1; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s 2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s 3; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 4 5; FUNC-LABEL: {{^}}sext_bool_icmp_eq_0: 6; GCN-NOT: v_cmp 7; GCN: v_cmp_ne_i32_e32 vcc, 8; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 9; GCN-NEXT:buffer_store_byte [[RESULT]] 10; GCN-NEXT: s_endpgm 11 12; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W 13; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1 14define void @sext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 15 %icmp0 = icmp eq i32 %a, %b 16 %ext = sext i1 %icmp0 to i32 17 %icmp1 = icmp eq i32 %ext, 0 18 store i1 %icmp1, i1 addrspace(1)* %out 19 ret void 20} 21 22; FUNC-LABEL: {{^}}sext_bool_icmp_ne_0: 23; GCN-NOT: v_cmp 24; GCN: v_cmp_ne_i32_e32 vcc, 25; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 26; GCN-NEXT: buffer_store_byte [[RESULT]] 27; GCN-NEXT: s_endpgm 28 29; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W 30; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1 31define void @sext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 32 %icmp0 = icmp ne i32 %a, %b 33 %ext = sext i1 %icmp0 to i32 34 %icmp1 = icmp ne i32 %ext, 0 35 store i1 %icmp1, i1 addrspace(1)* %out 36 ret void 37} 38 39; This really folds away to false 40; FUNC-LABEL: {{^}}sext_bool_icmp_eq_1: 41; GCN: v_cmp_eq_i32_e32 vcc, 42; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, -1, vcc 43; GCN-NEXT: v_cmp_eq_i32_e32 vcc, 1, [[TMP]]{{$}} 44; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, 1, 45; GCN-NEXT: buffer_store_byte [[TMP]] 46; GCN-NEXT: s_endpgm 47define void @sext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 48 %icmp0 = icmp eq i32 %a, %b 49 %ext = sext i1 %icmp0 to i32 50 %icmp1 = icmp eq i32 %ext, 1 51 store i1 %icmp1, i1 addrspace(1)* %out 52 ret void 53} 54 55; This really folds away to true 56; FUNC-LABEL: {{^}}sext_bool_icmp_ne_1: 57; GCN: v_cmp_ne_i32_e32 vcc, 58; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, -1, vcc 59; GCN-NEXT: v_cmp_ne_i32_e32 vcc, 1, [[TMP]]{{$}} 60; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, 1, 61; GCN-NEXT: buffer_store_byte [[TMP]] 62; GCN-NEXT: s_endpgm 63define void @sext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 64 %icmp0 = icmp ne i32 %a, %b 65 %ext = sext i1 %icmp0 to i32 66 %icmp1 = icmp ne i32 %ext, 1 67 store i1 %icmp1, i1 addrspace(1)* %out 68 ret void 69} 70 71; FUNC-LABEL: {{^}}zext_bool_icmp_eq_0: 72; GCN-NOT: v_cmp 73; GCN: v_cmp_ne_i32_e32 vcc, 74; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 75; GCN-NEXT: buffer_store_byte [[RESULT]] 76; GCN-NEXT: s_endpgm 77define void @zext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 78 %icmp0 = icmp eq i32 %a, %b 79 %ext = zext i1 %icmp0 to i32 80 %icmp1 = icmp eq i32 %ext, 0 81 store i1 %icmp1, i1 addrspace(1)* %out 82 ret void 83} 84 85; FUNC-LABEL: {{^}}zext_bool_icmp_ne_0: 86; GCN-NOT: v_cmp 87; GCN: v_cmp_ne_i32_e32 vcc, 88; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 89; GCN-NEXT: buffer_store_byte [[RESULT]] 90; GCN-NEXT: s_endpgm 91define void @zext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 92 %icmp0 = icmp ne i32 %a, %b 93 %ext = zext i1 %icmp0 to i32 94 %icmp1 = icmp ne i32 %ext, 0 95 store i1 %icmp1, i1 addrspace(1)* %out 96 ret void 97} 98 99; FUNC-LABEL: {{^}}zext_bool_icmp_eq_1: 100; GCN-NOT: v_cmp 101; GCN: v_cmp_eq_i32_e32 vcc, 102; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 103; GCN-NEXT: buffer_store_byte [[RESULT]] 104; GCN-NEXT: s_endpgm 105define void @zext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 106 %icmp0 = icmp eq i32 %a, %b 107 %ext = zext i1 %icmp0 to i32 108 %icmp1 = icmp eq i32 %ext, 1 109 store i1 %icmp1, i1 addrspace(1)* %out 110 ret void 111} 112 113; FUNC-LABEL: {{^}}zext_bool_icmp_ne_1: 114; GCN-NOT: v_cmp 115; GCN: v_cmp_eq_i32_e32 vcc, 116; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 117; GCN-NEXT: buffer_store_byte [[RESULT]] 118define void @zext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 119 %icmp0 = icmp ne i32 %a, %b 120 %ext = zext i1 %icmp0 to i32 121 %icmp1 = icmp ne i32 %ext, 1 122 store i1 %icmp1, i1 addrspace(1)* %out 123 ret void 124} 125 126; FUNC-LABEL: {{^}}sext_bool_icmp_ne_k: 127; SI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb 128; SI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xc 129; VI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c 130; VI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x30 131; GCN: v_mov_b32_e32 [[VB:v[0-9]+]], [[B]] 132; GCN: v_cmp_ne_i32_e32 vcc, 2, [[VB]]{{$}} 133; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 134; GCN: buffer_store_byte 135; GCN: s_endpgm 136define void @sext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 137 %icmp0 = icmp ne i32 %a, %b 138 %ext = sext i1 %icmp0 to i32 139 %icmp1 = icmp ne i32 %ext, 2 140 store i1 %icmp1, i1 addrspace(1)* %out 141 ret void 142} 143 144; FUNC-LABEL: {{^}}cmp_zext_k_i8max: 145; GCN: buffer_load_ubyte [[B:v[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0 offset:44 146; GCN: v_mov_b32_e32 [[K255:v[0-9]+]], 0xff{{$}} 147; GCN: v_cmp_ne_i32_e32 vcc, [[K255]], [[B]] 148; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 149; GCN-NEXT: buffer_store_byte [[RESULT]] 150; GCN: s_endpgm 151define void @cmp_zext_k_i8max(i1 addrspace(1)* %out, i8 %b) nounwind { 152 %b.ext = zext i8 %b to i32 153 %icmp0 = icmp ne i32 %b.ext, 255 154 store i1 %icmp0, i1 addrspace(1)* %out 155 ret void 156} 157 158; FUNC-LABEL: {{^}}cmp_sext_k_neg1: 159; GCN: buffer_load_sbyte [[B:v[0-9]+]] 160; GCN: v_cmp_ne_i32_e32 vcc, -1, [[B]]{{$}} 161; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 162; GCN-NEXT: buffer_store_byte [[RESULT]] 163; GCN: s_endpgm 164define void @cmp_sext_k_neg1(i1 addrspace(1)* %out, i8 addrspace(1)* %b.ptr) nounwind { 165 %b = load i8, i8 addrspace(1)* %b.ptr 166 %b.ext = sext i8 %b to i32 167 %icmp0 = icmp ne i32 %b.ext, -1 168 store i1 %icmp0, i1 addrspace(1)* %out 169 ret void 170} 171 172; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_sext_arg: 173; GCN: s_load_dword [[B:s[0-9]+]] 174; GCN: v_cmp_ne_i32_e64 [[CMP:s\[[0-9]+:[0-9]+\]]], -1, [[B]] 175; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[CMP]] 176; GCN-NEXT: buffer_store_byte [[RESULT]] 177; GCN: s_endpgm 178define void @cmp_sext_k_neg1_i8_sext_arg(i1 addrspace(1)* %out, i8 signext %b) nounwind { 179 %b.ext = sext i8 %b to i32 180 %icmp0 = icmp ne i32 %b.ext, -1 181 store i1 %icmp0, i1 addrspace(1)* %out 182 ret void 183} 184 185; FIXME: This ends up doing a buffer_load_ubyte, and and compare to 186; 255. Seems to be because of ordering problems when not allowing load widths to be reduced. 187; Should do a buffer_load_sbyte and compare with -1 188 189; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_arg: 190; GCN-DAG: buffer_load_ubyte [[B:v[0-9]+]] 191; GCN-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0xff{{$}} 192; GCN: v_cmp_ne_i32_e32 vcc, [[K]], [[B]]{{$}} 193; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 194; GCN-NEXT: buffer_store_byte [[RESULT]] 195; GCN: s_endpgm 196define void @cmp_sext_k_neg1_i8_arg(i1 addrspace(1)* %out, i8 %b) nounwind { 197 %b.ext = sext i8 %b to i32 198 %icmp0 = icmp ne i32 %b.ext, -1 199 store i1 %icmp0, i1 addrspace(1)* %out 200 ret void 201} 202 203; FUNC-LABEL: {{^}}cmp_zext_k_neg1: 204; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}} 205; GCN: buffer_store_byte [[RESULT]] 206; GCN: s_endpgm 207define void @cmp_zext_k_neg1(i1 addrspace(1)* %out, i8 %b) nounwind { 208 %b.ext = zext i8 %b to i32 209 %icmp0 = icmp ne i32 %b.ext, -1 210 store i1 %icmp0, i1 addrspace(1)* %out 211 ret void 212} 213 214; FUNC-LABEL: {{^}}zext_bool_icmp_ne_k: 215; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}} 216; GCN: buffer_store_byte [[RESULT]] 217; GCN-NEXT: s_endpgm 218define void @zext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 219 %icmp0 = icmp ne i32 %a, %b 220 %ext = zext i1 %icmp0 to i32 221 %icmp1 = icmp ne i32 %ext, 2 222 store i1 %icmp1, i1 addrspace(1)* %out 223 ret void 224} 225 226; FUNC-LABEL: {{^}}zext_bool_icmp_eq_k: 227; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}} 228; GCN: buffer_store_byte [[RESULT]] 229; GCN-NEXT: s_endpgm 230define void @zext_bool_icmp_eq_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 231 %icmp0 = icmp ne i32 %a, %b 232 %ext = zext i1 %icmp0 to i32 233 %icmp1 = icmp eq i32 %ext, 2 234 store i1 %icmp1, i1 addrspace(1)* %out 235 ret void 236} 237