; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt < %s -mattr=+avx512vnni -passes=msan -S | FileCheck %s

; Forked from llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll
;
; Handled strictly: (none)
;
; Handled heuristically: (none)

target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

declare <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32>, <64 x i8>, <64 x i8>)

define <16 x i32> @test_int_x86_avx512_ask_vpdpbusd_512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2) sanitize_memory {
; CHECK-LABEL: define <16 x i32> @test_int_x86_avx512_ask_vpdpbusd_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i8> [[X2:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT:    [[TMP24:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP5:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP23:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <64 x i8> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <64 x i8> [[TMP5]], zeroinitializer
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <64 x i8> [[X2]], zeroinitializer
; CHECK-NEXT:    [[TMP12:%.*]] = and <64 x i1> [[TMP10]], [[TMP11]]
; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i1> [[TMP6]], [[TMP11]]
; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP7]]
; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i1> [[TMP12]], [[TMP13]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i1> [[TMP15]], [[TMP14]]
; CHECK-NEXT:    [[TMP17:%.*]] = sext <64 x i1> [[TMP16]] to <64 x i8>
; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i8> [[TMP17]] to <16 x i32>
; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
; CHECK-NEXT:    [[TMP22:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP22]], [[TMP23]]
; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[X2]])
; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
;
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2)
  ret <16 x i32> %1
}

define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusd_512(<16 x i32> %x0, <64 x i8> %x1, ptr %x2p, <64 x i8> %x4, i16 %x3) sanitize_memory {
; CHECK-LABEL: define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusd_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <64 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <64 x i8> [[X4:%.*]], i16 [[X3:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP33:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 200), align 8
; CHECK-NEXT:    [[TMP40:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1:![0-9]+]]
; CHECK:       [[BB6]]:
; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR4:[0-9]+]]
; CHECK-NEXT:    unreachable
; CHECK:       [[BB7]]:
; CHECK-NEXT:    [[TMP31:%.*]] = load <64 x i8>, ptr [[X2P]], align 64
; CHECK-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64
; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
; CHECK-NEXT:    [[TMP30:%.*]] = load <64 x i8>, ptr [[TMP10]], align 64
; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <64 x i8> [[TMP33]], zeroinitializer
; CHECK-NEXT:    [[TMP36:%.*]] = icmp ne <64 x i8> [[TMP30]], zeroinitializer
; CHECK-NEXT:    [[TMP32:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP37:%.*]] = icmp ne <64 x i8> [[TMP31]], zeroinitializer
; CHECK-NEXT:    [[TMP38:%.*]] = and <64 x i1> [[TMP35]], [[TMP36]]
; CHECK-NEXT:    [[TMP58:%.*]] = and <64 x i1> [[TMP32]], [[TMP36]]
; CHECK-NEXT:    [[TMP59:%.*]] = and <64 x i1> [[TMP35]], [[TMP37]]
; CHECK-NEXT:    [[TMP60:%.*]] = or <64 x i1> [[TMP38]], [[TMP58]]
; CHECK-NEXT:    [[TMP61:%.*]] = or <64 x i1> [[TMP60]], [[TMP59]]
; CHECK-NEXT:    [[TMP62:%.*]] = sext <64 x i1> [[TMP61]] to <64 x i8>
; CHECK-NEXT:    [[TMP53:%.*]] = bitcast <64 x i8> [[TMP62]] to <16 x i32>
; CHECK-NEXT:    [[TMP54:%.*]] = icmp ne <16 x i32> [[TMP53]], zeroinitializer
; CHECK-NEXT:    [[TMP29:%.*]] = sext <16 x i1> [[TMP54]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP29]], [[TMP2]]
; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[TMP31]])
; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
; CHECK-NEXT:    [[TMP15:%.*]] = xor <16 x i32> [[TMP11]], [[X0]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <16 x i32> [[TMP15]], [[_MSPROP1]]
; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
; CHECK-NEXT:    [[TMP39:%.*]] = icmp ne <64 x i8> [[TMP33]], zeroinitializer
; CHECK-NEXT:    [[TMP46:%.*]] = icmp ne <64 x i8> [[TMP40]], zeroinitializer
; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <64 x i8> [[X4]], zeroinitializer
; CHECK-NEXT:    [[TMP47:%.*]] = and <64 x i1> [[TMP39]], [[TMP46]]
; CHECK-NEXT:    [[TMP48:%.*]] = and <64 x i1> [[TMP41]], [[TMP46]]
; CHECK-NEXT:    [[TMP49:%.*]] = and <64 x i1> [[TMP39]], [[TMP42]]
; CHECK-NEXT:    [[TMP50:%.*]] = or <64 x i1> [[TMP47]], [[TMP48]]
; CHECK-NEXT:    [[TMP51:%.*]] = or <64 x i1> [[TMP50]], [[TMP49]]
; CHECK-NEXT:    [[TMP52:%.*]] = sext <64 x i1> [[TMP51]] to <64 x i8>
; CHECK-NEXT:    [[TMP55:%.*]] = bitcast <64 x i8> [[TMP52]] to <16 x i32>
; CHECK-NEXT:    [[TMP56:%.*]] = icmp ne <16 x i32> [[TMP55]], zeroinitializer
; CHECK-NEXT:    [[TMP57:%.*]] = sext <16 x i1> [[TMP56]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP57]], [[TMP2]]
; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[X4]])
; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP23:%.*]] = xor <16 x i32> [[TMP19]], zeroinitializer
; CHECK-NEXT:    [[TMP24:%.*]] = or <16 x i32> [[TMP23]], [[_MSPROP3]]
; CHECK-NEXT:    [[TMP25:%.*]] = or <16 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[_MSPROP_SELECT4:%.*]] = select <16 x i1> [[TMP20]], <16 x i32> [[TMP25]], <16 x i32> [[TMP22]]
; CHECK-NEXT:    [[TMP26:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[TMP19]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <16 x i32>, <16 x i32> } { <16 x i32> splat (i32 -1), <16 x i32> splat (i32 -1) }, <16 x i32> [[_MSPROP_SELECT]], 0
; CHECK-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> [[TMP18]], 0
; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[TMP27]], <16 x i32> [[_MSPROP_SELECT4]], 1
; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES1]], <16 x i32> [[TMP26]], 1
; CHECK-NEXT:    store { <16 x i32>, <16 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES2]]
;
  %x2 = load <64 x i8>, ptr %x2p
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
  %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x4)
  %5 = bitcast i16 %x3 to <16 x i1>
  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
  ret { <16 x i32>, <16 x i32> } %res2
}

declare <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32>, <64 x i8>, <64 x i8>)

define <16 x i32>@test_int_x86_avx512_vpdpbusds_512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2) sanitize_memory {
; CHECK-LABEL: define <16 x i32> @test_int_x86_avx512_vpdpbusds_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i8> [[X2:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP24:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP5:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP23:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <64 x i8> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <64 x i8> [[TMP5]], zeroinitializer
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <64 x i8> [[X2]], zeroinitializer
; CHECK-NEXT:    [[TMP12:%.*]] = and <64 x i1> [[TMP10]], [[TMP11]]
; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i1> [[TMP6]], [[TMP11]]
; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP7]]
; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i1> [[TMP12]], [[TMP13]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i1> [[TMP15]], [[TMP14]]
; CHECK-NEXT:    [[TMP17:%.*]] = sext <64 x i1> [[TMP16]] to <64 x i8>
; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i8> [[TMP17]] to <16 x i32>
; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
; CHECK-NEXT:    [[TMP22:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP22]], [[TMP23]]
; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[X2]])
; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
;
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2)
  ret <16 x i32> %1
}

define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusds_512(<16 x i32> %x0, <64 x i8> %x1, ptr %x2p, <64 x i8> %x4, i16 %x3) sanitize_memory {
; CHECK-LABEL: define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusds_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <64 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <64 x i8> [[X4:%.*]], i16 [[X3:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP33:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 200), align 8
; CHECK-NEXT:    [[TMP40:%.*]] = load <64 x i8>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]
; CHECK:       [[BB6]]:
; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR4]]
; CHECK-NEXT:    unreachable
; CHECK:       [[BB7]]:
; CHECK-NEXT:    [[TMP31:%.*]] = load <64 x i8>, ptr [[X2P]], align 64
; CHECK-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64
; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
; CHECK-NEXT:    [[TMP30:%.*]] = load <64 x i8>, ptr [[TMP10]], align 64
; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <64 x i8> [[TMP33]], zeroinitializer
; CHECK-NEXT:    [[TMP36:%.*]] = icmp ne <64 x i8> [[TMP30]], zeroinitializer
; CHECK-NEXT:    [[TMP32:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP37:%.*]] = icmp ne <64 x i8> [[TMP31]], zeroinitializer
; CHECK-NEXT:    [[TMP38:%.*]] = and <64 x i1> [[TMP35]], [[TMP36]]
; CHECK-NEXT:    [[TMP58:%.*]] = and <64 x i1> [[TMP32]], [[TMP36]]
; CHECK-NEXT:    [[TMP59:%.*]] = and <64 x i1> [[TMP35]], [[TMP37]]
; CHECK-NEXT:    [[TMP60:%.*]] = or <64 x i1> [[TMP38]], [[TMP58]]
; CHECK-NEXT:    [[TMP61:%.*]] = or <64 x i1> [[TMP60]], [[TMP59]]
; CHECK-NEXT:    [[TMP62:%.*]] = sext <64 x i1> [[TMP61]] to <64 x i8>
; CHECK-NEXT:    [[TMP53:%.*]] = bitcast <64 x i8> [[TMP62]] to <16 x i32>
; CHECK-NEXT:    [[TMP54:%.*]] = icmp ne <16 x i32> [[TMP53]], zeroinitializer
; CHECK-NEXT:    [[TMP29:%.*]] = sext <16 x i1> [[TMP54]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP29]], [[TMP2]]
; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[TMP31]])
; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
; CHECK-NEXT:    [[TMP15:%.*]] = xor <16 x i32> [[TMP11]], [[X0]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <16 x i32> [[TMP15]], [[_MSPROP1]]
; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
; CHECK-NEXT:    [[TMP39:%.*]] = icmp ne <64 x i8> [[TMP33]], zeroinitializer
; CHECK-NEXT:    [[TMP46:%.*]] = icmp ne <64 x i8> [[TMP40]], zeroinitializer
; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <64 x i8> [[X1]], zeroinitializer
; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <64 x i8> [[X4]], zeroinitializer
; CHECK-NEXT:    [[TMP47:%.*]] = and <64 x i1> [[TMP39]], [[TMP46]]
; CHECK-NEXT:    [[TMP48:%.*]] = and <64 x i1> [[TMP41]], [[TMP46]]
; CHECK-NEXT:    [[TMP49:%.*]] = and <64 x i1> [[TMP39]], [[TMP42]]
; CHECK-NEXT:    [[TMP50:%.*]] = or <64 x i1> [[TMP47]], [[TMP48]]
; CHECK-NEXT:    [[TMP51:%.*]] = or <64 x i1> [[TMP50]], [[TMP49]]
; CHECK-NEXT:    [[TMP52:%.*]] = sext <64 x i1> [[TMP51]] to <64 x i8>
; CHECK-NEXT:    [[TMP55:%.*]] = bitcast <64 x i8> [[TMP52]] to <16 x i32>
; CHECK-NEXT:    [[TMP56:%.*]] = icmp ne <16 x i32> [[TMP55]], zeroinitializer
; CHECK-NEXT:    [[TMP57:%.*]] = sext <16 x i1> [[TMP56]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP57]], [[TMP2]]
; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> [[X0]], <64 x i8> [[X1]], <64 x i8> [[X4]])
; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP23:%.*]] = xor <16 x i32> [[TMP19]], zeroinitializer
; CHECK-NEXT:    [[TMP24:%.*]] = or <16 x i32> [[TMP23]], [[_MSPROP3]]
; CHECK-NEXT:    [[TMP25:%.*]] = or <16 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[_MSPROP_SELECT4:%.*]] = select <16 x i1> [[TMP20]], <16 x i32> [[TMP25]], <16 x i32> [[TMP22]]
; CHECK-NEXT:    [[TMP26:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[TMP19]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <16 x i32>, <16 x i32> } { <16 x i32> splat (i32 -1), <16 x i32> splat (i32 -1) }, <16 x i32> [[_MSPROP_SELECT]], 0
; CHECK-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> [[TMP18]], 0
; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[TMP27]], <16 x i32> [[_MSPROP_SELECT4]], 1
; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES1]], <16 x i32> [[TMP26]], 1
; CHECK-NEXT:    store { <16 x i32>, <16 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES2]]
;
  %x2 = load <64 x i8>, ptr %x2p
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
  %4 = call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %x0, <64 x i8> %x1, <64 x i8> %x4)
  %5 = bitcast i16 %x3 to <16 x i1>
  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
  ret { <16 x i32>, <16 x i32> } %res2
}

declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>)

define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) sanitize_memory {
; CHECK-LABEL: define <16 x i32> @test_int_x86_avx512_vpdpwssd_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <16 x i32> [[X1:%.*]], <16 x i32> [[X2:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[TMP5]], zeroinitializer
; CHECK-NEXT:    [[TMP12:%.*]] = and <32 x i1> [[TMP8]], [[TMP9]]
; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[TMP10]], [[TMP9]]
; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP8]], [[TMP11]]
; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i1> [[TMP12]], [[TMP13]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP15]], [[TMP14]]
; CHECK-NEXT:    [[TMP17:%.*]] = sext <32 x i1> [[TMP16]] to <32 x i16>
; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <32 x i16> [[TMP17]] to <16 x i32>
; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
;
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
  ret <16 x i32> %1
}

define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) sanitize_memory {
; CHECK-LABEL: define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <16 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i32> [[X4:%.*]], i16 [[X3:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 200), align 8
; CHECK-NEXT:    [[TMP5:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]
; CHECK:       [[BB6]]:
; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR4]]
; CHECK-NEXT:    unreachable
; CHECK:       [[BB7]]:
; CHECK-NEXT:    [[X2:%.*]] = load <16 x i32>, ptr [[X2P]], align 64
; CHECK-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64
; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
; CHECK-NEXT:    [[TMP36:%.*]] = icmp ne <32 x i16> [[TMP30]], zeroinitializer
; CHECK-NEXT:    [[TMP54:%.*]] = and <32 x i1> [[TMP33]], [[TMP34]]
; CHECK-NEXT:    [[TMP55:%.*]] = and <32 x i1> [[TMP35]], [[TMP34]]
; CHECK-NEXT:    [[TMP56:%.*]] = and <32 x i1> [[TMP33]], [[TMP36]]
; CHECK-NEXT:    [[TMP57:%.*]] = or <32 x i1> [[TMP54]], [[TMP55]]
; CHECK-NEXT:    [[TMP58:%.*]] = or <32 x i1> [[TMP57]], [[TMP56]]
; CHECK-NEXT:    [[TMP59:%.*]] = sext <32 x i1> [[TMP58]] to <32 x i16>
; CHECK-NEXT:    [[TMP60:%.*]] = bitcast <32 x i16> [[TMP59]] to <16 x i32>
; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
; CHECK-NEXT:    [[TMP15:%.*]] = xor <16 x i32> [[TMP11]], [[X0]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <16 x i32> [[TMP15]], [[_MSPROP1]]
; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
; CHECK-NEXT:    [[TMP44:%.*]] = icmp ne <32 x i16> [[TMP38]], zeroinitializer
; CHECK-NEXT:    [[TMP45:%.*]] = and <32 x i1> [[TMP41]], [[TMP42]]
; CHECK-NEXT:    [[TMP46:%.*]] = and <32 x i1> [[TMP43]], [[TMP42]]
; CHECK-NEXT:    [[TMP47:%.*]] = and <32 x i1> [[TMP41]], [[TMP44]]
; CHECK-NEXT:    [[TMP48:%.*]] = or <32 x i1> [[TMP45]], [[TMP46]]
; CHECK-NEXT:    [[TMP49:%.*]] = or <32 x i1> [[TMP48]], [[TMP47]]
; CHECK-NEXT:    [[TMP50:%.*]] = sext <32 x i1> [[TMP49]] to <32 x i16>
; CHECK-NEXT:    [[TMP51:%.*]] = bitcast <32 x i16> [[TMP50]] to <16 x i32>
; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP23:%.*]] = xor <16 x i32> [[TMP19]], zeroinitializer
; CHECK-NEXT:    [[TMP24:%.*]] = or <16 x i32> [[TMP23]], [[_MSPROP3]]
; CHECK-NEXT:    [[TMP25:%.*]] = or <16 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[_MSPROP_SELECT4:%.*]] = select <16 x i1> [[TMP20]], <16 x i32> [[TMP25]], <16 x i32> [[TMP22]]
; CHECK-NEXT:    [[TMP26:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[TMP19]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <16 x i32>, <16 x i32> } { <16 x i32> splat (i32 -1), <16 x i32> splat (i32 -1) }, <16 x i32> [[_MSPROP_SELECT]], 0
; CHECK-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> [[TMP18]], 0
; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[TMP27]], <16 x i32> [[_MSPROP_SELECT4]], 1
; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES1]], <16 x i32> [[TMP26]], 1
; CHECK-NEXT:    store { <16 x i32>, <16 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES2]]
;
  %x2 = load <16 x i32>, ptr %x2p
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
  %5 = bitcast i16 %x3 to <16 x i1>
  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
  ret { <16 x i32>, <16 x i32> } %res2
}

declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>)

define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) sanitize_memory {
; CHECK-LABEL: define <16 x i32> @test_int_x86_avx512_ask_vpdpwssds_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <16 x i32> [[X1:%.*]], <16 x i32> [[X2:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[TMP5]], zeroinitializer
; CHECK-NEXT:    [[TMP12:%.*]] = and <32 x i1> [[TMP8]], [[TMP9]]
; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[TMP10]], [[TMP9]]
; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP8]], [[TMP11]]
; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i1> [[TMP12]], [[TMP13]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP15]], [[TMP14]]
; CHECK-NEXT:    [[TMP17:%.*]] = sext <32 x i1> [[TMP16]] to <32 x i16>
; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <32 x i16> [[TMP17]] to <16 x i32>
; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
;
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
  ret <16 x i32> %1
}

define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) sanitize_memory {
; CHECK-LABEL: define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(
; CHECK-SAME: <16 x i32> [[X0:%.*]], <16 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i32> [[X4:%.*]], i16 [[X3:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 200), align 8
; CHECK-NEXT:    [[TMP5:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB6:.*]], label %[[BB7:.*]], !prof [[PROF1]]
; CHECK:       [[BB6]]:
; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR4]]
; CHECK-NEXT:    unreachable
; CHECK:       [[BB7]]:
; CHECK-NEXT:    [[X2:%.*]] = load <16 x i32>, ptr [[X2P]], align 64
; CHECK-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[X2P]] to i64
; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
; CHECK-NEXT:    [[TMP36:%.*]] = icmp ne <32 x i16> [[TMP30]], zeroinitializer
; CHECK-NEXT:    [[TMP54:%.*]] = and <32 x i1> [[TMP33]], [[TMP34]]
; CHECK-NEXT:    [[TMP55:%.*]] = and <32 x i1> [[TMP35]], [[TMP34]]
; CHECK-NEXT:    [[TMP56:%.*]] = and <32 x i1> [[TMP33]], [[TMP36]]
; CHECK-NEXT:    [[TMP57:%.*]] = or <32 x i1> [[TMP54]], [[TMP55]]
; CHECK-NEXT:    [[TMP58:%.*]] = or <32 x i1> [[TMP57]], [[TMP56]]
; CHECK-NEXT:    [[TMP59:%.*]] = sext <32 x i1> [[TMP58]] to <32 x i16>
; CHECK-NEXT:    [[TMP60:%.*]] = bitcast <32 x i16> [[TMP59]] to <16 x i32>
; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
; CHECK-NEXT:    [[TMP15:%.*]] = xor <16 x i32> [[TMP11]], [[X0]]
; CHECK-NEXT:    [[TMP16:%.*]] = or <16 x i32> [[TMP15]], [[_MSPROP1]]
; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
; CHECK-NEXT:    [[TMP44:%.*]] = icmp ne <32 x i16> [[TMP38]], zeroinitializer
; CHECK-NEXT:    [[TMP45:%.*]] = and <32 x i1> [[TMP41]], [[TMP42]]
; CHECK-NEXT:    [[TMP46:%.*]] = and <32 x i1> [[TMP43]], [[TMP42]]
; CHECK-NEXT:    [[TMP47:%.*]] = and <32 x i1> [[TMP41]], [[TMP44]]
; CHECK-NEXT:    [[TMP48:%.*]] = or <32 x i1> [[TMP45]], [[TMP46]]
; CHECK-NEXT:    [[TMP49:%.*]] = or <32 x i1> [[TMP48]], [[TMP47]]
; CHECK-NEXT:    [[TMP50:%.*]] = sext <32 x i1> [[TMP49]] to <32 x i16>
; CHECK-NEXT:    [[TMP51:%.*]] = bitcast <32 x i16> [[TMP50]] to <16 x i32>
; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP23:%.*]] = xor <16 x i32> [[TMP19]], zeroinitializer
; CHECK-NEXT:    [[TMP24:%.*]] = or <16 x i32> [[TMP23]], [[_MSPROP3]]
; CHECK-NEXT:    [[TMP25:%.*]] = or <16 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT:    [[_MSPROP_SELECT4:%.*]] = select <16 x i1> [[TMP20]], <16 x i32> [[TMP25]], <16 x i32> [[TMP22]]
; CHECK-NEXT:    [[TMP26:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[TMP19]], <16 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <16 x i32>, <16 x i32> } { <16 x i32> splat (i32 -1), <16 x i32> splat (i32 -1) }, <16 x i32> [[_MSPROP_SELECT]], 0
; CHECK-NEXT:    [[RES1:%.*]] = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> [[TMP18]], 0
; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[TMP27]], <16 x i32> [[_MSPROP_SELECT4]], 1
; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <16 x i32>, <16 x i32> } [[RES1]], <16 x i32> [[TMP26]], 1
; CHECK-NEXT:    store { <16 x i32>, <16 x i32> } [[TMP28]], ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret { <16 x i32>, <16 x i32> } [[RES2]]
;
  %x2 = load <16 x i32>, ptr %x2p
  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
  %5 = bitcast i16 %x3 to <16 x i1>
  %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
  %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
  %res2 = insertvalue { <16 x i32>, <16 x i32> }  %res1, <16 x i32> %6, 1
  ret { <16 x i32>, <16 x i32> } %res2
}
;.
; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 1048575}
;.