1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG %s 3; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s 4 5declare i32 @llvm.amdgcn.permlane64(i32) 6declare i32 @llvm.amdgcn.workitem.id.x() 7 8define amdgpu_kernel void @test_s(i32 addrspace(1)* %out, i32 %src0) { 9; GFX11-LABEL: test_s: 10; GFX11: ; %bb.0: 11; GFX11-NEXT: s_clause 0x1 12; GFX11-NEXT: s_load_b32 s2, s[0:1], 0x2c 13; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 14; GFX11-NEXT: s_waitcnt lgkmcnt(0) 15; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2 16; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) 17; GFX11-NEXT: v_permlane64_b32 v0, v0 18; GFX11-NEXT: global_store_b32 v1, v0, s[0:1] 19; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) 20; GFX11-NEXT: s_endpgm 21 %v = call i32 @llvm.amdgcn.permlane64(i32 %src0) 22 store i32 %v, i32 addrspace(1)* %out 23 ret void 24} 25 26define amdgpu_kernel void @test_i(i32 addrspace(1)* %out) { 27; GFX11-LABEL: test_i: 28; GFX11: ; %bb.0: 29; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 30; GFX11-NEXT: v_dual_mov_b32 v0, 0x63 :: v_dual_mov_b32 v1, 0 31; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) 32; GFX11-NEXT: v_permlane64_b32 v0, v0 33; GFX11-NEXT: s_waitcnt lgkmcnt(0) 34; GFX11-NEXT: global_store_b32 v1, v0, s[0:1] 35; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) 36; GFX11-NEXT: s_endpgm 37 %v = call i32 @llvm.amdgcn.permlane64(i32 99) 38 store i32 %v, i32 addrspace(1)* %out 39 ret void 40} 41 42define amdgpu_kernel void @test_v(i32 addrspace(1)* %out, i32 %src0) #1 { 43; GFX11-SDAG-LABEL: test_v: 44; GFX11-SDAG: ; %bb.0: 45; GFX11-SDAG-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 46; GFX11-SDAG-NEXT: v_mov_b32_e32 v1, 0 47; GFX11-SDAG-NEXT: v_permlane64_b32 v0, v0 48; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) 49; GFX11-SDAG-NEXT: global_store_b32 v1, v0, s[0:1] 50; GFX11-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) 51; GFX11-SDAG-NEXT: s_endpgm 52; 53; GFX11-GISEL-LABEL: test_v: 54; GFX11-GISEL: ; %bb.0: 55; GFX11-GISEL-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 56; GFX11-GISEL-NEXT: v_permlane64_b32 v0, v0 57; GFX11-GISEL-NEXT: v_mov_b32_e32 v1, 0 58; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) 59; GFX11-GISEL-NEXT: global_store_b32 v1, v0, s[0:1] 60; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) 61; GFX11-GISEL-NEXT: s_endpgm 62 %tidx = call i32 @llvm.amdgcn.workitem.id.x() 63 %v = call i32 @llvm.amdgcn.permlane64(i32 %tidx) 64 store i32 %v, i32 addrspace(1)* %out 65 ret void 66} 67