1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 4 5define i32 @knownbits_mask_extract_sext(<8 x i16> %a0) nounwind { 6; X32-LABEL: knownbits_mask_extract_sext: 7; X32: # %bb.0: 8; X32-NEXT: vmovd %xmm0, %eax 9; X32-NEXT: andl $15, %eax 10; X32-NEXT: retl 11; 12; X64-LABEL: knownbits_mask_extract_sext: 13; X64: # %bb.0: 14; X64-NEXT: vmovd %xmm0, %eax 15; X64-NEXT: andl $15, %eax 16; X64-NEXT: retq 17 %1 = and <8 x i16> %a0, <i16 15, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1> 18 %2 = extractelement <8 x i16> %1, i32 0 19 %3 = sext i16 %2 to i32 20 ret i32 %3 21} 22 23define float @knownbits_mask_extract_uitofp(<2 x i64> %a0) nounwind { 24; X32-LABEL: knownbits_mask_extract_uitofp: 25; X32: # %bb.0: 26; X32-NEXT: pushl %eax 27; X32-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero 28; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 29; X32-NEXT: vmovss %xmm0, (%esp) 30; X32-NEXT: flds (%esp) 31; X32-NEXT: popl %eax 32; X32-NEXT: retl 33; 34; X64-LABEL: knownbits_mask_extract_uitofp: 35; X64: # %bb.0: 36; X64-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero 37; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 38; X64-NEXT: retq 39 %1 = and <2 x i64> %a0, <i64 65535, i64 -1> 40 %2 = extractelement <2 x i64> %1, i32 0 41 %3 = uitofp i64 %2 to float 42 ret float %3 43} 44 45define <4 x float> @knownbits_insert_uitofp(<4 x i32> %a0, i16 %a1, i16 %a2) nounwind { 46; X32-LABEL: knownbits_insert_uitofp: 47; X32: # %bb.0: 48; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax 49; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx 50; X32-NEXT: vmovd %ecx, %xmm0 51; X32-NEXT: vpinsrd $2, %eax, %xmm0, %xmm0 52; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,2] 53; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 54; X32-NEXT: retl 55; 56; X64-LABEL: knownbits_insert_uitofp: 57; X64: # %bb.0: 58; X64-NEXT: movzwl %di, %eax 59; X64-NEXT: movzwl %si, %ecx 60; X64-NEXT: vmovd %eax, %xmm0 61; X64-NEXT: vpinsrd $2, %ecx, %xmm0, %xmm0 62; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,2] 63; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 64; X64-NEXT: retq 65 %1 = zext i16 %a1 to i32 66 %2 = zext i16 %a2 to i32 67 %3 = insertelement <4 x i32> %a0, i32 %1, i32 0 68 %4 = insertelement <4 x i32> %3, i32 %2, i32 2 69 %5 = shufflevector <4 x i32> %4, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2> 70 %6 = uitofp <4 x i32> %5 to <4 x float> 71 ret <4 x float> %6 72} 73 74define <4 x i32> @knownbits_mask_shuffle_sext(<8 x i16> %a0) nounwind { 75; X32-LABEL: knownbits_mask_shuffle_sext: 76; X32: # %bb.0: 77; X32-NEXT: vpand {{\.LCPI.*}}, %xmm0, %xmm0 78; X32-NEXT: vpxor %xmm1, %xmm1, %xmm1 79; X32-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 80; X32-NEXT: retl 81; 82; X64-LABEL: knownbits_mask_shuffle_sext: 83; X64: # %bb.0: 84; X64-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 85; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1 86; X64-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 87; X64-NEXT: retq 88 %1 = and <8 x i16> %a0, <i16 -1, i16 -1, i16 -1, i16 -1, i16 15, i16 15, i16 15, i16 15> 89 %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 90 %3 = sext <4 x i16> %2 to <4 x i32> 91 ret <4 x i32> %3 92} 93 94define <4 x i32> @knownbits_mask_shuffle_shuffle_sext(<8 x i16> %a0) nounwind { 95; X32-LABEL: knownbits_mask_shuffle_shuffle_sext: 96; X32: # %bb.0: 97; X32-NEXT: vpand {{\.LCPI.*}}, %xmm0, %xmm0 98; X32-NEXT: vpxor %xmm1, %xmm1, %xmm1 99; X32-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 100; X32-NEXT: retl 101; 102; X64-LABEL: knownbits_mask_shuffle_shuffle_sext: 103; X64: # %bb.0: 104; X64-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 105; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1 106; X64-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] 107; X64-NEXT: retq 108 %1 = and <8 x i16> %a0, <i16 -1, i16 -1, i16 -1, i16 -1, i16 15, i16 15, i16 15, i16 15> 109 %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef> 110 %3 = shufflevector <8 x i16> %2, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 111 %4 = sext <4 x i16> %3 to <4 x i32> 112 ret <4 x i32> %4 113} 114 115define <4 x i32> @knownbits_mask_shuffle_shuffle_undef_sext(<8 x i16> %a0) nounwind { 116; X32-LABEL: knownbits_mask_shuffle_shuffle_undef_sext: 117; X32: # %bb.0: 118; X32-NEXT: vpand {{\.LCPI.*}}, %xmm0, %xmm0 119; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3] 120; X32-NEXT: vpmovsxwd %xmm0, %xmm0 121; X32-NEXT: retl 122; 123; X64-LABEL: knownbits_mask_shuffle_shuffle_undef_sext: 124; X64: # %bb.0: 125; X64-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 126; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3] 127; X64-NEXT: vpmovsxwd %xmm0, %xmm0 128; X64-NEXT: retq 129 %1 = and <8 x i16> %a0, <i16 -1, i16 -1, i16 -1, i16 -1, i16 15, i16 15, i16 15, i16 15> 130 %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> 131 %3 = shufflevector <8 x i16> %2, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 132 %4 = sext <4 x i16> %3 to <4 x i32> 133 ret <4 x i32> %4 134} 135 136define <4 x float> @knownbits_mask_shuffle_uitofp(<4 x i32> %a0) nounwind { 137; X32-LABEL: knownbits_mask_shuffle_uitofp: 138; X32: # %bb.0: 139; X32-NEXT: vandps {{\.LCPI.*}}, %xmm0, %xmm0 140; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3] 141; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 142; X32-NEXT: retl 143; 144; X64-LABEL: knownbits_mask_shuffle_uitofp: 145; X64: # %bb.0: 146; X64-NEXT: vandps {{.*}}(%rip), %xmm0, %xmm0 147; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3] 148; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 149; X64-NEXT: retq 150 %1 = and <4 x i32> %a0, <i32 -1, i32 -1, i32 255, i32 4085> 151 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 2, i32 3, i32 3> 152 %3 = uitofp <4 x i32> %2 to <4 x float> 153 ret <4 x float> %3 154} 155 156define <4 x float> @knownbits_mask_or_shuffle_uitofp(<4 x i32> %a0) nounwind { 157; X32-LABEL: knownbits_mask_or_shuffle_uitofp: 158; X32: # %bb.0: 159; X32-NEXT: vmovaps {{.*#+}} xmm0 = [6.5535E+4,6.5535E+4,6.5535E+4,6.5535E+4] 160; X32-NEXT: retl 161; 162; X64-LABEL: knownbits_mask_or_shuffle_uitofp: 163; X64: # %bb.0: 164; X64-NEXT: vmovaps {{.*#+}} xmm0 = [6.5535E+4,6.5535E+4,6.5535E+4,6.5535E+4] 165; X64-NEXT: retq 166 %1 = and <4 x i32> %a0, <i32 -1, i32 -1, i32 255, i32 4085> 167 %2 = or <4 x i32> %1, <i32 65535, i32 65535, i32 65535, i32 65535> 168 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 2, i32 2, i32 3, i32 3> 169 %4 = uitofp <4 x i32> %3 to <4 x float> 170 ret <4 x float> %4 171} 172 173define <4 x float> @knownbits_mask_xor_shuffle_uitofp(<4 x i32> %a0) nounwind { 174; X32-LABEL: knownbits_mask_xor_shuffle_uitofp: 175; X32: # %bb.0: 176; X32-NEXT: vandps {{\.LCPI.*}}, %xmm0, %xmm0 177; X32-NEXT: vxorps {{\.LCPI.*}}, %xmm0, %xmm0 178; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3] 179; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 180; X32-NEXT: retl 181; 182; X64-LABEL: knownbits_mask_xor_shuffle_uitofp: 183; X64: # %bb.0: 184; X64-NEXT: vandps {{.*}}(%rip), %xmm0, %xmm0 185; X64-NEXT: vxorps {{.*}}(%rip), %xmm0, %xmm0 186; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3] 187; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 188; X64-NEXT: retq 189 %1 = and <4 x i32> %a0, <i32 -1, i32 -1, i32 255, i32 4085> 190 %2 = xor <4 x i32> %1, <i32 65535, i32 65535, i32 65535, i32 65535> 191 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 2, i32 2, i32 3, i32 3> 192 %4 = uitofp <4 x i32> %3 to <4 x float> 193 ret <4 x float> %4 194} 195 196define <4 x i32> @knownbits_mask_shl_shuffle_lshr(<4 x i32> %a0) nounwind { 197; X32-LABEL: knownbits_mask_shl_shuffle_lshr: 198; X32: # %bb.0: 199; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 200; X32-NEXT: retl 201; 202; X64-LABEL: knownbits_mask_shl_shuffle_lshr: 203; X64: # %bb.0: 204; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 205; X64-NEXT: retq 206 %1 = and <4 x i32> %a0, <i32 -65536, i32 -7, i32 -7, i32 -65536> 207 %2 = shl <4 x i32> %1, <i32 17, i32 17, i32 17, i32 17> 208 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 209 %4 = lshr <4 x i32> %3, <i32 15, i32 15, i32 15, i32 15> 210 ret <4 x i32> %4 211} 212 213define <4 x i32> @knownbits_mask_ashr_shuffle_lshr(<4 x i32> %a0) nounwind { 214; X32-LABEL: knownbits_mask_ashr_shuffle_lshr: 215; X32: # %bb.0: 216; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 217; X32-NEXT: retl 218; 219; X64-LABEL: knownbits_mask_ashr_shuffle_lshr: 220; X64: # %bb.0: 221; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 222; X64-NEXT: retq 223 %1 = and <4 x i32> %a0, <i32 131071, i32 -1, i32 -1, i32 131071> 224 %2 = ashr <4 x i32> %1, <i32 15, i32 15, i32 15, i32 15> 225 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 226 %4 = lshr <4 x i32> %3, <i32 30, i32 30, i32 30, i32 30> 227 ret <4 x i32> %4 228} 229 230define <4 x i32> @knownbits_mask_mul_shuffle_shl(<4 x i32> %a0, <4 x i32> %a1) nounwind { 231; X32-LABEL: knownbits_mask_mul_shuffle_shl: 232; X32: # %bb.0: 233; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 234; X32-NEXT: retl 235; 236; X64-LABEL: knownbits_mask_mul_shuffle_shl: 237; X64: # %bb.0: 238; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 239; X64-NEXT: retq 240 %1 = and <4 x i32> %a0, <i32 -65536, i32 -7, i32 -7, i32 -65536> 241 %2 = mul <4 x i32> %a1, %1 242 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 243 %4 = shl <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 244 ret <4 x i32> %4 245} 246 247define <4 x i32> @knownbits_mask_trunc_shuffle_shl(<4 x i64> %a0) nounwind { 248; X32-LABEL: knownbits_mask_trunc_shuffle_shl: 249; X32: # %bb.0: 250; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 251; X32-NEXT: retl 252; 253; X64-LABEL: knownbits_mask_trunc_shuffle_shl: 254; X64: # %bb.0: 255; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 256; X64-NEXT: retq 257 %1 = and <4 x i64> %a0, <i64 -65536, i64 -7, i64 7, i64 -65536> 258 %2 = trunc <4 x i64> %1 to <4 x i32> 259 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 260 %4 = shl <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 261 ret <4 x i32> %4 262} 263 264define <4 x i32> @knownbits_mask_add_shuffle_lshr(<4 x i32> %a0, <4 x i32> %a1) nounwind { 265; X32-LABEL: knownbits_mask_add_shuffle_lshr: 266; X32: # %bb.0: 267; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 268; X32-NEXT: retl 269; 270; X64-LABEL: knownbits_mask_add_shuffle_lshr: 271; X64: # %bb.0: 272; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 273; X64-NEXT: retq 274 %1 = and <4 x i32> %a0, <i32 32767, i32 -1, i32 -1, i32 32767> 275 %2 = and <4 x i32> %a1, <i32 32767, i32 -1, i32 -1, i32 32767> 276 %3 = add <4 x i32> %1, %2 277 %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 278 %5 = lshr <4 x i32> %4, <i32 17, i32 17, i32 17, i32 17> 279 ret <4 x i32> %5 280} 281 282define <4 x i32> @knownbits_mask_sub_shuffle_lshr(<4 x i32> %a0) nounwind { 283; X32-LABEL: knownbits_mask_sub_shuffle_lshr: 284; X32: # %bb.0: 285; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 286; X32-NEXT: retl 287; 288; X64-LABEL: knownbits_mask_sub_shuffle_lshr: 289; X64: # %bb.0: 290; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 291; X64-NEXT: retq 292 %1 = and <4 x i32> %a0, <i32 15, i32 -1, i32 -1, i32 15> 293 %2 = sub <4 x i32> <i32 255, i32 255, i32 255, i32 255>, %1 294 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 295 %4 = lshr <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 296 ret <4 x i32> %4 297} 298 299define <4 x i32> @knownbits_mask_udiv_shuffle_lshr(<4 x i32> %a0, <4 x i32> %a1) nounwind { 300; X32-LABEL: knownbits_mask_udiv_shuffle_lshr: 301; X32: # %bb.0: 302; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 303; X32-NEXT: retl 304; 305; X64-LABEL: knownbits_mask_udiv_shuffle_lshr: 306; X64: # %bb.0: 307; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 308; X64-NEXT: retq 309 %1 = and <4 x i32> %a0, <i32 32767, i32 -1, i32 -1, i32 32767> 310 %2 = udiv <4 x i32> %1, %a1 311 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 312 %4 = lshr <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 313 ret <4 x i32> %4 314} 315 316define <4 x i32> @knownbits_urem_lshr(<4 x i32> %a0) nounwind { 317; X32-LABEL: knownbits_urem_lshr: 318; X32: # %bb.0: 319; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 320; X32-NEXT: retl 321; 322; X64-LABEL: knownbits_urem_lshr: 323; X64: # %bb.0: 324; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 325; X64-NEXT: retq 326 %1 = urem <4 x i32> %a0, <i32 16, i32 16, i32 16, i32 16> 327 %2 = lshr <4 x i32> %1, <i32 22, i32 22, i32 22, i32 22> 328 ret <4 x i32> %2 329} 330 331define <4 x i32> @knownbits_mask_urem_shuffle_lshr(<4 x i32> %a0, <4 x i32> %a1) nounwind { 332; X32-LABEL: knownbits_mask_urem_shuffle_lshr: 333; X32: # %bb.0: 334; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 335; X32-NEXT: retl 336; 337; X64-LABEL: knownbits_mask_urem_shuffle_lshr: 338; X64: # %bb.0: 339; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 340; X64-NEXT: retq 341 %1 = and <4 x i32> %a0, <i32 32767, i32 -1, i32 -1, i32 32767> 342 %2 = and <4 x i32> %a1, <i32 32767, i32 -1, i32 -1, i32 32767> 343 %3 = urem <4 x i32> %1, %2 344 %4 = shufflevector <4 x i32> %3, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 345 %5 = lshr <4 x i32> %4, <i32 22, i32 22, i32 22, i32 22> 346 ret <4 x i32> %5 347} 348 349define <4 x i32> @knownbits_mask_srem_shuffle_lshr(<4 x i32> %a0) nounwind { 350; X32-LABEL: knownbits_mask_srem_shuffle_lshr: 351; X32: # %bb.0: 352; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 353; X32-NEXT: retl 354; 355; X64-LABEL: knownbits_mask_srem_shuffle_lshr: 356; X64: # %bb.0: 357; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 358; X64-NEXT: retq 359 %1 = and <4 x i32> %a0, <i32 -32768, i32 -1, i32 -1, i32 -32768> 360 %2 = srem <4 x i32> %1, <i32 16, i32 16, i32 16, i32 16> 361 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 362 %4 = lshr <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 363 ret <4 x i32> %4 364} 365 366define <4 x i32> @knownbits_mask_bswap_shuffle_shl(<4 x i32> %a0) nounwind { 367; X32-LABEL: knownbits_mask_bswap_shuffle_shl: 368; X32: # %bb.0: 369; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 370; X32-NEXT: retl 371; 372; X64-LABEL: knownbits_mask_bswap_shuffle_shl: 373; X64: # %bb.0: 374; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 375; X64-NEXT: retq 376 %1 = and <4 x i32> %a0, <i32 32767, i32 -1, i32 -1, i32 32767> 377 %2 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %1) 378 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 379 %4 = shl <4 x i32> %3, <i32 22, i32 22, i32 22, i32 22> 380 ret <4 x i32> %4 381} 382declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>) 383 384define <8 x float> @knownbits_mask_concat_uitofp(<4 x i32> %a0, <4 x i32> %a1) nounwind { 385; X32-LABEL: knownbits_mask_concat_uitofp: 386; X32: # %bb.0: 387; X32-NEXT: vandps {{\.LCPI.*}}, %xmm0, %xmm0 388; X32-NEXT: vandps {{\.LCPI.*}}, %xmm1, %xmm1 389; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,2,0,2] 390; X32-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[1,3,1,3] 391; X32-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 392; X32-NEXT: vcvtdq2ps %ymm0, %ymm0 393; X32-NEXT: retl 394; 395; X64-LABEL: knownbits_mask_concat_uitofp: 396; X64: # %bb.0: 397; X64-NEXT: vandps {{.*}}(%rip), %xmm0, %xmm0 398; X64-NEXT: vandps {{.*}}(%rip), %xmm1, %xmm1 399; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,2,0,2] 400; X64-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[1,3,1,3] 401; X64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 402; X64-NEXT: vcvtdq2ps %ymm0, %ymm0 403; X64-NEXT: retq 404 %1 = and <4 x i32> %a0, <i32 131071, i32 -1, i32 131071, i32 -1> 405 %2 = and <4 x i32> %a1, <i32 -1, i32 131071, i32 -1, i32 131071> 406 %3 = shufflevector <4 x i32> %1, <4 x i32> %2, <8 x i32> <i32 0, i32 2, i32 0, i32 2, i32 5, i32 7, i32 5, i32 7> 407 %4 = uitofp <8 x i32> %3 to <8 x float> 408 ret <8 x float> %4 409} 410 411define <4 x float> @knownbits_lshr_bitcast_shuffle_uitofp(<2 x i64> %a0, <4 x i32> %a1) nounwind { 412; X32-LABEL: knownbits_lshr_bitcast_shuffle_uitofp: 413; X32: # %bb.0: 414; X32-NEXT: vpsrlq $1, %xmm0, %xmm0 415; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3] 416; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 417; X32-NEXT: retl 418; 419; X64-LABEL: knownbits_lshr_bitcast_shuffle_uitofp: 420; X64: # %bb.0: 421; X64-NEXT: vpsrlq $1, %xmm0, %xmm0 422; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3] 423; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 424; X64-NEXT: retq 425 %1 = lshr <2 x i64> %a0, <i64 1, i64 1> 426 %2 = bitcast <2 x i64> %1 to <4 x i32> 427 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 3, i32 3> 428 %4 = uitofp <4 x i32> %3 to <4 x float> 429 ret <4 x float> %4 430} 431 432define <4 x float> @knownbits_smax_smin_shuffle_uitofp(<4 x i32> %a0) { 433; X32-LABEL: knownbits_smax_smin_shuffle_uitofp: 434; X32: # %bb.0: 435; X32-NEXT: vpminsd {{\.LCPI.*}}, %xmm0, %xmm0 436; X32-NEXT: vpmaxsd {{\.LCPI.*}}, %xmm0, %xmm0 437; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 438; X32-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 439; X32-NEXT: vpsrld $16, %xmm0, %xmm0 440; X32-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 441; X32-NEXT: vsubps {{\.LCPI.*}}, %xmm0, %xmm0 442; X32-NEXT: vaddps %xmm0, %xmm1, %xmm0 443; X32-NEXT: retl 444; 445; X64-LABEL: knownbits_smax_smin_shuffle_uitofp: 446; X64: # %bb.0: 447; X64-NEXT: vpminsd {{.*}}(%rip), %xmm0, %xmm0 448; X64-NEXT: vpmaxsd {{.*}}(%rip), %xmm0, %xmm0 449; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 450; X64-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 451; X64-NEXT: vpsrld $16, %xmm0, %xmm0 452; X64-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 453; X64-NEXT: vsubps {{.*}}(%rip), %xmm0, %xmm0 454; X64-NEXT: vaddps %xmm0, %xmm1, %xmm0 455; X64-NEXT: retq 456 %1 = call <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32> %a0, <4 x i32> <i32 0, i32 -65535, i32 -65535, i32 0>) 457 %2 = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> %1, <4 x i32> <i32 65535, i32 -1, i32 -1, i32 131071>) 458 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 459 %4 = uitofp <4 x i32> %3 to <4 x float> 460 ret <4 x float> %4 461} 462declare <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32>, <4 x i32>) nounwind readnone 463declare <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32>, <4 x i32>) nounwind readnone 464 465define <4 x float> @knownbits_umin_shuffle_uitofp(<4 x i32> %a0) { 466; X32-LABEL: knownbits_umin_shuffle_uitofp: 467; X32: # %bb.0: 468; X32-NEXT: vpminud {{\.LCPI.*}}, %xmm0, %xmm0 469; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 470; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 471; X32-NEXT: retl 472; 473; X64-LABEL: knownbits_umin_shuffle_uitofp: 474; X64: # %bb.0: 475; X64-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm0 476; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 477; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 478; X64-NEXT: retq 479 %1 = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %a0, <4 x i32> <i32 65535, i32 -1, i32 -1, i32 262143>) 480 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 481 %3 = uitofp <4 x i32> %2 to <4 x float> 482 ret <4 x float> %3 483} 484declare <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32>, <4 x i32>) nounwind readnone 485declare <4 x i32> @llvm.x86.sse41.pminud(<4 x i32>, <4 x i32>) nounwind readnone 486 487define <4 x i32> @knownbits_umax_shuffle_ashr(<4 x i32> %a0) { 488; X32-LABEL: knownbits_umax_shuffle_ashr: 489; X32: # %bb.0: 490; X32-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0 491; X32-NEXT: retl 492; 493; X64-LABEL: knownbits_umax_shuffle_ashr: 494; X64: # %bb.0: 495; X64-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0 496; X64-NEXT: retq 497 %1 = call <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32> %a0, <4 x i32> <i32 65535, i32 -1, i32 -1, i32 262143>) 498 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2> 499 %3 = ashr <4 x i32> %2, <i32 31, i32 31, i32 31, i32 31> 500 ret <4 x i32> %3 501} 502 503define <4 x float> @knownbits_mask_umax_shuffle_uitofp(<4 x i32> %a0) { 504; X32-LABEL: knownbits_mask_umax_shuffle_uitofp: 505; X32: # %bb.0: 506; X32-NEXT: vpand {{\.LCPI.*}}, %xmm0, %xmm0 507; X32-NEXT: vpmaxud {{\.LCPI.*}}, %xmm0, %xmm0 508; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 509; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 510; X32-NEXT: retl 511; 512; X64-LABEL: knownbits_mask_umax_shuffle_uitofp: 513; X64: # %bb.0: 514; X64-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 515; X64-NEXT: vpmaxud {{.*}}(%rip), %xmm0, %xmm0 516; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3] 517; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 518; X64-NEXT: retq 519 %1 = and <4 x i32> %a0, <i32 65535, i32 -1, i32 -1, i32 262143> 520 %2 = call <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32> %1, <4 x i32> <i32 255, i32 -1, i32 -1, i32 1023>) 521 %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 3, i32 3> 522 %4 = uitofp <4 x i32> %3 to <4 x float> 523 ret <4 x float> %4 524} 525 526define <4 x i32> @knownbits_mask_bitreverse_ashr(<4 x i32> %a0) { 527; X32-LABEL: knownbits_mask_bitreverse_ashr: 528; X32: # %bb.0: 529; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 530; X32-NEXT: retl 531; 532; X64-LABEL: knownbits_mask_bitreverse_ashr: 533; X64: # %bb.0: 534; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 535; X64-NEXT: retq 536 %1 = and <4 x i32> %a0, <i32 -2, i32 -2, i32 -2, i32 -2> 537 %2 = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> %1) 538 %3 = ashr <4 x i32> %2, <i32 31, i32 31, i32 31, i32 31> 539 ret <4 x i32> %3 540} 541declare <4 x i32> @llvm.bitreverse.v4i32(<4 x i32>) nounwind readnone 542 543; If we don't know that the input isn't INT_MIN we can't combine to sitofp 544define <4 x float> @knownbits_abs_uitofp(<4 x i32> %a0) { 545; X32-LABEL: knownbits_abs_uitofp: 546; X32: # %bb.0: 547; X32-NEXT: vpabsd %xmm0, %xmm0 548; X32-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 549; X32-NEXT: vpsrld $16, %xmm0, %xmm0 550; X32-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 551; X32-NEXT: vsubps {{\.LCPI.*}}, %xmm0, %xmm0 552; X32-NEXT: vaddps %xmm0, %xmm1, %xmm0 553; X32-NEXT: retl 554; 555; X64-LABEL: knownbits_abs_uitofp: 556; X64: # %bb.0: 557; X64-NEXT: vpabsd %xmm0, %xmm0 558; X64-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 559; X64-NEXT: vpsrld $16, %xmm0, %xmm0 560; X64-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7] 561; X64-NEXT: vsubps {{.*}}(%rip), %xmm0, %xmm0 562; X64-NEXT: vaddps %xmm0, %xmm1, %xmm0 563; X64-NEXT: retq 564 %1 = sub <4 x i32> zeroinitializer, %a0 565 %2 = icmp slt <4 x i32> %a0, zeroinitializer 566 %3 = select <4 x i1> %2, <4 x i32> %1, <4 x i32> %a0 567 %4 = uitofp <4 x i32> %3 to <4 x float> 568 ret <4 x float> %4 569} 570 571define <4 x float> @knownbits_or_abs_uitofp(<4 x i32> %a0) { 572; X32-LABEL: knownbits_or_abs_uitofp: 573; X32: # %bb.0: 574; X32-NEXT: vpor {{\.LCPI.*}}, %xmm0, %xmm0 575; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,0,2] 576; X32-NEXT: vpabsd %xmm0, %xmm0 577; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 578; X32-NEXT: retl 579; 580; X64-LABEL: knownbits_or_abs_uitofp: 581; X64: # %bb.0: 582; X64-NEXT: vpor {{.*}}(%rip), %xmm0, %xmm0 583; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,0,2] 584; X64-NEXT: vpabsd %xmm0, %xmm0 585; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 586; X64-NEXT: retq 587 %1 = or <4 x i32> %a0, <i32 1, i32 0, i32 3, i32 0> 588 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2> 589 %3 = sub <4 x i32> zeroinitializer, %2 590 %4 = icmp slt <4 x i32> %2, zeroinitializer 591 %5 = select <4 x i1> %4, <4 x i32> %3, <4 x i32> %2 592 %6 = uitofp <4 x i32> %5 to <4 x float> 593 ret <4 x float> %6 594} 595 596define <4 x float> @knownbits_and_select_shuffle_uitofp(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) nounwind { 597; X32-LABEL: knownbits_and_select_shuffle_uitofp: 598; X32: # %bb.0: 599; X32-NEXT: pushl %ebp 600; X32-NEXT: movl %esp, %ebp 601; X32-NEXT: andl $-16, %esp 602; X32-NEXT: subl $16, %esp 603; X32-NEXT: vmovaps 8(%ebp), %xmm3 604; X32-NEXT: vandps {{\.LCPI.*}}, %xmm2, %xmm2 605; X32-NEXT: vandps {{\.LCPI.*}}, %xmm3, %xmm3 606; X32-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 607; X32-NEXT: vblendvps %xmm0, %xmm2, %xmm3, %xmm0 608; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,2,2] 609; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 610; X32-NEXT: movl %ebp, %esp 611; X32-NEXT: popl %ebp 612; X32-NEXT: retl 613; 614; X64-LABEL: knownbits_and_select_shuffle_uitofp: 615; X64: # %bb.0: 616; X64-NEXT: vandps {{.*}}(%rip), %xmm2, %xmm2 617; X64-NEXT: vandps {{.*}}(%rip), %xmm3, %xmm3 618; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 619; X64-NEXT: vblendvps %xmm0, %xmm2, %xmm3, %xmm0 620; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,2,2] 621; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 622; X64-NEXT: retq 623 %1 = and <4 x i32> %a2, <i32 65535, i32 -1, i32 255, i32 -1> 624 %2 = and <4 x i32> %a3, <i32 255, i32 -1, i32 65535, i32 -1> 625 %3 = icmp eq <4 x i32> %a0, %a1 626 %4 = select <4 x i1> %3, <4 x i32> %1, <4 x i32> %2 627 %5 = shufflevector <4 x i32> %4, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2> 628 %6 = uitofp <4 x i32> %5 to <4 x float> 629 ret <4 x float> %6 630} 631 632define <4 x float> @knownbits_lshr_and_select_shuffle_uitofp(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) nounwind { 633; X32-LABEL: knownbits_lshr_and_select_shuffle_uitofp: 634; X32: # %bb.0: 635; X32-NEXT: pushl %ebp 636; X32-NEXT: movl %esp, %ebp 637; X32-NEXT: andl $-16, %esp 638; X32-NEXT: subl $16, %esp 639; X32-NEXT: vmovaps 8(%ebp), %xmm3 640; X32-NEXT: vpsrld $5, %xmm2, %xmm2 641; X32-NEXT: vandps {{\.LCPI.*}}, %xmm3, %xmm3 642; X32-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 643; X32-NEXT: vblendvps %xmm0, %xmm2, %xmm3, %xmm0 644; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,2,2] 645; X32-NEXT: vcvtdq2ps %xmm0, %xmm0 646; X32-NEXT: movl %ebp, %esp 647; X32-NEXT: popl %ebp 648; X32-NEXT: retl 649; 650; X64-LABEL: knownbits_lshr_and_select_shuffle_uitofp: 651; X64: # %bb.0: 652; X64-NEXT: vpsrld $5, %xmm2, %xmm2 653; X64-NEXT: vandps {{.*}}(%rip), %xmm3, %xmm3 654; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 655; X64-NEXT: vblendvps %xmm0, %xmm2, %xmm3, %xmm0 656; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,2,2] 657; X64-NEXT: vcvtdq2ps %xmm0, %xmm0 658; X64-NEXT: retq 659 %1 = lshr <4 x i32> %a2, <i32 5, i32 1, i32 5, i32 1> 660 %2 = and <4 x i32> %a3, <i32 255, i32 -1, i32 65535, i32 -1> 661 %3 = icmp eq <4 x i32> %a0, %a1 662 %4 = select <4 x i1> %3, <4 x i32> %1, <4 x i32> %2 663 %5 = shufflevector <4 x i32> %4, <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2> 664 %6 = uitofp <4 x i32> %5 to <4 x float> 665 ret <4 x float> %6 666} 667 668define <2 x double> @knownbits_lshr_subvector_uitofp(<4 x i32> %x) { 669; X32-LABEL: knownbits_lshr_subvector_uitofp: 670; X32: # %bb.0: 671; X32-NEXT: vpsrld $2, %xmm0, %xmm1 672; X32-NEXT: vpsrld $1, %xmm0, %xmm0 673; X32-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7] 674; X32-NEXT: vcvtdq2pd %xmm0, %xmm0 675; X32-NEXT: retl 676; 677; X64-LABEL: knownbits_lshr_subvector_uitofp: 678; X64: # %bb.0: 679; X64-NEXT: vpsrld $2, %xmm0, %xmm1 680; X64-NEXT: vpsrld $1, %xmm0, %xmm0 681; X64-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7] 682; X64-NEXT: vcvtdq2pd %xmm0, %xmm0 683; X64-NEXT: retq 684 %1 = lshr <4 x i32> %x, <i32 1, i32 2, i32 0, i32 0> 685 %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <2 x i32> <i32 0, i32 1> 686 %3 = uitofp <2 x i32> %2 to <2 x double> 687 ret <2 x double> %3 688} 689