131 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11-SDAG %s3 4define amdgpu_kernel void @test_p0(ptr addrspace(1) %out, ptr %src0) {5; GFX11-SDAG-LABEL: test_p0:6; GFX11-SDAG: ; %bb.0:7; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x248; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)9; GFX11-SDAG-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s310; GFX11-SDAG-NEXT: v_mov_b32_e32 v0, s211; GFX11-SDAG-NEXT: global_store_b64 v2, v[0:1], s[0:1]12; GFX11-SDAG-NEXT: s_endpgm13 %v = call ptr @llvm.amdgcn.permlane64.p0(ptr %src0)14 store ptr %v, ptr addrspace(1) %out15 ret void16}17 18define amdgpu_kernel void @test_v3p0(ptr addrspace(1) %out, <3 x ptr> %src0) {19; GFX11-SDAG-LABEL: test_v3p0:20; GFX11-SDAG: ; %bb.0:21; GFX11-SDAG-NEXT: s_clause 0x222; GFX11-SDAG-NEXT: s_load_b64 s[6:7], s[4:5], 0x5423; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x4424; GFX11-SDAG-NEXT: s_load_b64 s[4:5], s[4:5], 0x2425; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)26; GFX11-SDAG-NEXT: v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v5, s727; GFX11-SDAG-NEXT: v_dual_mov_b32 v4, s6 :: v_dual_mov_b32 v1, s128; GFX11-SDAG-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s329; GFX11-SDAG-NEXT: v_mov_b32_e32 v2, s230; GFX11-SDAG-NEXT: s_clause 0x131; GFX11-SDAG-NEXT: global_store_b64 v6, v[4:5], s[4:5] offset:1632; GFX11-SDAG-NEXT: global_store_b128 v6, v[0:3], s[4:5]33; GFX11-SDAG-NEXT: s_endpgm34 %v = call <3 x ptr> @llvm.amdgcn.permlane64.v3p0(<3 x ptr> %src0)35 store <3 x ptr> %v, ptr addrspace(1) %out36 ret void37}38 39define amdgpu_kernel void @test_p3(ptr addrspace(1) %out, ptr addrspace(3) %src0) {40; GFX11-SDAG-LABEL: test_p3:41; GFX11-SDAG: ; %bb.0:42; GFX11-SDAG-NEXT: s_clause 0x143; GFX11-SDAG-NEXT: s_load_b32 s2, s[4:5], 0x2c44; GFX11-SDAG-NEXT: s_load_b64 s[0:1], s[4:5], 0x2445; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)46; GFX11-SDAG-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s247; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]48; GFX11-SDAG-NEXT: s_endpgm49 %v = call ptr addrspace(3) @llvm.amdgcn.permlane64.v3p0(ptr addrspace(3) %src0)50 store ptr addrspace(3) %v, ptr addrspace(1) %out51 ret void52}53 54define amdgpu_kernel void @test_v3p3(ptr addrspace(1) %out, <3 x ptr addrspace(3)> %src0) {55; GFX11-SDAG-LABEL: test_v3p3:56; GFX11-SDAG: ; %bb.0:57; GFX11-SDAG-NEXT: s_clause 0x158; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x3459; GFX11-SDAG-NEXT: s_load_b64 s[4:5], s[4:5], 0x2460; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)61; GFX11-SDAG-NEXT: v_dual_mov_b32 v3, 0 :: v_dual_mov_b32 v0, s062; GFX11-SDAG-NEXT: v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s263; GFX11-SDAG-NEXT: global_store_b96 v3, v[0:2], s[4:5]64; GFX11-SDAG-NEXT: s_endpgm65 %v = call <3 x ptr addrspace(3)> @llvm.amdgcn.permlane64.v3p3(<3 x ptr addrspace(3)> %src0)66 store <3 x ptr addrspace(3)> %v, ptr addrspace(1) %out67 ret void68}69 70define amdgpu_kernel void @test_p5(ptr addrspace(1) %out, ptr addrspace(5) %src0) {71; GFX11-SDAG-LABEL: test_p5:72; GFX11-SDAG: ; %bb.0:73; GFX11-SDAG-NEXT: s_clause 0x174; GFX11-SDAG-NEXT: s_load_b32 s2, s[4:5], 0x2c75; GFX11-SDAG-NEXT: s_load_b64 s[0:1], s[4:5], 0x2476; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)77; GFX11-SDAG-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s278; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]79; GFX11-SDAG-NEXT: s_endpgm80 %v = call ptr addrspace(5) @llvm.amdgcn.permlane64.p5(ptr addrspace(5) %src0)81 store ptr addrspace(5) %v, ptr addrspace(1) %out82 ret void83}84 85define amdgpu_kernel void @test_v3p5(ptr addrspace(1) %out, <3 x ptr addrspace(5)> %src0) {86; GFX11-SDAG-LABEL: test_v3p5:87; GFX11-SDAG: ; %bb.0:88; GFX11-SDAG-NEXT: s_clause 0x189; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x3490; GFX11-SDAG-NEXT: s_load_b64 s[4:5], s[4:5], 0x2491; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)92; GFX11-SDAG-NEXT: v_dual_mov_b32 v3, 0 :: v_dual_mov_b32 v0, s093; GFX11-SDAG-NEXT: v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s294; GFX11-SDAG-NEXT: global_store_b96 v3, v[0:2], s[4:5]95; GFX11-SDAG-NEXT: s_endpgm96 %v = call <3 x ptr addrspace(5)> @llvm.amdgcn.permlane64.v3p5(<3 x ptr addrspace(5)> %src0)97 store <3 x ptr addrspace(5)> %v, ptr addrspace(1) %out98 ret void99}100 101define amdgpu_kernel void @test_p6(ptr addrspace(1) %out, ptr addrspace(6) %src0) {102; GFX11-SDAG-LABEL: test_p6:103; GFX11-SDAG: ; %bb.0:104; GFX11-SDAG-NEXT: s_clause 0x1105; GFX11-SDAG-NEXT: s_load_b32 s2, s[4:5], 0x2c106; GFX11-SDAG-NEXT: s_load_b64 s[0:1], s[4:5], 0x24107; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)108; GFX11-SDAG-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2109; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]110; GFX11-SDAG-NEXT: s_endpgm111 %v = call ptr addrspace(6) @llvm.amdgcn.permlane64.p6(ptr addrspace(6) %src0)112 store ptr addrspace(6) %v, ptr addrspace(1) %out113 ret void114}115 116define amdgpu_kernel void @test_v3p6(ptr addrspace(1) %out, <3 x ptr addrspace(6)> %src0) {117; GFX11-SDAG-LABEL: test_v3p6:118; GFX11-SDAG: ; %bb.0:119; GFX11-SDAG-NEXT: s_clause 0x1120; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x34121; GFX11-SDAG-NEXT: s_load_b64 s[4:5], s[4:5], 0x24122; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0)123; GFX11-SDAG-NEXT: v_dual_mov_b32 v3, 0 :: v_dual_mov_b32 v0, s0124; GFX11-SDAG-NEXT: v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2125; GFX11-SDAG-NEXT: global_store_b96 v3, v[0:2], s[4:5]126; GFX11-SDAG-NEXT: s_endpgm127 %v = call <3 x ptr addrspace(6)> @llvm.amdgcn.permlane64.v3p6(<3 x ptr addrspace(6)> %src0)128 store <3 x ptr addrspace(6)> %v, ptr addrspace(1) %out129 ret void130}131