1;RUN: llc < %s -march=amdgcn -mcpu=verde -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=SICI 2;RUN: llc < %s -march=amdgcn -mcpu=tonga -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=VI 3 4;CHECK-LABEL: {{^}}test1: 5;CHECK-NOT: s_waitcnt 6;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen glc 7;CHECK: s_waitcnt vmcnt(0) 8;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen glc 9;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc 10;CHECK: s_waitcnt vmcnt(0) 11;CHECK: buffer_atomic_swap v0, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc 12;CHECK: s_waitcnt vmcnt(0) 13;CHECK: buffer_atomic_swap v0, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc 14;CHECK: s_waitcnt vmcnt(0) 15;CHECK: buffer_atomic_swap v0, v[1:2], s[0:3], 0 idxen offen offset:42 glc 16;CHECK-DAG: s_waitcnt vmcnt(0) 17;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], [[SOFS]] idxen offset:4 glc 18;CHECK: s_waitcnt vmcnt(0) 19;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen{{$}} 20;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen glc 21define amdgpu_ps float @test1(<4 x i32> inreg %rsrc, i32 %data, i32 %vindex, i32 %voffset) { 22main_body: 23 %o1 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %data, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0) 24 %o2 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o1, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 25 %o3 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o2, <4 x i32> %rsrc, i32 0, i32 %voffset, i32 0, i32 0) 26 %o4 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o3, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 0, i32 0) 27 %ofs.5 = add i32 %voffset, 42 28 %o5 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o4, <4 x i32> %rsrc, i32 0, i32 %ofs.5, i32 0, i32 0) 29 %o6 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o5, <4 x i32> %rsrc, i32 0, i32 4, i32 8188, i32 0) 30 %unused = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o6, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0) 31 %o7 = bitcast i32 %o6 to float 32 %out = call float @llvm.amdgcn.struct.buffer.atomic.swap.f32(float %o7, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0) 33 ret float %out 34} 35 36;CHECK-LABEL: {{^}}test2: 37;CHECK-NOT: s_waitcnt 38;CHECK: buffer_atomic_add v0, v1, s[0:3], 0 idxen glc{{$}} 39;CHECK: s_waitcnt vmcnt(0) 40;CHECK: buffer_atomic_sub v0, v1, s[0:3], 0 idxen glc slc 41;CHECK: s_waitcnt vmcnt(0) 42;CHECK: buffer_atomic_smin v0, v1, s[0:3], 0 idxen glc{{$}} 43;CHECK: s_waitcnt vmcnt(0) 44;CHECK: buffer_atomic_umin v0, v1, s[0:3], 0 idxen glc slc 45;CHECK: s_waitcnt vmcnt(0) 46;CHECK: buffer_atomic_smax v0, v1, s[0:3], 0 idxen glc{{$}} 47;CHECK: s_waitcnt vmcnt(0) 48;CHECK: buffer_atomic_umax v0, v1, s[0:3], 0 idxen glc slc 49;CHECK: s_waitcnt vmcnt(0) 50;CHECK: buffer_atomic_and v0, v1, s[0:3], 0 idxen glc{{$}} 51;CHECK: s_waitcnt vmcnt(0) 52;CHECK: buffer_atomic_or v0, v1, s[0:3], 0 idxen glc slc 53;CHECK: s_waitcnt vmcnt(0) 54;CHECK: buffer_atomic_xor v0, v1, s[0:3], 0 idxen glc 55;CHECK: s_waitcnt vmcnt(0) 56;CHECK: buffer_atomic_inc v0, v1, s[0:3], 0 idxen glc 57;CHECK: s_waitcnt vmcnt(0) 58;CHECK: buffer_atomic_dec v0, v1, s[0:3], 0 idxen glc 59define amdgpu_ps float @test2(<4 x i32> inreg %rsrc, i32 %data, i32 %vindex) { 60main_body: 61 %t1 = call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 62 %t2 = call i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32 %t1, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2) 63 %t3 = call i32 @llvm.amdgcn.struct.buffer.atomic.smin.i32(i32 %t2, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 64 %t4 = call i32 @llvm.amdgcn.struct.buffer.atomic.umin.i32(i32 %t3, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2) 65 %t5 = call i32 @llvm.amdgcn.struct.buffer.atomic.smax.i32(i32 %t4, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 66 %t6 = call i32 @llvm.amdgcn.struct.buffer.atomic.umax.i32(i32 %t5, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2) 67 %t7 = call i32 @llvm.amdgcn.struct.buffer.atomic.and.i32(i32 %t6, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 68 %t8 = call i32 @llvm.amdgcn.struct.buffer.atomic.or.i32(i32 %t7, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2) 69 %t9 = call i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32 %t8, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 70 %t10 = call i32 @llvm.amdgcn.struct.buffer.atomic.inc.i32(i32 %t9, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 71 %t11 = call i32 @llvm.amdgcn.struct.buffer.atomic.dec.i32(i32 %t10, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 72 %out = bitcast i32 %t11 to float 73 ret float %out 74} 75 76; Ideally, we would teach tablegen & friends that cmpswap only modifies the 77; first vgpr. Since we don't do that yet, the register allocator will have to 78; create copies which we don't bother to track here. 79; 80;CHECK-LABEL: {{^}}test3: 81;CHECK-NOT: s_waitcnt 82;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[0:3], 0 idxen glc 83;CHECK: s_waitcnt vmcnt(0) 84;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[0:3], 0 idxen glc 85;CHECK: s_waitcnt vmcnt(0) 86;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc 87;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc 88;CHECK: s_waitcnt vmcnt(0) 89;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc 90;CHECK: s_waitcnt vmcnt(0) 91;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen offset:44 glc 92;CHECK-DAG: s_waitcnt vmcnt(0) 93;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[0:3], [[SOFS]] idxen offset:4 glc 94define amdgpu_ps float @test3(<4 x i32> inreg %rsrc, i32 %data, i32 %cmp, i32 %vindex, i32 %voffset) { 95main_body: 96 %o1 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %data, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0) 97 %o2 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o1, i32 %cmp, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0) 98 %o3 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o2, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 %voffset, i32 0, i32 0) 99 %o4 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o3, i32 %cmp, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 0, i32 0) 100 %offs.5 = add i32 %voffset, 44 101 %o5 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o4, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 %offs.5, i32 0, i32 0) 102 %o6 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o5, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 4, i32 8188, i32 0) 103 104; Detecting the no-return variant doesn't work right now because of how the 105; intrinsic is replaced by an instruction that feeds into an EXTRACT_SUBREG. 106; Since there probably isn't a reasonable use-case of cmpswap that discards 107; the return value, that seems okay. 108; 109; %unused = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o6, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0) 110 %out = bitcast i32 %o6 to float 111 ret float %out 112} 113 114;CHECK-LABEL: {{^}}test4: 115;CHECK: buffer_atomic_add v0, 116define amdgpu_ps float @test4() { 117main_body: 118 %v = call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 1, <4 x i32> undef, i32 0, i32 4, i32 0, i32 0) 119 %v.float = bitcast i32 %v to float 120 ret float %v.float 121} 122 123declare i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 124declare float @llvm.amdgcn.struct.buffer.atomic.swap.f32(float, <4 x i32>, i32, i32, i32, i32) #0 125declare i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 126declare i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 127declare i32 @llvm.amdgcn.struct.buffer.atomic.smin.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 128declare i32 @llvm.amdgcn.struct.buffer.atomic.umin.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 129declare i32 @llvm.amdgcn.struct.buffer.atomic.smax.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 130declare i32 @llvm.amdgcn.struct.buffer.atomic.umax.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 131declare i32 @llvm.amdgcn.struct.buffer.atomic.and.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 132declare i32 @llvm.amdgcn.struct.buffer.atomic.or.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 133declare i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 134declare i32 @llvm.amdgcn.struct.buffer.atomic.inc.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 135declare i32 @llvm.amdgcn.struct.buffer.atomic.dec.i32(i32, <4 x i32>, i32, i32, i32, i32) #0 136declare i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32, i32, <4 x i32>, i32, i32, i32, i32) #0 137 138attributes #0 = { nounwind } 139