4_lz_2d_v4f32_f32_r(float4 v4f32, float f32, int i32, __amdgpu_texture_t tex, int4 vec4i32) {
diff --git a/clang/test/SemaOpenCL/builtins-extended-image-param-gfx942-err.cl b/clang/test/SemaOpenCL/builtins-extended-image-param-gfx942-err.cl
index e60f8c70dc7c4..95818b0aea383 100644
--- a/clang/test/SemaOpenCL/builtins-extended-image-param-gfx942-err.cl
+++ b/clang/test/SemaOpenCL/builtins-extended-image-param-gfx942-err.cl
@@ -5,7 +5,7 @@
typedef int int4 __attribute__((ext_vector_type(4)));
typedef float float4 __attribute__((ext_vector_type(4)));
-typedef half half4 __attribute__((ext_vector_type(4)));
+typedef _Float16 half4 __attribute__((ext_vector_type(4)));
float4 test_amdgcn_image_gather4_lz_2d_v4f32_f32_r(float4 v4f32, float f32, int i32, __amdgpu_texture_t tex, int4 vec4i32) {
|
jmmartinez
reviewed
Nov 12, 2025
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
ranapratap55
force-pushed
the
users/ranapratap55/extended-image-fp16-openCL-fix
branch
from
f2d1093 to
2f0b67f
Compare
ranapratap55
force-pushed
the
users/ranapratap55/extended-image-fp16-openCL-fix
branch
from
2f0b67f to
4d54f17
Compare
arsenm
reviewed
Nov 12, 2025
| typedef int int4 __attribute__((ext_vector_type(4))); | ||
| typedef float float4 __attribute__((ext_vector_type(4))); | ||
| typedef half half4 __attribute__((ext_vector_type(4))); | ||
| typedef _Float16 half4 __attribute__((ext_vector_type(4))); |
Contributor
There was a problem hiding this comment.
I would expect these tests to not need changing. Can you add some reference tests for what happens when you use half
Contributor
Author
There was a problem hiding this comment.
I've tested it using 'half' but got below error.
for test:
typedef half half4 __attribute__((ext_vector_type(4)));
return __builtin_amdgcn_image_sample_d_2darray_v4f16_f32(100, f32, f32, f32, f32, f32, f32, f32, tex, vec4i32, 0, 101, 121);
output:
error: returning '__attribute__((__vector_size__(4 * sizeof(_Float16)))) _Float16' (vector of 4 '_Float16' values) from a function with incompatible result type 'half4' (vector of 4 'half' values)
Contributor
There was a problem hiding this comment.
Ouch that's a problem. So _Float16 is convertible to half but it's not the case for the vector types (and similarly for other implicit conversions in fact: https://godbolt.org/z/3Yx9zo63Y).
Contributor
There was a problem hiding this comment.
There is a -flax-vector-conversions flag that we could extend for the _Float16<->half case; but in general vector conversions are disabled in OpenCL.
ExprResult Sema::CheckExtVectorCast(SourceRange R, QualType DestTy,
Expr *CastExpr, CastKind &Kind) {
assert(DestTy->isExtVectorType() && "Not an extended vector type!");
QualType SrcTy = CastExpr->getType();
// If SrcTy is a VectorType, the total size must match to explicitly cast to
// an ExtVectorType.
// In OpenCL, casts between vectors of different types are not allowed.
// (See OpenCL 6.2).
if (SrcTy->isVectorType()) {
if (!areLaxCompatibleVectorTypes(SrcTy, DestTy) ||
(getLangOpts().OpenCL &&
!Context.hasSameUnqualifiedType(DestTy, SrcTy))) {
Diag(R.getBegin(),diag::err_invalid_conversion_between_ext_vectors)
<< DestTy << SrcTy << R;
return ExprError();
}
Kind = CK_BitCast;
return CastExpr;
}
// All non-pointer scalars can be cast to ExtVector type. The appropriate
// conversion will take place first from scalar to elt type, and then
// splat from elt type to vector.
if (SrcTy->isPointerType())
return Diag(R.getBegin(),
diag::err_invalid_conversion_between_vector_and_scalar)
<< DestTy << SrcTy << R;
Kind = CK_VectorSplat;
return prepareVectorSplat(DestTy, CastExpr);
}
Contributor
Author
There was a problem hiding this comment.
Should we proceed with "_Float16" in the OpenCL test? @jmmartinez @arsenm
Contributor
There was a problem hiding this comment.
Should we proceed with "_Float16" in the OpenCL test?
Not yet. We have to fix the vector conversion first. (Sorry I didn't have much time this week to work on it)
Contributor
Author
There was a problem hiding this comment.
Added code to treat _Float16 same as half. With this change from _Float16 to half conversion working fine.
shiltian
reviewed
Nov 12, 2025
🐧 Linux x64 Test Results
|
ranapratap55
force-pushed
the
users/ranapratap55/extended-image-fp16-openCL-fix
branch
from
c3fdba0 to
cff4602
Compare
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.
For extended imges insts amdgcn_image_sample_/gather4 builtins, using 'x' in the builtin def so that it will take _Float16 for both HIP/C++ and OpenCL.