llvm/test/Transforms/SLPVectorizer/AArch64/loop-splat-inst-count.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S --passes=slp-vectorizer -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s

define void @test(ptr noalias %a, ptr noalias %b, ptr noalias %c) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    br label %[[OUTER:.*]]
; CHECK:       [[OUTER]]:
; CHECK-NEXT:    [[QZ:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[QZ_NEXT:%.*]], %[[OUTER_LATCH:.*]] ]
; CHECK-NEXT:    [[ABASE:%.*]] = mul i64 [[QZ]], 12
; CHECK-NEXT:    [[PA00:%.*]] = getelementptr inbounds double, ptr [[A]], i64 [[ABASE]]
; CHECK-NEXT:    [[PA10:%.*]] = getelementptr inbounds double, ptr [[PA00]], i64 3
; CHECK-NEXT:    [[PA20:%.*]] = getelementptr inbounds double, ptr [[PA00]], i64 6
; CHECK-NEXT:    [[PA30:%.*]] = getelementptr inbounds double, ptr [[PA00]], i64 9
; CHECK-NEXT:    [[TMP0:%.*]] = load <2 x double>, ptr [[PA00]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PA10]], align 8
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PA20]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = load <2 x double>, ptr [[PA30]], align 8
; CHECK-NEXT:    [[CBASE:%.*]] = mul i64 [[QZ]], 6
; CHECK-NEXT:    br label %[[INNER:.*]]
; CHECK:       [[INNER]]:
; CHECK-NEXT:    [[DY:%.*]] = phi i64 [ 0, %[[OUTER]] ], [ [[DY_NEXT:%.*]], %[[INNER]] ]
; CHECK-NEXT:    [[PB0:%.*]] = getelementptr inbounds double, ptr [[B]], i64 [[DY]]
; CHECK-NEXT:    [[S0:%.*]] = load double, ptr [[PB0]], align 8
; CHECK-NEXT:    [[PB1:%.*]] = getelementptr inbounds double, ptr [[PB0]], i64 3
; CHECK-NEXT:    [[S1:%.*]] = load double, ptr [[PB1]], align 8
; CHECK-NEXT:    [[PB2:%.*]] = getelementptr inbounds double, ptr [[PB0]], i64 6
; CHECK-NEXT:    [[S2:%.*]] = load double, ptr [[PB2]], align 8
; CHECK-NEXT:    [[PB3:%.*]] = getelementptr inbounds double, ptr [[PB0]], i64 9
; CHECK-NEXT:    [[S3:%.*]] = load double, ptr [[PB3]], align 8
; CHECK-NEXT:    [[DY2:%.*]] = shl i64 [[DY]], 1
; CHECK-NEXT:    [[CIDX:%.*]] = add i64 [[CBASE]], [[DY2]]
; CHECK-NEXT:    [[PC0:%.*]] = getelementptr inbounds double, ptr [[C]], i64 [[CIDX]]
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> poison, double [[S0]], i64 0
; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x double> [[TMP4]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP6:%.*]] = fmul fast <2 x double> [[TMP0]], [[TMP5]]
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> poison, double [[S1]], i64 0
; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <2 x double> [[TMP7]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP9:%.*]] = fmul fast <2 x double> [[TMP1]], [[TMP8]]
; CHECK-NEXT:    [[TMP10:%.*]] = fadd fast <2 x double> [[TMP9]], [[TMP6]]
; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x double> poison, double [[S2]], i64 0
; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x double> [[TMP11]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP13:%.*]] = fmul fast <2 x double> [[TMP2]], [[TMP12]]
; CHECK-NEXT:    [[TMP14:%.*]] = fadd fast <2 x double> [[TMP13]], [[TMP10]]
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x double> poison, double [[S3]], i64 0
; CHECK-NEXT:    [[TMP16:%.*]] = shufflevector <2 x double> [[TMP15]], <2 x double> poison, <2 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP17:%.*]] = fmul fast <2 x double> [[TMP3]], [[TMP16]]
; CHECK-NEXT:    [[TMP18:%.*]] = fadd fast <2 x double> [[TMP17]], [[TMP14]]
; CHECK-NEXT:    store <2 x double> [[TMP18]], ptr [[PC0]], align 8
; CHECK-NEXT:    [[DY_NEXT]] = add nuw nsw i64 [[DY]], 1
; CHECK-NEXT:    [[DY_CMP:%.*]] = icmp ult i64 [[DY_NEXT]], 3
; CHECK-NEXT:    br i1 [[DY_CMP]], label %[[INNER]], label %[[OUTER_LATCH]]
; CHECK:       [[OUTER_LATCH]]:
; CHECK-NEXT:    [[QZ_NEXT]] = add nuw nsw i64 [[QZ]], 1
; CHECK-NEXT:    [[QZ_CMP:%.*]] = icmp ult i64 [[QZ_NEXT]], 4
; CHECK-NEXT:    br i1 [[QZ_CMP]], label %[[OUTER]], label %[[EXIT:.*]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  br label %outer

outer:
  %qz = phi i64 [ 0, %entry ], [ %qz.next, %outer.latch ]
  %abase = mul i64 %qz, 12
  %pa00 = getelementptr inbounds double, ptr %a, i64 %abase
  %a00 = load double, ptr %pa00, align 8
  %pa01 = getelementptr inbounds double, ptr %pa00, i64 1
  %a01 = load double, ptr %pa01, align 8
  %pa10 = getelementptr inbounds double, ptr %pa00, i64 3
  %a10 = load double, ptr %pa10, align 8
  %pa11 = getelementptr inbounds double, ptr %pa00, i64 4
  %a11 = load double, ptr %pa11, align 8
  %pa20 = getelementptr inbounds double, ptr %pa00, i64 6
  %a20 = load double, ptr %pa20, align 8
  %pa21 = getelementptr inbounds double, ptr %pa00, i64 7
  %a21 = load double, ptr %pa21, align 8
  %pa30 = getelementptr inbounds double, ptr %pa00, i64 9
  %a30 = load double, ptr %pa30, align 8
  %pa31 = getelementptr inbounds double, ptr %pa00, i64 10
  %a31 = load double, ptr %pa31, align 8
  %cbase = mul i64 %qz, 6
  br label %inner

inner:
  %dy = phi i64 [ 0, %outer ], [ %dy.next, %inner ]
  %pb0 = getelementptr inbounds double, ptr %b, i64 %dy
  %s0 = load double, ptr %pb0, align 8
  %pb1 = getelementptr inbounds double, ptr %pb0, i64 3
  %s1 = load double, ptr %pb1, align 8
  %pb2 = getelementptr inbounds double, ptr %pb0, i64 6
  %s2 = load double, ptr %pb2, align 8
  %pb3 = getelementptr inbounds double, ptr %pb0, i64 9
  %s3 = load double, ptr %pb3, align 8
  %m00 = fmul fast double %a00, %s0
  %m10 = fmul fast double %a10, %s1
  %u10 = fadd fast double %m10, %m00
  %m20 = fmul fast double %a20, %s2
  %u20 = fadd fast double %m20, %u10
  %m30 = fmul fast double %a30, %s3
  %u0 = fadd fast double %m30, %u20
  %m01 = fmul fast double %a01, %s0
  %m11 = fmul fast double %a11, %s1
  %u11 = fadd fast double %m11, %m01
  %m21 = fmul fast double %a21, %s2
  %u21 = fadd fast double %m21, %u11
  %m31 = fmul fast double %a31, %s3
  %u1 = fadd fast double %m31, %u21
  %dy2 = shl i64 %dy, 1
  %cidx = add i64 %cbase, %dy2
  %pc0 = getelementptr inbounds double, ptr %c, i64 %cidx
  store double %u0, ptr %pc0, align 8
  %pc1 = getelementptr inbounds double, ptr %pc0, i64 1
  store double %u1, ptr %pc1, align 8
  %dy.next = add nuw nsw i64 %dy, 1
  %dy.cmp = icmp ult i64 %dy.next, 3
  br i1 %dy.cmp, label %inner, label %outer.latch

outer.latch:
  %qz.next = add nuw nsw i64 %qz, 1
  %qz.cmp = icmp ult i64 %qz.next, 4
  br i1 %qz.cmp, label %outer, label %exit

exit:
  ret void
}