; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=lower-matrix-intrinsics,instcombine -fuse-matrix-use-loops=false -fuse-matrix-tile-size=1 -matrix-allow-contract -force-fuse-matrix -verify-dom-info %s -S | FileCheck %s

; REQUIRES: aarch64-registered-target

target datalayout = "e-m:o-i64:64-f80:128-n8:4:32:64-S128"
target triple = "aarch64-apple-ios"

define void @multiply_can_hoist_cast(<4 x double>* noalias %A, <4 x double> * %B, [4 x double]* %C) {
; CHECK-LABEL: @multiply_can_hoist_cast(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[STORE_BEGIN:%.*]] = ptrtoint [4 x double]* [[C:%.*]] to i64
; CHECK-NEXT:    [[STORE_END:%.*]] = add nuw nsw i64 [[STORE_BEGIN]], 32
; CHECK-NEXT:    [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK:       alias_cont:
; CHECK-NEXT:    [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT:    [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK:       copy:
; CHECK-NEXT:    [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT:    br label [[NO_ALIAS]]
; CHECK:       no_alias:
; CHECK-NEXT:    [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT:    [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST5:%.*]] = bitcast double* [[TMP7]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD6:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST5]], align 8
; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST8:%.*]] = bitcast double* [[TMP8]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD9:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST8]], align 8
; CHECK-NEXT:    [[TMP9:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD6]], <1 x double> [[COL_LOAD9]], <1 x double> [[TMP6]])
; CHECK-NEXT:    [[VEC_CAST15:%.*]] = bitcast [4 x double]* [[C]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP9]], <1 x double>* [[VEC_CAST15]], align 8
; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[TMP10]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD18:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_CAST20:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD21:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST20]], align 8
; CHECK-NEXT:    [[TMP11:%.*]] = fmul contract <1 x double> [[COL_LOAD18]], [[COL_LOAD21]]
; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST26:%.*]] = bitcast double* [[TMP12]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD27:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST26]], align 8
; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST29:%.*]] = bitcast double* [[TMP13]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD30:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST29]], align 8
; CHECK-NEXT:    [[TMP14:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD27]], <1 x double> [[COL_LOAD30]], <1 x double> [[TMP11]])
; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST36:%.*]] = bitcast double* [[TMP15]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP14]], <1 x double>* [[VEC_CAST36]], align 8
; CHECK-NEXT:    [[VEC_CAST38:%.*]] = bitcast <4 x double>* [[A]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD39:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST38]], align 8
; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST41:%.*]] = bitcast double* [[TMP16]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD42:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST41]], align 8
; CHECK-NEXT:    [[TMP17:%.*]] = fmul contract <1 x double> [[COL_LOAD39]], [[COL_LOAD42]]
; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST47:%.*]] = bitcast double* [[TMP18]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD48:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST47]], align 8
; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST50:%.*]] = bitcast double* [[TMP19]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD51:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST50]], align 8
; CHECK-NEXT:    [[TMP20:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD48]], <1 x double> [[COL_LOAD51]], <1 x double> [[TMP17]])
; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST57:%.*]] = bitcast double* [[TMP21]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP20]], <1 x double>* [[VEC_CAST57]], align 8
; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST59:%.*]] = bitcast double* [[TMP22]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD60:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST59]], align 8
; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST62:%.*]] = bitcast double* [[TMP23]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD63:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST62]], align 8
; CHECK-NEXT:    [[TMP24:%.*]] = fmul contract <1 x double> [[COL_LOAD60]], [[COL_LOAD63]]
; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST68:%.*]] = bitcast double* [[TMP25]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD69:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST68]], align 8
; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST71:%.*]] = bitcast double* [[TMP26]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD72:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST71]], align 8
; CHECK-NEXT:    [[TMP27:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD69]], <1 x double> [[COL_LOAD72]], <1 x double> [[TMP24]])
; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST78:%.*]] = bitcast double* [[TMP28]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP27]], <1 x double>* [[VEC_CAST78]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %c.cast = bitcast [4 x double]* %C to <4 x double>*
  store <4 x double> %c, <4 x double>* %c.cast, align 8
  ret void
}

define void @multiply_can_hoist_multiple_insts(<4 x double>* noalias %A, <4 x double> * %B, [4 x double]* %C) {
; CHECK-LABEL: @multiply_can_hoist_multiple_insts(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr [4 x double], [4 x double]* [[C:%.*]], i64 2
; CHECK-NEXT:    [[STORE_BEGIN:%.*]] = ptrtoint [4 x double]* [[GEP]] to i64
; CHECK-NEXT:    [[STORE_END:%.*]] = add nuw nsw i64 [[STORE_BEGIN]], 32
; CHECK-NEXT:    [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK:       alias_cont:
; CHECK-NEXT:    [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT:    [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK:       copy:
; CHECK-NEXT:    [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT:    br label [[NO_ALIAS]]
; CHECK:       no_alias:
; CHECK-NEXT:    [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT:    [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST5:%.*]] = bitcast double* [[TMP7]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD6:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST5]], align 8
; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST8:%.*]] = bitcast double* [[TMP8]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD9:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST8]], align 8
; CHECK-NEXT:    [[TMP9:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD6]], <1 x double> [[COL_LOAD9]], <1 x double> [[TMP6]])
; CHECK-NEXT:    [[VEC_CAST15:%.*]] = bitcast [4 x double]* [[GEP]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP9]], <1 x double>* [[VEC_CAST15]], align 8
; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[TMP10]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD18:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_CAST20:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD21:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST20]], align 8
; CHECK-NEXT:    [[TMP11:%.*]] = fmul contract <1 x double> [[COL_LOAD18]], [[COL_LOAD21]]
; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST26:%.*]] = bitcast double* [[TMP12]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD27:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST26]], align 8
; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST29:%.*]] = bitcast double* [[TMP13]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD30:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST29]], align 8
; CHECK-NEXT:    [[TMP14:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD27]], <1 x double> [[COL_LOAD30]], <1 x double> [[TMP11]])
; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 2, i64 1
; CHECK-NEXT:    [[VEC_CAST36:%.*]] = bitcast double* [[TMP15]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP14]], <1 x double>* [[VEC_CAST36]], align 8
; CHECK-NEXT:    [[VEC_CAST38:%.*]] = bitcast <4 x double>* [[A]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD39:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST38]], align 8
; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST41:%.*]] = bitcast double* [[TMP16]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD42:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST41]], align 8
; CHECK-NEXT:    [[TMP17:%.*]] = fmul contract <1 x double> [[COL_LOAD39]], [[COL_LOAD42]]
; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST47:%.*]] = bitcast double* [[TMP18]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD48:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST47]], align 8
; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST50:%.*]] = bitcast double* [[TMP19]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD51:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST50]], align 8
; CHECK-NEXT:    [[TMP20:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD48]], <1 x double> [[COL_LOAD51]], <1 x double> [[TMP17]])
; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 2, i64 2
; CHECK-NEXT:    [[VEC_CAST57:%.*]] = bitcast double* [[TMP21]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP20]], <1 x double>* [[VEC_CAST57]], align 8
; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST59:%.*]] = bitcast double* [[TMP22]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD60:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST59]], align 8
; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST62:%.*]] = bitcast double* [[TMP23]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD63:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST62]], align 8
; CHECK-NEXT:    [[TMP24:%.*]] = fmul contract <1 x double> [[COL_LOAD60]], [[COL_LOAD63]]
; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST68:%.*]] = bitcast double* [[TMP25]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD69:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST68]], align 8
; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST71:%.*]] = bitcast double* [[TMP26]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD72:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST71]], align 8
; CHECK-NEXT:    [[TMP27:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD69]], <1 x double> [[COL_LOAD72]], <1 x double> [[TMP24]])
; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 2, i64 3
; CHECK-NEXT:    [[VEC_CAST78:%.*]] = bitcast double* [[TMP28]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP27]], <1 x double>* [[VEC_CAST78]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %gep = getelementptr [4 x double], [4 x double]* %C, i32 2
  %c.cast = bitcast [4 x double]* %gep to <4 x double>*
  store <4 x double> %c, <4 x double>* %c.cast, align 8
  ret void
}

; Make sure the correct instruction order is preserved when hoisting.
define void @multiply_can_hoist_multiple_insts2(<4 x double>* noalias %A, <4 x double> * %B, [4 x double]* %C) {
; CHECK-LABEL: @multiply_can_hoist_multiple_insts2(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[GEP_179:%.*]] = getelementptr [4 x double], [4 x double]* [[C:%.*]], i64 42
; CHECK-NEXT:    [[STORE_BEGIN:%.*]] = ptrtoint [4 x double]* [[GEP_179]] to i64
; CHECK-NEXT:    [[STORE_END:%.*]] = add nuw nsw i64 [[STORE_BEGIN]], 32
; CHECK-NEXT:    [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK:       alias_cont:
; CHECK-NEXT:    [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT:    [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT:    br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK:       copy:
; CHECK-NEXT:    [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT:    br label [[NO_ALIAS]]
; CHECK:       no_alias:
; CHECK-NEXT:    [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT:    [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST5:%.*]] = bitcast double* [[TMP7]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD6:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST5]], align 8
; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST8:%.*]] = bitcast double* [[TMP8]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD9:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST8]], align 8
; CHECK-NEXT:    [[TMP9:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD6]], <1 x double> [[COL_LOAD9]], <1 x double> [[TMP6]])
; CHECK-NEXT:    [[VEC_CAST15:%.*]] = bitcast [4 x double]* [[GEP_179]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP9]], <1 x double>* [[VEC_CAST15]], align 8
; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[TMP10]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD18:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_CAST20:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD21:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST20]], align 8
; CHECK-NEXT:    [[TMP11:%.*]] = fmul contract <1 x double> [[COL_LOAD18]], [[COL_LOAD21]]
; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST26:%.*]] = bitcast double* [[TMP12]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD27:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST26]], align 8
; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST29:%.*]] = bitcast double* [[TMP13]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD30:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST29]], align 8
; CHECK-NEXT:    [[TMP14:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD27]], <1 x double> [[COL_LOAD30]], <1 x double> [[TMP11]])
; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 42, i64 1
; CHECK-NEXT:    [[VEC_CAST36:%.*]] = bitcast double* [[TMP15]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP14]], <1 x double>* [[VEC_CAST36]], align 8
; CHECK-NEXT:    [[VEC_CAST38:%.*]] = bitcast <4 x double>* [[A]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD39:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST38]], align 8
; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST41:%.*]] = bitcast double* [[TMP16]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD42:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST41]], align 8
; CHECK-NEXT:    [[TMP17:%.*]] = fmul contract <1 x double> [[COL_LOAD39]], [[COL_LOAD42]]
; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST47:%.*]] = bitcast double* [[TMP18]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD48:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST47]], align 8
; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST50:%.*]] = bitcast double* [[TMP19]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD51:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST50]], align 8
; CHECK-NEXT:    [[TMP20:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD48]], <1 x double> [[COL_LOAD51]], <1 x double> [[TMP17]])
; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 42, i64 2
; CHECK-NEXT:    [[VEC_CAST57:%.*]] = bitcast double* [[TMP21]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP20]], <1 x double>* [[VEC_CAST57]], align 8
; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 1
; CHECK-NEXT:    [[VEC_CAST59:%.*]] = bitcast double* [[TMP22]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD60:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST59]], align 8
; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST62:%.*]] = bitcast double* [[TMP23]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD63:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST62]], align 8
; CHECK-NEXT:    [[TMP24:%.*]] = fmul contract <1 x double> [[COL_LOAD60]], [[COL_LOAD63]]
; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST68:%.*]] = bitcast double* [[TMP25]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD69:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST68]], align 8
; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr <4 x double>, <4 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT:    [[VEC_CAST71:%.*]] = bitcast double* [[TMP26]] to <1 x double>*
; CHECK-NEXT:    [[COL_LOAD72:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST71]], align 8
; CHECK-NEXT:    [[TMP27:%.*]] = call contract <1 x double> @llvm.fmuladd.v1f64(<1 x double> [[COL_LOAD69]], <1 x double> [[COL_LOAD72]], <1 x double> [[TMP24]])
; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 42, i64 3
; CHECK-NEXT:    [[VEC_CAST78:%.*]] = bitcast double* [[TMP28]] to <1 x double>*
; CHECK-NEXT:    store <1 x double> [[TMP27]], <1 x double>* [[VEC_CAST78]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %c.cast = bitcast [4 x double]* %C to <4 x double>*
  %off.0 = add i32 10, 10
  %off.1 = add i32 %off.0, 2
  %off.2 = add i32 %off.0, %off.1
  %gep.1 = getelementptr <4 x double>, <4 x double>* %c.cast, i32 %off.2
  store <4 x double> %c, <4 x double>* %gep.1, align 8
  ret void
}

define void @multiply_dont_hoist_phi(<4 x double>* noalias %A, <4 x double> * %B, [4 x double]* %C) {
; CHECK-LABEL: @multiply_dont_hoist_phi(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br label [[NEXT:%.*]]
; CHECK:       next:
; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr <4 x double>, <4 x double>* [[A:%.*]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST1]], align 8
; CHECK-NEXT:    [[VEC_GEP5:%.*]] = getelementptr <4 x double>, <4 x double>* [[B:%.*]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[VEC_GEP5]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD7:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST6]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP0:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT13]]
; CHECK-NEXT:    [[TMP1:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT16]], <2 x double> [[TMP0]])
; CHECK-NEXT:    [[VEC_CAST3:%.*]] = bitcast <4 x double>* [[B]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD4:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST3]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP2:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT]]
; CHECK-NEXT:    [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT10]], <2 x double> [[TMP2]])
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr [4 x double], [4 x double]* [[C:%.*]], i64 26, i64 0
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[TMP4]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP3]], <2 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_GEP18:%.*]] = getelementptr [4 x double], [4 x double]* [[C]], i64 26, i64 2
; CHECK-NEXT:    [[VEC_CAST19:%.*]] = bitcast double* [[VEC_GEP18]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP1]], <2 x double>* [[VEC_CAST19]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %c.cast = bitcast [4 x double]* %C to <4 x double>*
  br label %next

next:
  %p = phi i32 [ 2, %entry ]
  %off.0 = add i32 10, %p
  %off.1 = add i32 %off.0, 2
  %off.2 = add i32 %off.0, %off.1
  %gep.1 = getelementptr <4 x double>, <4 x double>* %c.cast, i32 %off.2
  store <4 x double> %c, <4 x double>* %gep.1, align 8
  ret void
}

; The address load may alias, so avoid moving it for now.
define void @multiply_dont_hoist_cast_due_to_operand(<4 x double>* noalias %A, <4 x double> * %B, [4 x double]** %C.ptr) {
; CHECK-LABEL: @multiply_dont_hoist_cast_due_to_operand(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST1]], align 8
; CHECK-NEXT:    [[VEC_CAST3:%.*]] = bitcast <4 x double>* [[B:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD4:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST3]], align 8
; CHECK-NEXT:    [[VEC_GEP5:%.*]] = getelementptr <4 x double>, <4 x double>* [[B]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[VEC_GEP5]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD7:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST6]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP0:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT]]
; CHECK-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP1:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT10]], <2 x double> [[TMP0]])
; CHECK-NEXT:    [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP2:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT13]]
; CHECK-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT16]], <2 x double> [[TMP2]])
; CHECK-NEXT:    [[TMP4:%.*]] = bitcast [4 x double]** [[C_PTR:%.*]] to double**
; CHECK-NEXT:    [[C2021:%.*]] = load double*, double** [[TMP4]], align 8
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[C2021]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP1]], <2 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_GEP18:%.*]] = getelementptr double, double* [[C2021]], i64 2
; CHECK-NEXT:    [[VEC_CAST19:%.*]] = bitcast double* [[VEC_GEP18]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP3]], <2 x double>* [[VEC_CAST19]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %C = load [4 x double]*, [4 x double]** %C.ptr
  %c.cast = bitcast [4 x double]* %C to <4 x double>*
  store <4 x double> %c, <4 x double>* %c.cast, align 8
  ret void
}

; The address load may alias, so avoid moving it for now.
define void @multiply_dont_hoist_load(<4 x double>* noalias %A, <4 x double> * %B, <4 x double>** %C.ptr) {
; CHECK-LABEL: @multiply_dont_hoist_load(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST1]], align 8
; CHECK-NEXT:    [[VEC_CAST3:%.*]] = bitcast <4 x double>* [[B:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD4:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST3]], align 8
; CHECK-NEXT:    [[VEC_GEP5:%.*]] = getelementptr <4 x double>, <4 x double>* [[B]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[VEC_GEP5]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD7:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST6]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP0:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT]]
; CHECK-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP1:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT10]], <2 x double> [[TMP0]])
; CHECK-NEXT:    [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP2:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT13]]
; CHECK-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT16]], <2 x double> [[TMP2]])
; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x double>** [[C_PTR:%.*]] to double**
; CHECK-NEXT:    [[C20:%.*]] = load double*, double** [[TMP4]], align 8
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast double* [[C20]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP1]], <2 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_GEP18:%.*]] = getelementptr double, double* [[C20]], i64 2
; CHECK-NEXT:    [[VEC_CAST19:%.*]] = bitcast double* [[VEC_GEP18]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP3]], <2 x double>* [[VEC_CAST19]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %C = load <4 x double>*, <4 x double>** %C.ptr
  store <4 x double> %c, <4 x double>* %C, align 8
  ret void
}

; The call to @get_adress may clobber memory, avoid moving it for now.
define void @multiply_dont_hoist_call(<4 x double>* noalias %A, <4 x double> * %B) {
; CHECK-LABEL: @multiply_dont_hoist_call(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr <4 x double>, <4 x double>* [[A]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST1]], align 8
; CHECK-NEXT:    [[VEC_CAST3:%.*]] = bitcast <4 x double>* [[B:%.*]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD4:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST3]], align 8
; CHECK-NEXT:    [[VEC_GEP5:%.*]] = getelementptr <4 x double>, <4 x double>* [[B]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[VEC_GEP5]] to <2 x double>*
; CHECK-NEXT:    [[COL_LOAD7:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST6]], align 8
; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP0:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT]]
; CHECK-NEXT:    [[SPLAT_SPLAT10:%.*]] = shufflevector <2 x double> [[COL_LOAD4]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP1:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT10]], <2 x double> [[TMP0]])
; CHECK-NEXT:    [[SPLAT_SPLAT13:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP2:%.*]] = fmul contract <2 x double> [[COL_LOAD]], [[SPLAT_SPLAT13]]
; CHECK-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <2 x double> [[COL_LOAD7]], <2 x double> undef, <2 x i32> <i32 1, i32 1>
; CHECK-NEXT:    [[TMP3:%.*]] = call contract <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[COL_LOAD2]], <2 x double> [[SPLAT_SPLAT16]], <2 x double> [[TMP2]])
; CHECK-NEXT:    [[C:%.*]] = call <4 x double>* @get_address()
; CHECK-NEXT:    [[VEC_CAST17:%.*]] = bitcast <4 x double>* [[C]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP1]], <2 x double>* [[VEC_CAST17]], align 8
; CHECK-NEXT:    [[VEC_GEP18:%.*]] = getelementptr <4 x double>, <4 x double>* [[C]], i64 0, i64 2
; CHECK-NEXT:    [[VEC_CAST19:%.*]] = bitcast double* [[VEC_GEP18]] to <2 x double>*
; CHECK-NEXT:    store <2 x double> [[TMP3]], <2 x double>* [[VEC_CAST19]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %a = load <4 x double>, <4 x double>* %A, align 8
  %b = load <4 x double>, <4 x double>* %B, align 8
  %c = call <4 x double> @llvm.matrix.multiply(<4 x double> %a, <4 x double> %b, i32 2, i32 2, i32 2)
  %C = call <4 x double>* @get_address()
  store <4 x double> %c, <4 x double>* %C, align 8
  ret void
}

declare <4 x double>* @get_address()


declare <4 x double> @llvm.matrix.multiply(<4 x double>, <4 x double>, i32, i32, i32)