127 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 62; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s3 4define i32 @select_load_i32_p0(i1 %cond, ptr %a, ptr %b) {5; CHECK-LABEL: select_load_i32_p0:6; CHECK: ; %bb.0:7; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)8; CHECK-NEXT: v_and_b32_e32 v0, 1, v09; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v010; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc11; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc12; CHECK-NEXT: flat_load_dword v0, v[1:2]13; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)14; CHECK-NEXT: s_setpc_b64 s[30:31]15 %ld0 = load i32, ptr %a16 %ld1 = load i32, ptr %b17 %select = select i1 %cond, i32 %ld0, i32 %ld118 ret i32 %select19}20 21define i32 @select_load_i32_p1(i1 %cond, ptr addrspace(1) %a, ptr addrspace(1) %b) {22; CHECK-LABEL: select_load_i32_p1:23; CHECK: ; %bb.0:24; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)25; CHECK-NEXT: v_and_b32_e32 v0, 1, v026; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v027; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc28; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc29; CHECK-NEXT: global_load_dword v0, v[1:2], off30; CHECK-NEXT: s_waitcnt vmcnt(0)31; CHECK-NEXT: s_setpc_b64 s[30:31]32 %ld0 = load i32, ptr addrspace(1) %a33 %ld1 = load i32, ptr addrspace(1) %b34 %select = select i1 %cond, i32 %ld0, i32 %ld135 ret i32 %select36}37 38define i32 @select_load_i32_p3(i1 %cond, ptr addrspace(3) %a, ptr addrspace(3) %b) {39; CHECK-LABEL: select_load_i32_p3:40; CHECK: ; %bb.0:41; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)42; CHECK-NEXT: v_and_b32_e32 v0, 1, v043; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v044; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc45; CHECK-NEXT: ds_read_b32 v0, v046; CHECK-NEXT: s_waitcnt lgkmcnt(0)47; CHECK-NEXT: s_setpc_b64 s[30:31]48 %ld0 = load i32, ptr addrspace(3) %a49 %ld1 = load i32, ptr addrspace(3) %b50 %select = select i1 %cond, i32 %ld0, i32 %ld151 ret i32 %select52}53 54define i32 @select_load_i32_p0_p1(i1 %cond, ptr %a, ptr addrspace(1) %b) {55; CHECK-LABEL: select_load_i32_p0_p1:56; CHECK: ; %bb.0:57; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)58; CHECK-NEXT: flat_load_dword v5, v[1:2]59; CHECK-NEXT: global_load_dword v6, v[3:4], off60; CHECK-NEXT: v_and_b32_e32 v0, 1, v061; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v062; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)63; CHECK-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc64; CHECK-NEXT: s_setpc_b64 s[30:31]65 %ld0 = load i32, ptr %a66 %ld1 = load i32, ptr addrspace(1) %b67 %select = select i1 %cond, i32 %ld0, i32 %ld168 ret i32 %select69}70 71define i32 @select_load_i32_p1_p0(i1 %cond, ptr addrspace(1) %a, ptr %b) {72; CHECK-LABEL: select_load_i32_p1_p0:73; CHECK: ; %bb.0:74; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)75; CHECK-NEXT: global_load_dword v5, v[1:2], off76; CHECK-NEXT: flat_load_dword v6, v[3:4]77; CHECK-NEXT: v_and_b32_e32 v0, 1, v078; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v079; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)80; CHECK-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc81; CHECK-NEXT: s_setpc_b64 s[30:31]82 %ld0 = load i32, ptr addrspace(1) %a83 %ld1 = load i32, ptr %b84 %select = select i1 %cond, i32 %ld0, i32 %ld185 ret i32 %select86}87 88define i8 @select_load_i8_p1(i1 %cond, ptr addrspace(1) %a, ptr addrspace(1) %b) {89; CHECK-LABEL: select_load_i8_p1:90; CHECK: ; %bb.0:91; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)92; CHECK-NEXT: v_and_b32_e32 v0, 1, v093; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v094; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc95; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc96; CHECK-NEXT: global_load_ubyte v0, v[1:2], off97; CHECK-NEXT: s_waitcnt vmcnt(0)98; CHECK-NEXT: s_setpc_b64 s[30:31]99 %ld0 = load i8, ptr addrspace(1) %a100 %ld1 = load i8, ptr addrspace(1) %b101 %select = select i1 %cond, i8 %ld0, i8 %ld1102 ret i8 %select103}104 105define i32 @select_load_i32_p1_offset(i1 %cond, ptr addrspace(1) %a, ptr addrspace(1) %b) {106; CHECK-LABEL: select_load_i32_p1_offset:107; CHECK: ; %bb.0:108; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)109; CHECK-NEXT: v_add_co_u32_e32 v3, vcc, 0x100, v1110; CHECK-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v2, vcc111; CHECK-NEXT: v_add_co_u32_e32 v5, vcc, 0x200, v1112; CHECK-NEXT: v_and_b32_e32 v0, 1, v0113; CHECK-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v2, vcc114; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0115; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v4, vcc116; CHECK-NEXT: v_cndmask_b32_e32 v0, v5, v3, vcc117; CHECK-NEXT: global_load_dword v0, v[0:1], off118; CHECK-NEXT: s_waitcnt vmcnt(0)119; CHECK-NEXT: s_setpc_b64 s[30:31]120 %gep.a = getelementptr i8, ptr addrspace(1) %a, i64 256121 %gep.b = getelementptr i8, ptr addrspace(1) %a, i64 512122 %ld0 = load i32, ptr addrspace(1) %gep.a123 %ld1 = load i32, ptr addrspace(1) %gep.b124 %select = select i1 %cond, i32 %ld0, i32 %ld1125 ret i32 %select126}127