110 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512ifma,+avx512vl | FileCheck %s --check-prefixes=AVX512VL4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avxifma | FileCheck %s --check-prefixes=AVXIFMA5 6; High-52 path7 8declare <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(<2 x i64>, <2 x i64>, <2 x i64>)9declare <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(<4 x i64>, <4 x i64>, <4 x i64>)10 11; High-52, 25x25 masked inputs, accumulator = 1, expected constant fold.12define <2 x i64> @kb52h_128_mask25_and1(<2 x i64> %x, <2 x i64> %y) {13; AVX512VL-LABEL: kb52h_128_mask25_and1:14; AVX512VL: # %bb.0:15; AVX512VL-NEXT: vmovddup {{.*#+}} xmm0 = [1,1]16; AVX512VL-NEXT: # xmm0 = mem[0,0]17; AVX512VL-NEXT: retq18;19; AVXIFMA-LABEL: kb52h_128_mask25_and1:20; AVXIFMA: # %bb.0:21; AVXIFMA-NEXT: vmovddup {{.*#+}} xmm0 = [1,1]22; AVXIFMA-NEXT: # xmm0 = mem[0,0]23; AVXIFMA-NEXT: retq24 %mx = and <2 x i64> %x, splat (i64 33554431) ; (1<<25)-125 %my = and <2 x i64> %y, splat (i64 33554431) ; (1<<25)-126 %r = call <2 x i64> @llvm.x86.avx512.vpmadd52h.uq.128(27 <2 x i64> splat (i64 1),28 <2 x i64> %mx,29 <2 x i64> %my)30 %ret = and <2 x i64> %r, splat (i64 1)31 ret <2 x i64> %ret32}33 34; High-52, 25x26 masked inputs, accumulator = 1, expected constant fold.35define <4 x i64> @kb52h_256_mask25x26_acc1(<4 x i64> %x, <4 x i64> %y) {36; AVX512VL-LABEL: kb52h_256_mask25x26_acc1:37; AVX512VL: # %bb.0:38; AVX512VL-NEXT: vbroadcastsd {{.*#+}} ymm0 = [1,1,1,1]39; AVX512VL-NEXT: retq40;41; AVXIFMA-LABEL: kb52h_256_mask25x26_acc1:42; AVXIFMA: # %bb.0:43; AVXIFMA-NEXT: vbroadcastsd {{.*#+}} ymm0 = [1,1,1,1]44; AVXIFMA-NEXT: retq45 %mx = and <4 x i64> %x, splat (i64 33554431) ; (1<<25)-146 %my = and <4 x i64> %y, splat (i64 67108863) ; (1<<26)-147 %r = call <4 x i64> @llvm.x86.avx512.vpmadd52h.uq.256(48 <4 x i64> splat (i64 1),49 <4 x i64> %mx, <4 x i64> %my)50 ret <4 x i64> %r51}52 53; Low-52 path54 55declare <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(<2 x i64>, <2 x i64>, <2 x i64>)56declare <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(<4 x i64>, <4 x i64>, <4 x i64>)57 58; Low-52, 26x26 masked inputs, add with accumulator.59define <2 x i64> @kb52l_128_mask26x26_add_intrin(<2 x i64> %x, <2 x i64> %y, <2 x i64> %acc) {60; AVX512VL-LABEL: kb52l_128_mask26x26_add_intrin:61; AVX512VL: # %bb.0:62; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [67108863,67108863]63; AVX512VL-NEXT: vpand %xmm3, %xmm0, %xmm064; AVX512VL-NEXT: vpand %xmm3, %xmm1, %xmm165; AVX512VL-NEXT: vpmadd52luq %xmm1, %xmm0, %xmm266; AVX512VL-NEXT: vmovdqa %xmm2, %xmm067; AVX512VL-NEXT: retq68;69; AVXIFMA-LABEL: kb52l_128_mask26x26_add_intrin:70; AVXIFMA: # %bb.0:71; AVXIFMA-NEXT: vpbroadcastq {{.*#+}} xmm3 = [67108863,67108863]72; AVXIFMA-NEXT: vpand %xmm3, %xmm0, %xmm073; AVXIFMA-NEXT: vpand %xmm3, %xmm1, %xmm174; AVXIFMA-NEXT: {vex} vpmadd52luq %xmm1, %xmm0, %xmm275; AVXIFMA-NEXT: vmovdqa %xmm2, %xmm076; AVXIFMA-NEXT: retq77 %xm = and <2 x i64> %x, splat (i64 67108863) ; (1<<26)-178 %ym = and <2 x i64> %y, splat (i64 67108863) ; (1<<26)-179 %r = call <2 x i64> @llvm.x86.avx512.vpmadd52l.uq.128(80 <2 x i64> %acc, <2 x i64> %xm, <2 x i64> %ym)81 ret <2 x i64> %r82}83 84; Low-52, 50-bit × 2-bit masked inputs, add with accumulator.85define <4 x i64> @kb52l_256_mask50x3_add_intrin(<4 x i64> %x, <4 x i64> %y, <4 x i64> %acc) {86; AVX512VL-LABEL: kb52l_256_mask50x3_add_intrin:87; AVX512VL: # %bb.0:88; AVX512VL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm089; AVX512VL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm190; AVX512VL-NEXT: vpmadd52luq %ymm1, %ymm0, %ymm291; AVX512VL-NEXT: vmovdqa %ymm2, %ymm092; AVX512VL-NEXT: retq93;94; AVXIFMA-LABEL: kb52l_256_mask50x3_add_intrin:95; AVXIFMA: # %bb.0:96; AVXIFMA-NEXT: vpbroadcastq {{.*#+}} ymm3 = [1125899906842623,1125899906842623,1125899906842623,1125899906842623]97; AVXIFMA-NEXT: vpand %ymm3, %ymm0, %ymm098; AVXIFMA-NEXT: vpbroadcastq {{.*#+}} ymm3 = [3,3,3,3]99; AVXIFMA-NEXT: vpand %ymm3, %ymm1, %ymm1100; AVXIFMA-NEXT: {vex} vpmadd52luq %ymm1, %ymm0, %ymm2101; AVXIFMA-NEXT: vmovdqa %ymm2, %ymm0102; AVXIFMA-NEXT: retq103 %xm = and <4 x i64> %x, splat (i64 1125899906842623) ; (1<<50)-1104 %ym = and <4 x i64> %y, splat (i64 3) ; (1<<2)-1105 %r = call <4 x i64> @llvm.x86.avx512.vpmadd52l.uq.256(106 <4 x i64> %acc, <4 x i64> %xm, <4 x i64> %ym)107 ret <4 x i64> %r108}109 110