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: {{^}}sext_bool_icmp_eq_neg1: 72; GCN-NOT: v_cmp 73; GCN: v_cmp_eq_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 @sext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 78 %icmp0 = icmp eq i32 %a, %b 79 %ext = sext i1 %icmp0 to i32 80 %icmp1 = icmp eq i32 %ext, -1 81 store i1 %icmp1, i1 addrspace(1)* %out 82 ret void 83} 84 85; FUNC-LABEL: {{^}}sext_bool_icmp_ne_neg1: 86; GCN-NOT: v_cmp 87; GCN: v_cmp_eq_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 @sext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 92 %icmp0 = icmp ne i32 %a, %b 93 %ext = sext i1 %icmp0 to i32 94 %icmp1 = icmp ne i32 %ext, -1 95 store i1 %icmp1, i1 addrspace(1)* %out 96 ret void 97} 98 99; FUNC-LABEL: {{^}}zext_bool_icmp_eq_0: 100; GCN-NOT: v_cmp 101; GCN: v_cmp_ne_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_0(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, 0 109 store i1 %icmp1, i1 addrspace(1)* %out 110 ret void 111} 112 113; FUNC-LABEL: {{^}}zext_bool_icmp_ne_0: 114; GCN-NOT: v_cmp 115; GCN: v_cmp_ne_i32_e32 vcc, 116; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 117; GCN-NEXT: buffer_store_byte [[RESULT]] 118; GCN-NEXT: s_endpgm 119define void @zext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 120 %icmp0 = icmp ne i32 %a, %b 121 %ext = zext i1 %icmp0 to i32 122 %icmp1 = icmp ne i32 %ext, 0 123 store i1 %icmp1, i1 addrspace(1)* %out 124 ret void 125} 126 127; FUNC-LABEL: {{^}}zext_bool_icmp_eq_1: 128; GCN-NOT: v_cmp 129; GCN: v_cmp_eq_i32_e32 vcc, 130; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 131; GCN-NEXT: buffer_store_byte [[RESULT]] 132; GCN-NEXT: s_endpgm 133define void @zext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 134 %icmp0 = icmp eq i32 %a, %b 135 %ext = zext i1 %icmp0 to i32 136 %icmp1 = icmp eq i32 %ext, 1 137 store i1 %icmp1, i1 addrspace(1)* %out 138 ret void 139} 140 141; FUNC-LABEL: {{^}}zext_bool_icmp_ne_1: 142; GCN-NOT: v_cmp 143; GCN: v_cmp_eq_i32_e32 vcc, 144; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 145; GCN-NEXT: buffer_store_byte [[RESULT]] 146define void @zext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 147 %icmp0 = icmp ne i32 %a, %b 148 %ext = zext i1 %icmp0 to i32 149 %icmp1 = icmp ne i32 %ext, 1 150 store i1 %icmp1, i1 addrspace(1)* %out 151 ret void 152} 153 154; Reduces to false: 155; FUNC-LABEL: {{^}}zext_bool_icmp_eq_neg1: 156; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 0{{$}} 157; GCN: buffer_store_byte [[TMP]] 158; GCN-NEXT: s_endpgm 159define void @zext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 160 %icmp0 = icmp eq i32 %a, %b 161 %ext = zext i1 %icmp0 to i32 162 %icmp1 = icmp eq i32 %ext, -1 163 store i1 %icmp1, i1 addrspace(1)* %out 164 ret void 165} 166 167; Reduces to true: 168; FUNC-LABEL: {{^}}zext_bool_icmp_ne_neg1: 169; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 1{{$}} 170; GCN: buffer_store_byte [[TMP]] 171; GCN-NEXT: s_endpgm 172define void @zext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 173 %icmp0 = icmp ne i32 %a, %b 174 %ext = zext i1 %icmp0 to i32 175 %icmp1 = icmp ne i32 %ext, -1 176 store i1 %icmp1, i1 addrspace(1)* %out 177 ret void 178} 179 180; FUNC-LABEL: {{^}}sext_bool_icmp_ne_k: 181; SI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb 182; SI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xc 183; VI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c 184; VI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x30 185; GCN: v_mov_b32_e32 [[VB:v[0-9]+]], [[B]] 186; GCN: v_cmp_ne_i32_e32 vcc, 2, [[VB]]{{$}} 187; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 188; GCN: buffer_store_byte 189; GCN: s_endpgm 190define void @sext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 191 %icmp0 = icmp ne i32 %a, %b 192 %ext = sext i1 %icmp0 to i32 193 %icmp1 = icmp ne i32 %ext, 2 194 store i1 %icmp1, i1 addrspace(1)* %out 195 ret void 196} 197 198; FUNC-LABEL: {{^}}cmp_zext_k_i8max: 199; SI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb 200; VI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c 201; GCN: s_movk_i32 [[K255:s[0-9]+]], 0xff 202; GCN: s_and_b32 [[B:s[0-9]+]], [[VALUE]], [[K255]] 203; GCN: v_mov_b32_e32 [[VK255:v[0-9]+]], [[K255]] 204; GCN: v_cmp_ne_i32_e32 vcc, [[B]], [[VK255]] 205; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 206; GCN: buffer_store_byte [[RESULT]] 207; GCN: s_endpgm 208define void @cmp_zext_k_i8max(i1 addrspace(1)* %out, i8 %b) nounwind { 209 %b.ext = zext i8 %b to i32 210 %icmp0 = icmp ne i32 %b.ext, 255 211 store i1 %icmp0, i1 addrspace(1)* %out 212 ret void 213} 214 215; FUNC-LABEL: {{^}}cmp_sext_k_neg1: 216; GCN: buffer_load_sbyte [[B:v[0-9]+]] 217; GCN: v_cmp_ne_i32_e32 vcc, -1, [[B]]{{$}} 218; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 219; GCN: buffer_store_byte [[RESULT]] 220; GCN: s_endpgm 221define void @cmp_sext_k_neg1(i1 addrspace(1)* %out, i8 addrspace(1)* %b.ptr) nounwind { 222 %b = load i8, i8 addrspace(1)* %b.ptr 223 %b.ext = sext i8 %b to i32 224 %icmp0 = icmp ne i32 %b.ext, -1 225 store i1 %icmp0, i1 addrspace(1)* %out 226 ret void 227} 228 229; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_sext_arg: 230; GCN: s_load_dword [[B:s[0-9]+]] 231; GCN: v_cmp_ne_i32_e64 [[CMP:s\[[0-9]+:[0-9]+\]]], -1, [[B]] 232; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[CMP]] 233; GCN-NEXT: buffer_store_byte [[RESULT]] 234; GCN: s_endpgm 235define void @cmp_sext_k_neg1_i8_sext_arg(i1 addrspace(1)* %out, i8 signext %b) nounwind { 236 %b.ext = sext i8 %b to i32 237 %icmp0 = icmp ne i32 %b.ext, -1 238 store i1 %icmp0, i1 addrspace(1)* %out 239 ret void 240} 241 242; FIXME: This ends up doing a buffer_load_ubyte, and and compare to 243; 255. Seems to be because of ordering problems when not allowing load widths to be reduced. 244; Should do a buffer_load_sbyte and compare with -1 245 246; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_arg: 247; SI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb 248; VI: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c 249; GCN: s_movk_i32 [[K:s[0-9]+]], 0xff 250; GCN: s_and_b32 [[B:s[0-9]+]], [[VAL]], [[K]] 251; GCN: v_mov_b32_e32 [[VK:v[0-9]+]], [[K]] 252; GCN: v_cmp_ne_i32_e32 vcc, [[B]], [[VK]]{{$}} 253; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc 254; GCN: buffer_store_byte [[RESULT]] 255; GCN: s_endpgm 256define void @cmp_sext_k_neg1_i8_arg(i1 addrspace(1)* %out, i8 %b) nounwind { 257 %b.ext = sext i8 %b to i32 258 %icmp0 = icmp ne i32 %b.ext, -1 259 store i1 %icmp0, i1 addrspace(1)* %out 260 ret void 261} 262 263; FUNC-LABEL: {{^}}cmp_zext_k_neg1: 264; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}} 265; GCN: buffer_store_byte [[RESULT]] 266; GCN: s_endpgm 267define void @cmp_zext_k_neg1(i1 addrspace(1)* %out, i8 %b) nounwind { 268 %b.ext = zext i8 %b to i32 269 %icmp0 = icmp ne i32 %b.ext, -1 270 store i1 %icmp0, i1 addrspace(1)* %out 271 ret void 272} 273 274; FUNC-LABEL: {{^}}zext_bool_icmp_ne_k: 275; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}} 276; GCN: buffer_store_byte [[RESULT]] 277; GCN-NEXT: s_endpgm 278define void @zext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 279 %icmp0 = icmp ne i32 %a, %b 280 %ext = zext i1 %icmp0 to i32 281 %icmp1 = icmp ne i32 %ext, 2 282 store i1 %icmp1, i1 addrspace(1)* %out 283 ret void 284} 285 286; FUNC-LABEL: {{^}}zext_bool_icmp_eq_k: 287; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}} 288; GCN: buffer_store_byte [[RESULT]] 289; GCN-NEXT: s_endpgm 290define void @zext_bool_icmp_eq_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 291 %icmp0 = icmp ne i32 %a, %b 292 %ext = zext i1 %icmp0 to i32 293 %icmp1 = icmp eq i32 %ext, 2 294 store i1 %icmp1, i1 addrspace(1)* %out 295 ret void 296} 297