Skip to content

Commit c15295e

Browse files
arsenmarsenm
committed
clang/AMDGPU: Emit atomicrmw for flat/global atomic min/max f64 builtins
1 parent 8b78624 commit c15295e

File tree

2 files changed

+21
-33
lines changed

2 files changed

+21
-33
lines changed

‎clang/lib/CodeGen/CGBuiltin.cpp‎

Lines changed: 9 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -18744,32 +18744,6 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1874418744
Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() });
1874518745
return Builder.CreateCall(F, { Src0, Builder.getFalse() });
1874618746
}
18747-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
18748-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
18749-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
18750-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
18751-
Intrinsic::ID IID;
18752-
llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
18753-
switch (BuiltinID) {
18754-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
18755-
IID = Intrinsic::amdgcn_global_atomic_fmin;
18756-
break;
18757-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
18758-
IID = Intrinsic::amdgcn_global_atomic_fmax;
18759-
break;
18760-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
18761-
IID = Intrinsic::amdgcn_flat_atomic_fmin;
18762-
break;
18763-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
18764-
IID = Intrinsic::amdgcn_flat_atomic_fmax;
18765-
break;
18766-
}
18767-
llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
18768-
llvm::Value *Val = EmitScalarExpr(E->getArg(1));
18769-
llvm::Function *F =
18770-
CGM.getIntrinsic(IID, {ArgTy, Addr->getType(), Val->getType()});
18771-
return Builder.CreateCall(F, {Addr, Val});
18772-
}
1877318747
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_i32:
1877418748
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_v2i32:
1877518749
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v4i16:
@@ -19153,7 +19127,11 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1915319127
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32:
1915419128
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
1915519129
case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2bf16:
19156-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16: {
19130+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16:
19131+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
19132+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
19133+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
19134+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
1915719135
llvm::AtomicRMWInst::BinOp BinOp;
1915819136
switch (BuiltinID) {
1915919137
case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
@@ -19180,8 +19158,12 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1918019158
BinOp = llvm::AtomicRMWInst::FAdd;
1918119159
break;
1918219160
case AMDGPU::BI__builtin_amdgcn_ds_fminf:
19161+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
19162+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
1918319163
BinOp = llvm::AtomicRMWInst::FMin;
1918419164
break;
19165+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
19166+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
1918519167
case AMDGPU::BI__builtin_amdgcn_ds_fmaxf:
1918619168
BinOp = llvm::AtomicRMWInst::FMax;
1918719169
break;

‎clang/test/CodeGenOpenCL/builtins-fp-atomics-gfx90a.cl‎

Lines changed: 12 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -27,7 +27,8 @@ void test_global_add_half2(__global half2 *addr, half2 x) {
2727
}
2828

2929
// CHECK-LABEL: test_global_global_min_f64
30-
// CHECK: call double @llvm.amdgcn.global.atomic.fmin.f64.p1.f64(ptr addrspace(1) %{{.*}}, double %{{.*}})
30+
// CHECK: = atomicrmw fmin ptr addrspace(1) {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
31+
3132
// GFX90A-LABEL: test_global_global_min_f64$local
3233
// GFX90A: global_atomic_min_f64
3334
void test_global_global_min_f64(__global double *addr, double x){
@@ -36,7 +37,8 @@ void test_global_global_min_f64(__global double *addr, double x){
3637
}
3738

3839
// CHECK-LABEL: test_global_max_f64
39-
// CHECK: call double @llvm.amdgcn.global.atomic.fmax.f64.p1.f64(ptr addrspace(1) %{{.*}}, double %{{.*}})
40+
// CHECK: = atomicrmw fmax ptr addrspace(1) {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
41+
4042
// GFX90A-LABEL: test_global_max_f64$local
4143
// GFX90A: global_atomic_max_f64
4244
void test_global_max_f64(__global double *addr, double x){
@@ -65,7 +67,8 @@ void test_flat_global_add_f64(__global double *addr, double x){
6567
}
6668

6769
// CHECK-LABEL: test_flat_min_flat_f64
68-
// CHECK: call double @llvm.amdgcn.flat.atomic.fmin.f64.p0.f64(ptr %{{.*}}, double %{{.*}})
70+
// CHECK: = atomicrmw fmin ptr {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
71+
6972
// GFX90A-LABEL: test_flat_min_flat_f64$local
7073
// GFX90A: flat_atomic_min_f64
7174
void test_flat_min_flat_f64(__generic double *addr, double x){
@@ -74,7 +77,8 @@ void test_flat_min_flat_f64(__generic double *addr, double x){
7477
}
7578

7679
// CHECK-LABEL: test_flat_global_min_f64
77-
// CHECK: call double @llvm.amdgcn.flat.atomic.fmin.f64.p1.f64(ptr addrspace(1) %{{.*}}, double %{{.*}})
80+
// CHECK: = atomicrmw fmin ptr addrspace(1) {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
81+
7882
// GFX90A: test_flat_global_min_f64$local
7983
// GFX90A: global_atomic_min_f64
8084
void test_flat_global_min_f64(__global double *addr, double x){
@@ -83,7 +87,8 @@ void test_flat_global_min_f64(__global double *addr, double x){
8387
}
8488

8589
// CHECK-LABEL: test_flat_max_flat_f64
86-
// CHECK: call double @llvm.amdgcn.flat.atomic.fmax.f64.p0.f64(ptr %{{.*}}, double %{{.*}})
90+
// CHECK: = atomicrmw fmax ptr {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
91+
8792
// GFX90A-LABEL: test_flat_max_flat_f64$local
8893
// GFX90A: flat_atomic_max_f64
8994
void test_flat_max_flat_f64(__generic double *addr, double x){
@@ -92,7 +97,8 @@ void test_flat_max_flat_f64(__generic double *addr, double x){
9297
}
9398

9499
// CHECK-LABEL: test_flat_global_max_f64
95-
// CHECK: call double @llvm.amdgcn.flat.atomic.fmax.f64.p1.f64(ptr addrspace(1) %{{.*}}, double %{{.*}})
100+
// CHECK: = atomicrmw fmax ptr addrspace(1) {{.+}}, double %{{.+}} syncscope("agent") monotonic, align 8, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
101+
96102
// GFX90A-LABEL: test_flat_global_max_f64$local
97103
// GFX90A: global_atomic_max_f64
98104
void test_flat_global_max_f64(__global double *addr, double x){

0 commit comments

Comments
 (0)