llvm/test/Transforms/SLPVectorizer/X86/dot-product.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -mtriple=x86_64-unknown -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE2
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE4
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX

;
; dot4(ptr x, ptr y) - ((xptr y[0])+(xptr y[1])+(xptr y[2])+(xptr y[3]))
;

define double @dot4f64(ptr dereferenceable(32) %ptrx, ptr dereferenceable(32) %ptry) {
; CHECK-LABEL: @dot4f64(
; CHECK-NEXT:    [[PTRX2:%.*]] = getelementptr inbounds double, ptr [[PTRX:%.*]], i64 2
; CHECK-NEXT:    [[PTRY2:%.*]] = getelementptr inbounds double, ptr [[PTRY:%.*]], i64 2
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <2 x double>, ptr [[PTRX2]], align 4
; CHECK-NEXT:    [[TMP5:%.*]] = load <2 x double>, ptr [[PTRY2]], align 4
; CHECK-NEXT:    [[TMP6:%.*]] = fmul <2 x double> [[TMP4]], [[TMP5]]
; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT01:%.*]] = fadd double [[TMP7]], [[TMP8]]
; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x double> [[TMP6]], i32 0
; CHECK-NEXT:    [[DOT012:%.*]] = fadd double [[DOT01]], [[TMP9]]
; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x double> [[TMP6]], i32 1
; CHECK-NEXT:    [[DOT0123:%.*]] = fadd double [[DOT012]], [[TMP10]]
; CHECK-NEXT:    ret double [[DOT0123]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds double, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds double, ptr %ptry, i64 2
  %ptrx3 = getelementptr inbounds double, ptr %ptrx, i64 3
  %ptry3 = getelementptr inbounds double, ptr %ptry, i64 3
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %x2 = load double, ptr %ptrx2, align 4
  %y2 = load double, ptr %ptry2, align 4
  %x3 = load double, ptr %ptrx3, align 4
  %y3 = load double, ptr %ptry3, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %mul2 = fmul double %x2, %y2
  %mul3 = fmul double %x3, %y3
  %dot01 = fadd double %mul0, %mul1
  %dot012 = fadd double %dot01, %mul2
  %dot0123 = fadd double %dot012, %mul3
  ret double %dot0123
}

define float @dot4f32(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot4f32(
; CHECK-NEXT:    [[PTRX2:%.*]] = getelementptr inbounds float, ptr [[PTRX:%.*]], i64 2
; CHECK-NEXT:    [[PTRY2:%.*]] = getelementptr inbounds float, ptr [[PTRY:%.*]], i64 2
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <2 x float>, ptr [[PTRX2]], align 4
; CHECK-NEXT:    [[TMP5:%.*]] = load <2 x float>, ptr [[PTRY2]], align 4
; CHECK-NEXT:    [[TMP6:%.*]] = fmul <2 x float> [[TMP4]], [[TMP5]]
; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT01:%.*]] = fadd float [[TMP7]], [[TMP8]]
; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[TMP6]], i32 0
; CHECK-NEXT:    [[DOT012:%.*]] = fadd float [[DOT01]], [[TMP9]]
; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[TMP6]], i32 1
; CHECK-NEXT:    [[DOT0123:%.*]] = fadd float [[DOT012]], [[TMP10]]
; CHECK-NEXT:    ret float [[DOT0123]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds float, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds float, ptr %ptry, i64 2
  %ptrx3 = getelementptr inbounds float, ptr %ptrx, i64 3
  %ptry3 = getelementptr inbounds float, ptr %ptry, i64 3
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %x2 = load float, ptr %ptrx2, align 4
  %y2 = load float, ptr %ptry2, align 4
  %x3 = load float, ptr %ptrx3, align 4
  %y3 = load float, ptr %ptry3, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %mul2 = fmul float %x2, %y2
  %mul3 = fmul float %x3, %y3
  %dot01 = fadd float %mul0, %mul1
  %dot012 = fadd float %dot01, %mul2
  %dot0123 = fadd float %dot012, %mul3
  ret float %dot0123
}

define double @dot4f64_fast(ptr dereferenceable(32) %ptrx, ptr dereferenceable(32) %ptry) {
; CHECK-LABEL: @dot4f64_fast(
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x double>, ptr [[PTRX:%.*]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x double>, ptr [[PTRY:%.*]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <4 x double> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = call fast double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> [[TMP3]])
; CHECK-NEXT:    ret double [[TMP4]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds double, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds double, ptr %ptry, i64 2
  %ptrx3 = getelementptr inbounds double, ptr %ptrx, i64 3
  %ptry3 = getelementptr inbounds double, ptr %ptry, i64 3
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %x2 = load double, ptr %ptrx2, align 4
  %y2 = load double, ptr %ptry2, align 4
  %x3 = load double, ptr %ptrx3, align 4
  %y3 = load double, ptr %ptry3, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %mul2 = fmul double %x2, %y2
  %mul3 = fmul double %x3, %y3
  %dot01 = fadd fast double %mul0, %mul1
  %dot012 = fadd fast double %dot01, %mul2
  %dot0123 = fadd fast double %dot012, %mul3
  ret double %dot0123
}

define float @dot4f32_fast(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot4f32_fast(
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, ptr [[PTRX:%.*]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x float>, ptr [[PTRY:%.*]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <4 x float> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[TMP3]])
; CHECK-NEXT:    ret float [[TMP4]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds float, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds float, ptr %ptry, i64 2
  %ptrx3 = getelementptr inbounds float, ptr %ptrx, i64 3
  %ptry3 = getelementptr inbounds float, ptr %ptry, i64 3
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %x2 = load float, ptr %ptrx2, align 4
  %y2 = load float, ptr %ptry2, align 4
  %x3 = load float, ptr %ptrx3, align 4
  %y3 = load float, ptr %ptry3, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %mul2 = fmul float %x2, %y2
  %mul3 = fmul float %x3, %y3
  %dot01 = fadd fast float %mul0, %mul1
  %dot012 = fadd fast float %dot01, %mul2
  %dot0123 = fadd fast float %dot012, %mul3
  ret float %dot0123
}

;
; dot3(ptr x, ptr y) - ((xptr y[0])+(xptr y[1])+(xptr y[2]))
;

define double @dot3f64(ptr dereferenceable(32) %ptrx, ptr dereferenceable(32) %ptry) {
; CHECK-LABEL: @dot3f64(
; CHECK-NEXT:    [[PTRX1:%.*]] = getelementptr inbounds double, ptr [[PTRX:%.*]], i64 1
; CHECK-NEXT:    [[PTRY1:%.*]] = getelementptr inbounds double, ptr [[PTRY:%.*]], i64 1
; CHECK-NEXT:    [[X0:%.*]] = load double, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[Y0:%.*]] = load double, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[MUL0:%.*]] = fmul double [[X0]], [[Y0]]
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX1]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY1]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; CHECK-NEXT:    [[DOT01:%.*]] = fadd double [[MUL0]], [[TMP4]]
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT012:%.*]] = fadd double [[DOT01]], [[TMP5]]
; CHECK-NEXT:    ret double [[DOT012]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds double, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds double, ptr %ptry, i64 2
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %x2 = load double, ptr %ptrx2, align 4
  %y2 = load double, ptr %ptry2, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %mul2 = fmul double %x2, %y2
  %dot01 = fadd double %mul0, %mul1
  %dot012 = fadd double %dot01, %mul2
  ret double %dot012
}

define float @dot3f32(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot3f32(
; CHECK-NEXT:    [[PTRX1:%.*]] = getelementptr inbounds float, ptr [[PTRX:%.*]], i64 1
; CHECK-NEXT:    [[PTRY1:%.*]] = getelementptr inbounds float, ptr [[PTRY:%.*]], i64 1
; CHECK-NEXT:    [[X0:%.*]] = load float, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[Y0:%.*]] = load float, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[MUL0:%.*]] = fmul float [[X0]], [[Y0]]
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX1]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY1]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; CHECK-NEXT:    [[DOT01:%.*]] = fadd float [[MUL0]], [[TMP4]]
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT012:%.*]] = fadd float [[DOT01]], [[TMP5]]
; CHECK-NEXT:    ret float [[DOT012]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds float, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds float, ptr %ptry, i64 2
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %x2 = load float, ptr %ptrx2, align 4
  %y2 = load float, ptr %ptry2, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %mul2 = fmul float %x2, %y2
  %dot01 = fadd float %mul0, %mul1
  %dot012 = fadd float %dot01, %mul2
  ret float %dot012
}

define double @dot3f64_fast(ptr dereferenceable(32) %ptrx, ptr dereferenceable(32) %ptry) {
; CHECK-LABEL: @dot3f64_fast(
; CHECK-NEXT:    [[PTRX1:%.*]] = getelementptr inbounds double, ptr [[PTRX:%.*]], i64 1
; CHECK-NEXT:    [[PTRY1:%.*]] = getelementptr inbounds double, ptr [[PTRY:%.*]], i64 1
; CHECK-NEXT:    [[X0:%.*]] = load double, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[Y0:%.*]] = load double, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[MUL0:%.*]] = fmul double [[X0]], [[Y0]]
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX1]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY1]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; CHECK-NEXT:    [[DOT01:%.*]] = fadd fast double [[MUL0]], [[TMP4]]
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT012:%.*]] = fadd fast double [[DOT01]], [[TMP5]]
; CHECK-NEXT:    ret double [[DOT012]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds double, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds double, ptr %ptry, i64 2
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %x2 = load double, ptr %ptrx2, align 4
  %y2 = load double, ptr %ptry2, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %mul2 = fmul double %x2, %y2
  %dot01 = fadd fast double %mul0, %mul1
  %dot012 = fadd fast double %dot01, %mul2
  ret double %dot012
}

define float @dot3f32_fast(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot3f32_fast(
; CHECK-NEXT:    [[PTRX1:%.*]] = getelementptr inbounds float, ptr [[PTRX:%.*]], i64 1
; CHECK-NEXT:    [[PTRY1:%.*]] = getelementptr inbounds float, ptr [[PTRY:%.*]], i64 1
; CHECK-NEXT:    [[X0:%.*]] = load float, ptr [[PTRX]], align 4
; CHECK-NEXT:    [[Y0:%.*]] = load float, ptr [[PTRY]], align 4
; CHECK-NEXT:    [[MUL0:%.*]] = fmul float [[X0]], [[Y0]]
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX1]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY1]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; CHECK-NEXT:    [[DOT01:%.*]] = fadd fast float [[MUL0]], [[TMP4]]
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT012:%.*]] = fadd fast float [[DOT01]], [[TMP5]]
; CHECK-NEXT:    ret float [[DOT012]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %ptrx2 = getelementptr inbounds float, ptr %ptrx, i64 2
  %ptry2 = getelementptr inbounds float, ptr %ptry, i64 2
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %x2 = load float, ptr %ptrx2, align 4
  %y2 = load float, ptr %ptry2, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %mul2 = fmul float %x2, %y2
  %dot01 = fadd fast float %mul0, %mul1
  %dot012 = fadd fast float %dot01, %mul2
  ret float %dot012
}

;
; dot2(ptr x, ptr y) - ((xptr y[0])+(xptr y[1]))
;

define double @dot2f64(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot2f64(
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX:%.*]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY:%.*]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT01:%.*]] = fadd double [[TMP4]], [[TMP5]]
; CHECK-NEXT:    ret double [[DOT01]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %dot01 = fadd double %mul0, %mul1
  ret double %dot01
}

define float @dot2f32(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; CHECK-LABEL: @dot2f32(
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX:%.*]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY:%.*]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; CHECK-NEXT:    [[DOT01:%.*]] = fadd float [[TMP4]], [[TMP5]]
; CHECK-NEXT:    ret float [[DOT01]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %dot01 = fadd float %mul0, %mul1
  ret float %dot01
}

define double @dot2f64_fast(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; SSE2-LABEL: @dot2f64_fast(
; SSE2-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX:%.*]], align 4
; SSE2-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY:%.*]], align 4
; SSE2-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; SSE2-NEXT:    [[TMP4:%.*]] = call fast double @llvm.vector.reduce.fadd.v2f64(double 0.000000e+00, <2 x double> [[TMP3]])
; SSE2-NEXT:    ret double [[TMP4]]
;
; SSE4-LABEL: @dot2f64_fast(
; SSE4-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX:%.*]], align 4
; SSE4-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY:%.*]], align 4
; SSE4-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; SSE4-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; SSE4-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; SSE4-NEXT:    [[DOT01:%.*]] = fadd fast double [[TMP4]], [[TMP5]]
; SSE4-NEXT:    ret double [[DOT01]]
;
; AVX-LABEL: @dot2f64_fast(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[PTRX:%.*]], align 4
; AVX-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[PTRY:%.*]], align 4
; AVX-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[TMP1]], [[TMP2]]
; AVX-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; AVX-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; AVX-NEXT:    [[DOT01:%.*]] = fadd fast double [[TMP4]], [[TMP5]]
; AVX-NEXT:    ret double [[DOT01]]
;
  %ptrx1 = getelementptr inbounds double, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds double, ptr %ptry, i64 1
  %x0 = load double, ptr %ptrx, align 4
  %y0 = load double, ptr %ptry, align 4
  %x1 = load double, ptr %ptrx1, align 4
  %y1 = load double, ptr %ptry1, align 4
  %mul0 = fmul double %x0, %y0
  %mul1 = fmul double %x1, %y1
  %dot01 = fadd fast double %mul0, %mul1
  ret double %dot01
}

define float @dot2f32_fast(ptr dereferenceable(16) %ptrx, ptr dereferenceable(16) %ptry) {
; SSE2-LABEL: @dot2f32_fast(
; SSE2-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX:%.*]], align 4
; SSE2-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY:%.*]], align 4
; SSE2-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; SSE2-NEXT:    [[TMP4:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float 0.000000e+00, <2 x float> [[TMP3]])
; SSE2-NEXT:    ret float [[TMP4]]
;
; SSE4-LABEL: @dot2f32_fast(
; SSE4-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX:%.*]], align 4
; SSE4-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY:%.*]], align 4
; SSE4-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; SSE4-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; SSE4-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; SSE4-NEXT:    [[DOT01:%.*]] = fadd fast float [[TMP4]], [[TMP5]]
; SSE4-NEXT:    ret float [[DOT01]]
;
; AVX-LABEL: @dot2f32_fast(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[PTRX:%.*]], align 4
; AVX-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[PTRY:%.*]], align 4
; AVX-NEXT:    [[TMP3:%.*]] = fmul <2 x float> [[TMP1]], [[TMP2]]
; AVX-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; AVX-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; AVX-NEXT:    [[DOT01:%.*]] = fadd fast float [[TMP4]], [[TMP5]]
; AVX-NEXT:    ret float [[DOT01]]
;
  %ptrx1 = getelementptr inbounds float, ptr %ptrx, i64 1
  %ptry1 = getelementptr inbounds float, ptr %ptry, i64 1
  %x0 = load float, ptr %ptrx, align 4
  %y0 = load float, ptr %ptry, align 4
  %x1 = load float, ptr %ptrx1, align 4
  %y1 = load float, ptr %ptry1, align 4
  %mul0 = fmul float %x0, %y0
  %mul1 = fmul float %x1, %y1
  %dot01 = fadd fast float %mul0, %mul1
  ret float %dot01
}