1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512ifma | FileCheck %s 2*9880d681SAndroid Build Coastguard Worker 3*9880d681SAndroid Build Coastguard Workerdeclare <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64>, <8 x i64>, <8 x i64>, i8) 4*9880d681SAndroid Build Coastguard Worker 5*9880d681SAndroid Build Coastguard Workerdefine <8 x i64>@test_int_x86_avx512_mask_vpmadd52h_uq_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) { 6*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: test_int_x86_avx512_mask_vpmadd52h_uq_512: 7*9880d681SAndroid Build Coastguard Worker; CHECK: kmovw %edi, %k1 8*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm3 9*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm3 {%k1} 10*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm4 11*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm4 12*9880d681SAndroid Build Coastguard Worker; CHECK: vpxord %zmm2, %zmm2, %zmm2 13*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm0 {%k1} 14*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm2 {%k1} {z} 15*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm3, %zmm0 16*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm2, %zmm4, %zmm1 17*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm1, %zmm0 18*9880d681SAndroid Build Coastguard Worker 19*9880d681SAndroid Build Coastguard Worker %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) 20*9880d681SAndroid Build Coastguard Worker %res1 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 21*9880d681SAndroid Build Coastguard Worker %res2 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> zeroinitializer, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 22*9880d681SAndroid Build Coastguard Worker %res3 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1) 23*9880d681SAndroid Build Coastguard Worker %res4 = add <8 x i64> %res, %res1 24*9880d681SAndroid Build Coastguard Worker %res5 = add <8 x i64> %res3, %res2 25*9880d681SAndroid Build Coastguard Worker %res6 = add <8 x i64> %res5, %res4 26*9880d681SAndroid Build Coastguard Worker ret <8 x i64> %res6 27*9880d681SAndroid Build Coastguard Worker} 28*9880d681SAndroid Build Coastguard Worker 29*9880d681SAndroid Build Coastguard Workerdeclare <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64>, <8 x i64>, <8 x i64>, i8) 30*9880d681SAndroid Build Coastguard Worker 31*9880d681SAndroid Build Coastguard Workerdefine <8 x i64>@test_int_x86_avx512_maskz_vpmadd52h_uq_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) { 32*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: test_int_x86_avx512_maskz_vpmadd52h_uq_512: 33*9880d681SAndroid Build Coastguard Worker; CHECK: kmovw %edi, %k1 34*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm3 35*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm3 {%k1} {z} 36*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm4 37*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm4 38*9880d681SAndroid Build Coastguard Worker; CHECK: vpxord %zmm2, %zmm2, %zmm2 39*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm0 {%k1} {z} 40*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52huq %zmm2, %zmm1, %zmm2 {%k1} {z} 41*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm3, %zmm0 42*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm2, %zmm4, %zmm1 43*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm1, %zmm0 44*9880d681SAndroid Build Coastguard Worker 45*9880d681SAndroid Build Coastguard Worker %res = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) 46*9880d681SAndroid Build Coastguard Worker %res1 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 47*9880d681SAndroid Build Coastguard Worker %res2 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> zeroinitializer, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 48*9880d681SAndroid Build Coastguard Worker %res3 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52h.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1) 49*9880d681SAndroid Build Coastguard Worker %res4 = add <8 x i64> %res, %res1 50*9880d681SAndroid Build Coastguard Worker %res5 = add <8 x i64> %res3, %res2 51*9880d681SAndroid Build Coastguard Worker %res6 = add <8 x i64> %res5, %res4 52*9880d681SAndroid Build Coastguard Worker ret <8 x i64> %res6 53*9880d681SAndroid Build Coastguard Worker} 54*9880d681SAndroid Build Coastguard Worker 55*9880d681SAndroid Build Coastguard Workerdeclare <8 x i64> @llvm.x86.avx512.mask.vpmadd52l.uq.512(<8 x i64>, <8 x i64>, <8 x i64>, i8) 56*9880d681SAndroid Build Coastguard Worker 57*9880d681SAndroid Build Coastguard Workerdefine <8 x i64>@test_int_x86_avx512_mask_vpmadd52l_uq_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) { 58*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: test_int_x86_avx512_mask_vpmadd52l_uq_512: 59*9880d681SAndroid Build Coastguard Worker; CHECK: kmovw %edi, %k1 60*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm3 61*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm3 {%k1} 62*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm4 63*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm4 64*9880d681SAndroid Build Coastguard Worker; CHECK: vpxord %zmm2, %zmm2, %zmm2 65*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm0 {%k1} 66*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm2 {%k1} {z} 67*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm3, %zmm0 68*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm2, %zmm4, %zmm1 69*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm1, %zmm0 70*9880d681SAndroid Build Coastguard Worker 71*9880d681SAndroid Build Coastguard Worker %res = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) 72*9880d681SAndroid Build Coastguard Worker %res1 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 73*9880d681SAndroid Build Coastguard Worker %res2 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52l.uq.512(<8 x i64> zeroinitializer, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 74*9880d681SAndroid Build Coastguard Worker %res3 = call <8 x i64> @llvm.x86.avx512.mask.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1) 75*9880d681SAndroid Build Coastguard Worker %res4 = add <8 x i64> %res, %res1 76*9880d681SAndroid Build Coastguard Worker %res5 = add <8 x i64> %res3, %res2 77*9880d681SAndroid Build Coastguard Worker %res6 = add <8 x i64> %res5, %res4 78*9880d681SAndroid Build Coastguard Worker ret <8 x i64> %res6 79*9880d681SAndroid Build Coastguard Worker} 80*9880d681SAndroid Build Coastguard Worker 81*9880d681SAndroid Build Coastguard Workerdeclare <8 x i64> @llvm.x86.avx512.maskz.vpmadd52l.uq.512(<8 x i64>, <8 x i64>, <8 x i64>, i8) 82*9880d681SAndroid Build Coastguard Worker 83*9880d681SAndroid Build Coastguard Workerdefine <8 x i64>@test_int_x86_avx512_maskz_vpmadd52l_uq_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) { 84*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: test_int_x86_avx512_maskz_vpmadd52l_uq_512: 85*9880d681SAndroid Build Coastguard Worker; CHECK: kmovw %edi, %k1 86*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm3 87*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm3 {%k1} {z} 88*9880d681SAndroid Build Coastguard Worker; CHECK: vmovaps %zmm0, %zmm4 89*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm4 90*9880d681SAndroid Build Coastguard Worker; CHECK: vpxord %zmm2, %zmm2, %zmm2 91*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm0 {%k1} {z} 92*9880d681SAndroid Build Coastguard Worker; CHECK: vpmadd52luq %zmm2, %zmm1, %zmm2 {%k1} {z} 93*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm3, %zmm0 94*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm2, %zmm4, %zmm1 95*9880d681SAndroid Build Coastguard Worker; CHECK: vpaddq %zmm0, %zmm1, %zmm0 96*9880d681SAndroid Build Coastguard Worker 97*9880d681SAndroid Build Coastguard Worker %res = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) 98*9880d681SAndroid Build Coastguard Worker %res1 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 99*9880d681SAndroid Build Coastguard Worker %res2 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52l.uq.512(<8 x i64> zeroinitializer, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3) 100*9880d681SAndroid Build Coastguard Worker %res3 = call <8 x i64> @llvm.x86.avx512.maskz.vpmadd52l.uq.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1) 101*9880d681SAndroid Build Coastguard Worker %res4 = add <8 x i64> %res, %res1 102*9880d681SAndroid Build Coastguard Worker %res5 = add <8 x i64> %res3, %res2 103*9880d681SAndroid Build Coastguard Worker %res6 = add <8 x i64> %res5, %res4 104*9880d681SAndroid Build Coastguard Worker ret <8 x i64> %res6 105*9880d681SAndroid Build Coastguard Worker} 106