1;RUN: llc < %s -march=amdgcn -mcpu=gfx900 -verify-machineinstrs | FileCheck %s --check-prefix=GCN 2 3; GCN-LABEL: {{^}}full_mask: 4; GCN: s_mov_b64 exec, -1 5; GCN: v_add_f32_e32 v0, 6define amdgpu_ps float @full_mask(float %a, float %b) { 7main_body: 8 %s = fadd float %a, %b 9 call void @llvm.amdgcn.init.exec(i64 -1) 10 ret float %s 11} 12 13; GCN-LABEL: {{^}}partial_mask: 14; GCN: s_mov_b64 exec, 0x1e240 15; GCN: v_add_f32_e32 v0, 16define amdgpu_ps float @partial_mask(float %a, float %b) { 17main_body: 18 %s = fadd float %a, %b 19 call void @llvm.amdgcn.init.exec(i64 123456) 20 ret float %s 21} 22 23; GCN-LABEL: {{^}}input_s3off8: 24; GCN: s_bfe_u32 s0, s3, 0x70008 25; GCN: s_bfm_b64 exec, s0, 0 26; GCN: s_cmp_eq_u32 s0, 64 27; GCN: s_cmov_b64 exec, -1 28; GCN: v_add_f32_e32 v0, 29define amdgpu_ps float @input_s3off8(i32 inreg, i32 inreg, i32 inreg, i32 inreg %count, float %a, float %b) { 30main_body: 31 %s = fadd float %a, %b 32 call void @llvm.amdgcn.init.exec.from.input(i32 %count, i32 8) 33 ret float %s 34} 35 36; GCN-LABEL: {{^}}input_s0off19: 37; GCN: s_bfe_u32 s0, s0, 0x70013 38; GCN: s_bfm_b64 exec, s0, 0 39; GCN: s_cmp_eq_u32 s0, 64 40; GCN: s_cmov_b64 exec, -1 41; GCN: v_add_f32_e32 v0, 42define amdgpu_ps float @input_s0off19(i32 inreg %count, float %a, float %b) { 43main_body: 44 %s = fadd float %a, %b 45 call void @llvm.amdgcn.init.exec.from.input(i32 %count, i32 19) 46 ret float %s 47} 48 49; GCN-LABEL: {{^}}reuse_input: 50; GCN: s_bfe_u32 s1, s0, 0x70013 51; GCN: s_bfm_b64 exec, s1, 0 52; GCN: s_cmp_eq_u32 s1, 64 53; GCN: s_cmov_b64 exec, -1 54; GCN: v_add_u32_e32 v0, s0, v0 55define amdgpu_ps float @reuse_input(i32 inreg %count, i32 %a) { 56main_body: 57 call void @llvm.amdgcn.init.exec.from.input(i32 %count, i32 19) 58 %s = add i32 %a, %count 59 %f = sitofp i32 %s to float 60 ret float %f 61} 62 63; GCN-LABEL: {{^}}reuse_input2: 64; GCN: s_bfe_u32 s1, s0, 0x70013 65; GCN: s_bfm_b64 exec, s1, 0 66; GCN: s_cmp_eq_u32 s1, 64 67; GCN: s_cmov_b64 exec, -1 68; GCN: v_add_u32_e32 v0, s0, v0 69define amdgpu_ps float @reuse_input2(i32 inreg %count, i32 %a) { 70main_body: 71 %s = add i32 %a, %count 72 %f = sitofp i32 %s to float 73 call void @llvm.amdgcn.init.exec.from.input(i32 %count, i32 19) 74 ret float %f 75} 76 77; GCN-LABEL: {{^}}init_unreachable: 78; 79; This used to crash. 80define amdgpu_ps void @init_unreachable() { 81main_body: 82 call void @llvm.amdgcn.init.exec(i64 -1) 83 unreachable 84} 85 86declare void @llvm.amdgcn.init.exec(i64) #1 87declare void @llvm.amdgcn.init.exec.from.input(i32, i32) #1 88 89attributes #1 = { convergent } 90