[AMDGPU] Lower __builtin_amdgcn_read_exec_hi to use amdgcn_ballot #69567
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
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-clang-codegen Author: Rana Pratap Reddy (ranapratap55) ChangesCurrently __builtin_read_exec_hi lowers to llvm.read_register, this patch lowers it to use amdgcn_ballot. Full diff: https://github.com/llvm/llvm-project/pull/69567.diff 2 Files Affected:
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index db9f354fa8386d3..d60826f293f0c46 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -7997,14 +7997,26 @@ enum SpecialRegisterAccessKind {
static Value *EmitAMDGCNBallotForExec(CodeGenFunction &CGF, const CallExpr *E,
llvm::Type *RegisterType,
- llvm::Type *ValueType) {
+ llvm::Type *ValueType, bool isExecHi) {
CodeGen::CGBuilderTy &Builder = CGF.Builder;
CodeGen::CodeGenModule &CGM = CGF.CGM;
llvm::Type *ResultType = CGF.ConvertType(E->getType());
- Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_ballot, {ResultType});
- llvm::Value *Call = Builder.CreateCall(F, {Builder.getInt1(true)});
- return Call;
+ llvm::Value *Call;
+ Function *F;
+
+ if (isExecHi) {
+ F = CGM.getIntrinsic(Intrinsic::amdgcn_ballot, {RegisterType});
+ Call = Builder.CreateCall(F, {Builder.getInt1(true)});
+ Value *C1 = llvm::ConstantInt::get(ValueType, 32);
+ Value *Rt2 = Builder.CreateLShr(Call, C1);
+ Rt2 = Builder.CreateTruncOrBitCast(Rt2, CGF.Int32Ty);
+ return Rt2;
+ } else {
+ F = CGM.getIntrinsic(Intrinsic::amdgcn_ballot, {ResultType});
+ Call = Builder.CreateCall(F, {Builder.getInt1(true)});
+ return Call;
+ }
}
// Generates the IR for the read/write special register builtin,
@@ -17837,10 +17849,11 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
return Builder.CreateCall(F, {Addr, Val, ZeroI32, ZeroI32, ZeroI1});
}
case AMDGPU::BI__builtin_amdgcn_read_exec:
+ return EmitAMDGCNBallotForExec(*this, E, Int64Ty, Int64Ty, false);
case AMDGPU::BI__builtin_amdgcn_read_exec_lo:
- case AMDGPU::BI__builtin_amdgcn_read_exec_hi: {
- return EmitAMDGCNBallotForExec(*this, E, Int64Ty, Int64Ty);
- }
+ return EmitAMDGCNBallotForExec(*this, E, Int32Ty, Int32Ty, false);
+ case AMDGPU::BI__builtin_amdgcn_read_exec_hi:
+ return EmitAMDGCNBallotForExec(*this, E, Int64Ty, Int64Ty, true);
case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray:
case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_h:
case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_l:
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn.cl
index 8938642e3b19f8c..0bc9a54682d3e31 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn.cl
@@ -526,7 +526,9 @@ void test_read_exec_lo(global uint* out) {
// CHECK: declare i32 @llvm.amdgcn.ballot.i32(i1) #[[$NOUNWIND_READONLY:[0-9]+]]
// CHECK-LABEL: @test_read_exec_hi(
-// CHECK: call i32 @llvm.amdgcn.ballot.i32(i1 true)
+// CHECK: call i64 @llvm.amdgcn.ballot.i64(i1 true)
+// CHECK: lshr i64 [[A:%.*]], 32
+// CHECK: trunc i64 [[B:%.*]] to i32
void test_read_exec_hi(global uint* out) {
*out = __builtin_amdgcn_read_exec_hi();
}
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
ranapratap55
force-pushed
the
rpr55/read_reg_to_ballot
branch
from
340e633 to
1958244
Compare
ranapratap55
changed the title
arsenm
requested changes
Oct 24, 2023
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| Call = Builder.CreateCall(F, {Builder.getInt1(true)}); | ||
| Value *C1 = llvm::ConstantInt::get(ValueType, 32); | ||
| Value *Rt2 = Builder.CreateLShr(Call, C1); | ||
| Rt2 = Builder.CreateTruncOrBitCast(Rt2, CGF.Int32Ty); |
There was a problem hiding this comment.
Should just be CreateTrunc, the result should always be i32
There was a problem hiding this comment.
Updated in the latest patch.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
Comment on lines
8016
to
8017
| F = CGM.getIntrinsic(Intrinsic::amdgcn_ballot, {ResultType}); | ||
| Call = Builder.CreateCall(F, {Builder.getInt1(true)}); |
| @@ -526,7 +526,9 @@ void test_read_exec_lo(global uint* out) { | |||
| // CHECK: declare i32 @llvm.amdgcn.ballot.i32(i1) #[[$NOUNWIND_READONLY:[0-9]+]] | |||
|
|
|||
| // CHECK-LABEL: @test_read_exec_hi( | |||
| // CHECK: call i32 @llvm.amdgcn.ballot.i32(i1 true) | |||
| // CHECK: call i64 @llvm.amdgcn.ballot.i64(i1 true) | |||
There was a problem hiding this comment.
needs wave32 and wave64 tests. These ballot tests belong in builtins-amdgcn-wave32.cl and builtins-amdgcn-wave64.cl
There was a problem hiding this comment.
moved to respected wave32.cl and wave64.cl tests.
ranapratap55
force-pushed
the
rpr55/read_reg_to_ballot
branch
3 times, most recently
from
c2ed656 to
65c806a
Compare
arsenm
reviewed
Oct 25, 2023
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
|
|
||
| if (isExecHi) { | ||
| Value *C1 = llvm::ConstantInt::get(ValueType, 32); | ||
| Value *Rt2 = Builder.CreateLShr(Call, C1); |
There was a problem hiding this comment.
There's a CreateLShr overload that takes the constant integer, you don't need the ConstantInt::get
There was a problem hiding this comment.
Removed ConstantInt::get.
| } | ||
|
|
||
| // CHECK-LABEL: @test_read_exec_hi( | ||
| // CHECK: call i64 @llvm.amdgcn.ballot.i64(i1 true) |
There was a problem hiding this comment.
Does this codegen on wave32? Or should this directly emit 0?
There was a problem hiding this comment.
Yes, this is generated codegen on wave32.
ranapratap55
force-pushed
the
rpr55/read_reg_to_ballot
branch
from
65c806a to
8a5dfe6
Compare
arsenm
approved these changes
Oct 25, 2023
ranapratap55
force-pushed
the
rpr55/read_reg_to_ballot
branch
from
8a5dfe6 to
ebc8ebd
Compare
ranapratap55
force-pushed
the
rpr55/read_reg_to_ballot
branch
from
ebc8ebd to
bd6e144
Compare
zahiraam
pushed a commit
to zahiraam/llvm-project
that referenced
this pull request
Oct 26, 2023
…vm#69567) Currently __builtin_amdgcn_read_exec_hi lowers to llvm.read_register, this patch lowers it to use amdgcn_ballot.
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.
Currently __builtin_amdgcn_read_exec_hi lowers to llvm.read_register, this patch lowers it to use amdgcn_ballot.