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: %v = call i32 @llvm.amdgcn.permlane16.var(i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
25 define amdgpu_kernel void @v_permlane16_var_b32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) #0 {
26 %v = call i32 @llvm.amdgcn.permlane16.var(i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
27 store i32 %v, ptr addrspace(1) %out
31 ; CHECK: DIVERGENT: %v = call i32 @llvm.amdgcn.permlanex16.var(i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
32 define amdgpu_kernel void @v_permlanex16_var_b32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) #0 {
33 %v = call i32 @llvm.amdgcn.permlanex16.var(i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false) #0
34 store i32 %v, ptr addrspace(1) %out
38 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 false) #0
39 define amdgpu_kernel void @update_dpp(ptr addrspace(1) %out, i32 %in1, i32 %in2) #0 {
40 %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 false) #0
41 store i32 %tmp0, ptr addrspace(1) %out
45 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %in, i32 1, i32 1, i32 1, i1 true) #0
46 define amdgpu_kernel void @mov_dpp(ptr addrspace(1) %out, i32 %in) #0 {
47 %tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %in, i32 1, i32 1, i32 1, i1 true) #0
48 store i32 %tmp0, ptr addrspace(1) %out
52 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.mov.dpp8.i32(i32 %in, i32 1) #0
53 define amdgpu_kernel void @mov_dpp8(ptr addrspace(1) %out, i32 %in) #0 {
54 %tmp0 = call i32 @llvm.amdgcn.mov.dpp8.i32(i32 %in, i32 1) #0
55 store i32 %tmp0, ptr addrspace(1) %out
59 ; CHECK: DIVERGENT: %tmp0 = call i32 @llvm.amdgcn.writelane(i32 0, i32 1, i32 2)
60 define amdgpu_kernel void @writelane(ptr addrspace(1) %out) #0 {
61 %tmp0 = call i32 @llvm.amdgcn.writelane(i32 0, i32 1, i32 2)
62 store i32 %tmp0, ptr addrspace(1) %out
66 ; 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)
67 define amdgpu_kernel void @wmma_f32_16x16x16_f16(<16 x half> %A, <16 x half> %B, <8 x float> %C, ptr addrspace(1) %out) {
68 %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half> %A, <16 x half> %B, <8 x float> %C)
69 store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
73 ; 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)
74 define amdgpu_kernel void @wmma_f32_16x16x16_ibf16(<16 x i16> %A, <16 x i16> %B, <8 x float> %C, ptr addrspace(1) %out) {
75 %tmp0 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16> %A, <16 x i16> %B, <8 x float> %C)
76 store <8 x float> %tmp0, ptr addrspace(1) %out, align 32
80 ; 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)
81 define amdgpu_kernel void @wmma_f16_16x16x16_f16(<16 x half> %A, <16 x half> %B, <16 x half> %C, ptr addrspace(1) %out) {
83 %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)
84 store <16 x half> %tmp0, ptr addrspace(1) %out, align 32
88 ; 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)
89 define amdgpu_kernel void @wmma_f16_16x16x16_bf16(<16 x i16> %A, <16 x i16> %B, <16 x i16> %C, ptr addrspace(1) %out) {
91 %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)
92 store <16 x i16> %tmp0, ptr addrspace(1) %out, align 32
96 ; 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)
97 define amdgpu_kernel void @wmma_i32_16x16x16_ui8(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
99 %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)
100 store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
104 ; 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)
105 define amdgpu_kernel void @wmma_i32_16x16x16_ui4(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, ptr addrspace(1) %out) {
107 %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)
108 store <8 x i32> %tmp0, ptr addrspace(1) %out, align 32
112 declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
113 declare i32 @llvm.amdgcn.permlane16(i32, i32, i32, i32, i1, i1) #1
114 declare i32 @llvm.amdgcn.permlanex16(i32, i32, i32, i32, i1, i1) #1
115 declare i32 @llvm.amdgcn.permlane16.var(i32, i32, i32, i1, i1) #1
116 declare i32 @llvm.amdgcn.permlanex16.var(i32, i32, i32, i1, i1) #1
117 declare i32 @llvm.amdgcn.mov.dpp.i32(i32, i32, i32, i32, i1) #1
118 declare i32 @llvm.amdgcn.mov.dpp8.i32(i32, i32) #1
119 declare i32 @llvm.amdgcn.update.dpp.i32(i32, i32, i32, i32, i32, i1) #1
120 declare i32 @llvm.amdgcn.writelane(i32, i32, i32) #1
121 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16.v8f32(<16 x half>, <16 x half> , <8 x float>) #1
122 declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16.v8f32(<16 x i16>, <16 x i16> , <8 x float>) #1
123 declare <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16.v16f16(<16 x half>, <16 x half> , <16 x half>, i1 immarg) #1
124 declare <16 x i16> @llvm.amdgcn.wmma.bf16.16x16x16.bf16.v16i16(<16 x i16>, <16 x i16> , <16 x i16>, i1 immarg) #1
125 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
126 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
128 attributes #0 = { nounwind convergent }
129 attributes #1 = { nounwind readnone convergent }