[llvm-branch-commits] [clang] clang/AMDGPU: Emit atomicrmw from {global|flat}_atomic_fadd_v2f16 builtins (PR #96873)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Jun 27 02:11:12 PDT 2024
https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/96873
None
>From 65a690d80cf39df132cacff510371c9dcb1b97fd Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 26 Jun 2024 19:12:59 +0200
Subject: [PATCH] clang/AMDGPU: Emit atomicrmw from
{global|flat}_atomic_fadd_v2f16 builtins
---
clang/lib/CodeGen/CGBuiltin.cpp | 20 ++++++-------------
.../builtins-fp-atomics-gfx12.cl | 9 ++++++---
.../builtins-fp-atomics-gfx90a.cl | 2 +-
.../builtins-fp-atomics-gfx940.cl | 3 ++-
4 files changed, 15 insertions(+), 19 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index d90762748d925..2a1861e4413fd 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18655,22 +18655,15 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
Src0 = Builder.CreatePointerBitCastOrAddrSpaceCast(Src0, PTy);
return Builder.CreateCall(F, { Src0, Src1, Src2, Src3, Src4 });
}
- case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
- case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32:
- case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16: {
+ case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32: {
Intrinsic::ID IID;
llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
switch (BuiltinID) {
- case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
- ArgTy = llvm::FixedVectorType::get(
- llvm::Type::getHalfTy(getLLVMContext()), 2);
- IID = Intrinsic::amdgcn_global_atomic_fadd;
- break;
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
IID = Intrinsic::amdgcn_global_atomic_fmin;
break;
@@ -18690,11 +18683,6 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
ArgTy = llvm::Type::getFloatTy(getLLVMContext());
IID = Intrinsic::amdgcn_flat_atomic_fadd;
break;
- case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16:
- ArgTy = llvm::FixedVectorType::get(
- llvm::Type::getHalfTy(getLLVMContext()), 2);
- IID = Intrinsic::amdgcn_flat_atomic_fadd;
- break;
}
llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
llvm::Value *Val = EmitScalarExpr(E->getArg(1));
@@ -19085,7 +19073,9 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_v2f16:
case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_v2bf16:
case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
- case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64: {
+ case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
+ case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
+ case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16: {
llvm::AtomicRMWInst::BinOp BinOp;
switch (BuiltinID) {
case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
@@ -19103,6 +19093,8 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
case AMDGPU::BI__builtin_amdgcn_ds_atomic_fadd_v2bf16:
case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f32:
case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_f64:
+ case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2f16:
+ case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2f16:
BinOp = llvm::AtomicRMWInst::FAdd;
break;
}
diff --git a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx12.cl b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx12.cl
index 21c1c38bc78dc..cf304d7b0818a 100644
--- a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx12.cl
+++ b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx12.cl
@@ -48,7 +48,8 @@ void test_local_add_2f16_noret(__local half2 *addr, half2 x) {
}
// CHECK-LABEL: test_flat_add_2f16
-// CHECK: call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p0.v2f16(ptr %{{.*}}, <2 x half> %{{.*}})
+// CHECK: [[RMW:%.+]] = atomicrmw fadd ptr %{{.+}}, <2 x half> %{{.+}} syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+
// GFX12-LABEL: test_flat_add_2f16
// GFX12: flat_atomic_pk_add_f16
half2 test_flat_add_2f16(__generic half2 *addr, half2 x) {
@@ -64,7 +65,8 @@ short2 test_flat_add_2bf16(__generic short2 *addr, short2 x) {
}
// CHECK-LABEL: test_global_add_half2
-// CHECK: call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1.v2f16(ptr addrspace(1) %{{.*}}, <2 x half> %{{.*}})
+// CHECK: [[RMW:%.+]] = atomicrmw fadd ptr addrspace(1) %{{.+}}, <2 x half> %{{.+}} syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+
// GFX12-LABEL: test_global_add_half2
// GFX12: global_atomic_pk_add_f16 v2, v[0:1], v2, off th:TH_ATOMIC_RETURN
void test_global_add_half2(__global half2 *addr, half2 x) {
@@ -73,7 +75,8 @@ void test_global_add_half2(__global half2 *addr, half2 x) {
}
// CHECK-LABEL: test_global_add_half2_noret
-// CHECK: call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1.v2f16(ptr addrspace(1) %{{.*}}, <2 x half> %{{.*}})
+// CHECK: [[RMW:%.+]] = atomicrmw fadd ptr addrspace(1) %{{.+}}, <2 x half> %{{.+}} syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+
// GFX12-LABEL: test_global_add_half2_noret
// GFX12: global_atomic_pk_add_f16 v[0:1], v2, off
void test_global_add_half2_noret(__global half2 *addr, half2 x) {
diff --git a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl
index 3778f65feaad4..4980c44215743 100644
--- a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl
+++ b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl
@@ -18,7 +18,7 @@ void test_global_add_f64(__global double *addr, double x) {
}
// CHECK-LABEL: test_global_add_half2
-// CHECK: call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1.v2f16(ptr addrspace(1) %{{.*}}, <2 x half> %{{.*}})
+// CHECK: = atomicrmw fadd ptr addrspace(1) %{{.+}}, <2 x half> %{{.+}} syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
// GFX90A-LABEL: test_global_add_half2
// GFX90A: global_atomic_pk_add_f16 v2, v[0:1], v2, off glc
void test_global_add_half2(__global half2 *addr, half2 x) {
diff --git a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx940.cl b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx940.cl
index d4ca32b9b3cd6..a4f438bea33a6 100644
--- a/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx940.cl
+++ b/clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx940.cl
@@ -18,7 +18,8 @@ half2 test_flat_add_f32(__generic float *addr, float x) {
}
// CHECK-LABEL: test_flat_add_2f16
-// CHECK: call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p0.v2f16(ptr %{{.*}}, <2 x half> %{{.*}})
+// CHECK: [[RMW:%.+]] = atomicrmw fadd ptr %{{.+}}, <2 x half> %{{.+}} syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+
// GFX940-LABEL: test_flat_add_2f16
// GFX940: flat_atomic_pk_add_f16
half2 test_flat_add_2f16(__generic half2 *addr, half2 x) {
More information about the llvm-branch-commits
mailing list