Skip to content

Commit 2c9916f

Browse files
arsenmarsenm
committed
clang/AMDGPU: Emit atomicrmw for flat/global atomic min/max f64 builtins
1 parent 75cbc81 commit 2c9916f

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
@@ -18920,32 +18920,6 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1892018920
Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() });
1892118921
return Builder.CreateCall(F, { Src0, Builder.getFalse() });
1892218922
}
18923-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
18924-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
18925-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
18926-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
18927-
Intrinsic::ID IID;
18928-
llvm::Type *ArgTy = llvm::Type::getDoubleTy(getLLVMContext());
18929-
switch (BuiltinID) {
18930-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
18931-
IID = Intrinsic::amdgcn_global_atomic_fmin;
18932-
break;
18933-
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
18934-
IID = Intrinsic::amdgcn_global_atomic_fmax;
18935-
break;
18936-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
18937-
IID = Intrinsic::amdgcn_flat_atomic_fmin;
18938-
break;
18939-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
18940-
IID = Intrinsic::amdgcn_flat_atomic_fmax;
18941-
break;
18942-
}
18943-
llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
18944-
llvm::Value *Val = EmitScalarExpr(E->getArg(1));
18945-
llvm::Function *F =
18946-
CGM.getIntrinsic(IID, {ArgTy, Addr->getType(), Val->getType()});
18947-
return Builder.CreateCall(F, {Addr, Val});
18948-
}
1894918923
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_i32:
1895018924
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_v2i32:
1895118925
case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v4i16:
@@ -19329,7 +19303,11 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1932919303
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f32:
1933019304
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_f64:
1933119305
case AMDGPU::BI__builtin_amdgcn_global_atomic_fadd_v2bf16:
19332-
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16: {
19306+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fadd_v2bf16:
19307+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
19308+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
19309+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
19310+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64: {
1933319311
llvm::AtomicRMWInst::BinOp BinOp;
1933419312
switch (BuiltinID) {
1933519313
case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
@@ -19356,8 +19334,12 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
1935619334
BinOp = llvm::AtomicRMWInst::FAdd;
1935719335
break;
1935819336
case AMDGPU::BI__builtin_amdgcn_ds_fminf:
19337+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmin_f64:
19338+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmin_f64:
1935919339
BinOp = llvm::AtomicRMWInst::FMin;
1936019340
break;
19341+
case AMDGPU::BI__builtin_amdgcn_global_atomic_fmax_f64:
19342+
case AMDGPU::BI__builtin_amdgcn_flat_atomic_fmax_f64:
1936119343
case AMDGPU::BI__builtin_amdgcn_ds_fmaxf:
1936219344
BinOp = llvm::AtomicRMWInst::FMax;
1936319345
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)