1 ; RUN: opt -mtriple amdgcn-- -passes='print<uniformity>' -disable-output %s 2>&1 | FileCheck %s
3 ; CHECK: DIVERGENT: %swizzle = call i32 @llvm.amdgcn.ds.swizzle(i32 %src, i32 100) #0
4 define amdgpu_kernel void @ds_swizzle(ptr addrspace(1) %out, i32 %src) #0 {
5 %swizzle = call i32 @llvm.amdgcn.ds.swizzle(i32 %src, i32 100) #0
6 store i32 %swizzle, ptr addrspace(1) %out, align 4
10 ; CHECK: DIVERGENT: %v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
11 define amdgpu_kernel void @v_permlane16_b32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) #0 {
12 %v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
13 store i32 %v, ptr addrspace(1) %out
17 ; CHECK: DIVERGENT: %v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
18 define amdgpu_kernel void @v_permlanex16_b32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) #0 {
19 %v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
20 store i32 %v, ptr addrspace(1) %out
24 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 false) #0
25 define amdgpu_kernel void @update_dpp(ptr addrspace(1) %out, i32 %in1, i32 %in2) #0 {
26 %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 false) #0
27 store i32 %tmp0, ptr addrspace(1) %out
31 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %in, i32 1, i32 1, i32 1, i1 true) #0
32 define amdgpu_kernel void @mov_dpp(ptr addrspace(1) %out, i32 %in) #0 {
33 %tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %in, i32 1, i32 1, i32 1, i1 true) #0
34 store i32 %tmp0, ptr addrspace(1) %out
38 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.mov.dpp8.i32(i32 %in, i32 1) #0
39 define amdgpu_kernel void @mov_dpp8(ptr addrspace(1) %out, i32 %in) #0 {
40 %tmp0 = call i32 @llvm.amdgcn.mov.dpp8.i32(i32 %in, i32 1) #0
41 store i32 %tmp0, ptr addrspace(1) %out
45 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.writelane(i32 0, i32 1, i32 2)
46 define amdgpu_kernel void @writelane(ptr addrspace(1) %out) #0 {
47 %tmp0 = call i32 @llvm.amdgcn.writelane(i32 0, i32 1, i32 2)
48 store i32 %tmp0, ptr addrspace(1) %out
52 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half> %A, <16 x half> %B, <8 x float> %C)
53 define amdgpu_kernel void @wmma_f32_16x16x16_f16(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
54 %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half> %A, <16 x half> %B, <8 x float> %C)
55 store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
59 ; CHECK: DIVERGENT: %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16> %A, <16 x i16> %B, <8 x float> %C)
60 define amdgpu_kernel void @wmma_f32_16x16x16_ibf16(<16 x i16> %A, <16 x i16> %B, <8 x float> %C, ptr addrspace(1) %out) {
61 %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16> %A, <16 x i16> %B, <8 x float> %C)
62 store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
66 ; CHECK: DIVERGENT: %tmp0 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16(<16 x half> %A, <16 x half> %B, <16 x half> %C, i1 false)
67 define amdgpu_kernel void @wmma_f16_16x16x16_f16(<16 x half> %A, <16 x half> %B, <16 x half> %C, ptr addrspace(1) %out) {
69 %tmp0 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16(<16 x half> %A, <16 x half> %B, <16 x half> %C, i1 false)
70 store <16 x half> %tmp0, ptr addrspace(1) %out, align 32
74 ; CHECK: DIVERGENT: %tmp0 = call <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16(<16 x i16> %A, <16 x i16> %B, <16 x i16> %C, i1 false)
75 define amdgpu_kernel void @wmma_f16_16x16x16_bf16(<16 x i16> %A, <16 x i16> %B, <16 x i16> %C, ptr addrspace(1) %out) {
77 %tmp0 = call <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16(<16 x i16> %A, <16 x i16> %B, <16 x i16> %C, i1 false)
78 store <16 x i16> %tmp0, ptr addrspace(1) %out, align 32
82 ; CHECK: DIVERGENT: %tmp0 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32(i1 false, <4 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i1 false)
83 define amdgpu_kernel void @wmma_i32_16x16x16_ui8(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
85 %tmp0 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32(i1 false, <4 x i32> %A, i1 false, <4 x i32> %B, <8 x i32> %C, i1 false)
86 store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
90 ; CHECK: DIVERGENT: %tmp0 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32(i1 false, <2 x i32> %A, i1 false, <2 x i32> %B, <8 x i32> %C, i1 false)
91 define amdgpu_kernel void @wmma_i32_16x16x16_ui4(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
93 %tmp0 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32(i1 false, <2 x i32> %A, i1 false, <2 x i32> %B, <8 x i32> %C, i1 false)
94 store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
98 declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
99 declare i32 @llvm.amdgcn.permlane16(i32, i32, i32, i32, i1, i1) #1
100 declare i32 @llvm.amdgcn.permlanex16(i32, i32, i32, i32, i1, i1) #1
101 declare i32 @llvm.amdgcn.mov.dpp.i32(i32, i32, i32, i32, i1) #1
102 declare i32 @llvm.amdgcn.mov.dpp8.i32(i32, i32) #1
103 declare i32 @llvm.amdgcn.update.dpp.i32(i32, i32, i32, i32, i32, i1) #1
104 declare i32 @llvm.amdgcn.writelane(i32, i32, i32) #1
105 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half>, <16 x half> , <8 x float>) #1
106 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16>, <16 x i16> , <8 x float>) #1
107 declare <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16(<16 x half>, <16 x half> , <16 x half>, i1 immarg) #1
108 declare <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16(<16 x i16>, <16 x i16> , <16 x i16>, i1 immarg) #1
109 declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8.v8i32(i1 immarg, <4 x i32>, i1 immarg, <4 x i32> , <8 x i32>, i1 immarg) #1
110 declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4.v8i32(i1 immarg, <2 x i32>, i1 immarg, <2 x i32> , <8 x i32>, i1 immarg) #1
112 attributes #0 = { nounwind convergent }
113 attributes #1 = { nounwind readnone convergent }