xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/RISCV/vec3-base.ll (revision 3133acf1fbd1cc57ea8e74288ee9a0acd027d749)

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=slp-vectorizer -slp-vectorize-non-power-of-2 -mtriple=riscv64 -mattr=+v -S %s | FileCheck --check-prefixes=CHECK,NON-POW2 %s
; RUN: opt -passes=slp-vectorizer -slp-vectorize-non-power-of-2=false -mtriple=riscv64 -mattr=+v -S %s | FileCheck --check-prefixes=CHECK,POW2-ONLY %s

define void @v3_load_i32_mul_by_constant_store(ptr %src, ptr %dst) {
; NON-POW2-LABEL: @v3_load_i32_mul_by_constant_store(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP0:%.*]] = load <3 x i32>, ptr [[GEP_SRC_0]], align 4
; NON-POW2-NEXT:    [[TMP1:%.*]] = mul nsw <3 x i32> [[TMP0]], splat (i32 10)
; NON-POW2-NEXT:    store <3 x i32> [[TMP1]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @v3_load_i32_mul_by_constant_store(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2:%.*]] = load i32, ptr [[GEP_SRC_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_SRC_2]], 10
; POW2-ONLY-NEXT:    [[TMP0:%.*]] = load <2 x i32>, ptr [[GEP_SRC_0]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = mul nsw <2 x i32> [[TMP0]], splat (i32 10)
; POW2-ONLY-NEXT:    store <2 x i32> [[TMP1]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; POW2-ONLY-NEXT:    store i32 [[MUL_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %gep.src.0 = getelementptr inbounds i32, ptr %src, i32 0
  %l.src.0 = load i32, ptr %gep.src.0, align 4
  %mul.0 = mul nsw i32 %l.src.0, 10

  %gep.src.1 = getelementptr inbounds i32, ptr %src, i32 1
  %l.src.1 = load i32, ptr %gep.src.1, align 4
  %mul.1 = mul nsw i32 %l.src.1, 10

  %gep.src.2 = getelementptr inbounds i32, ptr %src, i32 2
  %l.src.2 = load i32, ptr %gep.src.2, align 4
  %mul.2 = mul nsw i32 %l.src.2, 10

  store i32 %mul.0, ptr %dst

  %dst.1 = getelementptr i32, ptr %dst, i32 1
  store i32 %mul.1, ptr %dst.1

  %dst.2 = getelementptr i32, ptr %dst, i32 2
  store i32 %mul.2, ptr %dst.2

  ret void
}

; Should no be vectorized with a undef/poison element as padding, as
; division by undef/poison may cause UB.  Must use VL predication or
; masking instead, where RISCV wins.
define void @v3_load_i32_udiv_by_constant_store(ptr %src, ptr %dst) {
; NON-POW2-LABEL: @v3_load_i32_udiv_by_constant_store(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP0:%.*]] = load <3 x i32>, ptr [[GEP_SRC_0]], align 4
; NON-POW2-NEXT:    [[TMP1:%.*]] = udiv <3 x i32> splat (i32 10), [[TMP0]]
; NON-POW2-NEXT:    store <3 x i32> [[TMP1]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @v3_load_i32_udiv_by_constant_store(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_SRC_0:%.*]] = load i32, ptr [[GEP_SRC_0]], align 4
; POW2-ONLY-NEXT:    [[MUL_0:%.*]] = udiv i32 10, [[L_SRC_0]]
; POW2-ONLY-NEXT:    [[GEP_SRC_1:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 1
; POW2-ONLY-NEXT:    [[L_SRC_1:%.*]] = load i32, ptr [[GEP_SRC_1]], align 4
; POW2-ONLY-NEXT:    [[MUL_1:%.*]] = udiv i32 10, [[L_SRC_1]]
; POW2-ONLY-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2:%.*]] = load i32, ptr [[GEP_SRC_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = udiv i32 10, [[L_SRC_2]]
; POW2-ONLY-NEXT:    store i32 [[MUL_0]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[DST_1:%.*]] = getelementptr i32, ptr [[DST]], i32 1
; POW2-ONLY-NEXT:    store i32 [[MUL_1]], ptr [[DST_1]], align 4
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; POW2-ONLY-NEXT:    store i32 [[MUL_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %gep.src.0 = getelementptr inbounds i32, ptr %src, i32 0
  %l.src.0 = load i32, ptr %gep.src.0, align 4
  %mul.0 = udiv i32 10, %l.src.0

  %gep.src.1 = getelementptr inbounds i32, ptr %src, i32 1
  %l.src.1 = load i32, ptr %gep.src.1, align 4
  %mul.1 = udiv i32 10, %l.src.1

  %gep.src.2 = getelementptr inbounds i32, ptr %src, i32 2
  %l.src.2 = load i32, ptr %gep.src.2, align 4
  %mul.2 = udiv i32 10, %l.src.2

  store i32 %mul.0, ptr %dst

  %dst.1 = getelementptr i32, ptr %dst, i32 1
  store i32 %mul.1, ptr %dst.1

  %dst.2 = getelementptr i32, ptr %dst, i32 2
  store i32 %mul.2, ptr %dst.2

  ret void
}



define void @v3_load_i32_mul_store(ptr %src.1, ptr %src.2, ptr %dst) {
; NON-POW2-LABEL: @v3_load_i32_mul_store(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[GEP_SRC_1_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_1:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_SRC_2_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_2:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP0:%.*]] = load <3 x i32>, ptr [[GEP_SRC_1_0]], align 4
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x i32>, ptr [[GEP_SRC_2_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = mul nsw <3 x i32> [[TMP0]], [[TMP1]]
; NON-POW2-NEXT:    store <3 x i32> [[TMP2]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @v3_load_i32_mul_store(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[GEP_SRC_1_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_1:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_2_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_2:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_1_2:%.*]] = getelementptr inbounds i32, ptr [[SRC_1]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_1_2:%.*]] = load i32, ptr [[GEP_SRC_1_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_2_2:%.*]] = getelementptr inbounds i32, ptr [[SRC_2]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2_2:%.*]] = load i32, ptr [[GEP_SRC_2_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_SRC_1_2]], [[L_SRC_2_2]]
; POW2-ONLY-NEXT:    [[TMP0:%.*]] = load <2 x i32>, ptr [[GEP_SRC_1_0]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr [[GEP_SRC_2_0]], align 4
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = mul nsw <2 x i32> [[TMP0]], [[TMP1]]
; POW2-ONLY-NEXT:    store <2 x i32> [[TMP2]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; POW2-ONLY-NEXT:    store i32 [[MUL_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %gep.src.1.0 = getelementptr inbounds i32, ptr %src.1, i32 0
  %l.src.1.0 = load i32, ptr %gep.src.1.0, align 4
  %gep.src.2.0 = getelementptr inbounds i32, ptr %src.2, i32 0
  %l.src.2.0 = load i32, ptr %gep.src.2.0, align 4
  %mul.0 = mul nsw i32 %l.src.1.0, %l.src.2.0

  %gep.src.1.1 = getelementptr inbounds i32, ptr %src.1, i32 1
  %l.src.1.1 = load i32, ptr %gep.src.1.1, align 4
  %gep.src.2.1 = getelementptr inbounds i32, ptr %src.2, i32 1
  %l.src.2.1 = load i32, ptr %gep.src.2.1, align 4
  %mul.1 = mul nsw i32 %l.src.1.1, %l.src.2.1

  %gep.src.1.2 = getelementptr inbounds i32, ptr %src.1, i32 2
  %l.src.1.2 = load i32, ptr %gep.src.1.2, align 4
  %gep.src.2.2 = getelementptr inbounds i32, ptr %src.2, i32 2
  %l.src.2.2 = load i32, ptr %gep.src.2.2, align 4
  %mul.2 = mul nsw i32 %l.src.1.2, %l.src.2.2

  store i32 %mul.0, ptr %dst

  %dst.1 = getelementptr i32, ptr %dst, i32 1
  store i32 %mul.1, ptr %dst.1

  %dst.2 = getelementptr i32, ptr %dst, i32 2
  store i32 %mul.2, ptr %dst.2

  ret void
}

define void @v3_load_i32_mul_add_const_store(ptr %src.1, ptr %src.2, ptr %dst) {
; NON-POW2-LABEL: @v3_load_i32_mul_add_const_store(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[GEP_SRC_1_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_1:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_SRC_2_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_2:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP0:%.*]] = load <3 x i32>, ptr [[GEP_SRC_1_0]], align 4
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x i32>, ptr [[GEP_SRC_2_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = mul nsw <3 x i32> [[TMP0]], [[TMP1]]
; NON-POW2-NEXT:    [[TMP3:%.*]] = add <3 x i32> [[TMP2]], splat (i32 9)
; NON-POW2-NEXT:    store <3 x i32> [[TMP3]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @v3_load_i32_mul_add_const_store(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[GEP_SRC_1_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_1:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_2_0:%.*]] = getelementptr inbounds i32, ptr [[SRC_2:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_1_2:%.*]] = getelementptr inbounds i32, ptr [[SRC_1]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_1_2:%.*]] = load i32, ptr [[GEP_SRC_1_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_2_2:%.*]] = getelementptr inbounds i32, ptr [[SRC_2]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2_2:%.*]] = load i32, ptr [[GEP_SRC_2_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_SRC_1_2]], [[L_SRC_2_2]]
; POW2-ONLY-NEXT:    [[ADD_2:%.*]] = add i32 [[MUL_2]], 9
; POW2-ONLY-NEXT:    [[TMP0:%.*]] = load <2 x i32>, ptr [[GEP_SRC_1_0]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = load <2 x i32>, ptr [[GEP_SRC_2_0]], align 4
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = mul nsw <2 x i32> [[TMP0]], [[TMP1]]
; POW2-ONLY-NEXT:    [[TMP3:%.*]] = add <2 x i32> [[TMP2]], splat (i32 9)
; POW2-ONLY-NEXT:    store <2 x i32> [[TMP3]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr i32, ptr [[DST]], i32 2
; POW2-ONLY-NEXT:    store i32 [[ADD_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %gep.src.1.0 = getelementptr inbounds i32, ptr %src.1, i32 0
  %l.src.1.0 = load i32, ptr %gep.src.1.0, align 4
  %gep.src.2.0 = getelementptr inbounds i32, ptr %src.2, i32 0
  %l.src.2.0 = load i32, ptr %gep.src.2.0, align 4
  %mul.0 = mul nsw i32 %l.src.1.0, %l.src.2.0
  %add.0 = add i32 %mul.0, 9

  %gep.src.1.1 = getelementptr inbounds i32, ptr %src.1, i32 1
  %l.src.1.1 = load i32, ptr %gep.src.1.1, align 4
  %gep.src.2.1 = getelementptr inbounds i32, ptr %src.2, i32 1
  %l.src.2.1 = load i32, ptr %gep.src.2.1, align 4
  %mul.1 = mul nsw i32 %l.src.1.1, %l.src.2.1
  %add.1 = add i32 %mul.1, 9

  %gep.src.1.2 = getelementptr inbounds i32, ptr %src.1, i32 2
  %l.src.1.2 = load i32, ptr %gep.src.1.2, align 4
  %gep.src.2.2 = getelementptr inbounds i32, ptr %src.2, i32 2
  %l.src.2.2 = load i32, ptr %gep.src.2.2, align 4
  %mul.2 = mul nsw i32 %l.src.1.2, %l.src.2.2
  %add.2 = add i32 %mul.2, 9

  store i32 %add.0, ptr %dst

  %dst.1 = getelementptr i32, ptr %dst, i32 1
  store i32 %add.1, ptr %dst.1

  %dst.2 = getelementptr i32, ptr %dst, i32 2
  store i32 %add.2, ptr %dst.2

  ret void
}

define void @v3_load_f32_fadd_fadd_by_constant_store(ptr %src, ptr %dst) {
; NON-POW2-LABEL: @v3_load_f32_fadd_fadd_by_constant_store(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds float, ptr [[SRC:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP0:%.*]] = load <3 x float>, ptr [[GEP_SRC_0]], align 4
; NON-POW2-NEXT:    [[TMP1:%.*]] = fadd <3 x float> [[TMP0]], splat (float 1.000000e+01)
; NON-POW2-NEXT:    store <3 x float> [[TMP1]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @v3_load_f32_fadd_fadd_by_constant_store(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds float, ptr [[SRC:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds float, ptr [[SRC]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2:%.*]] = load float, ptr [[GEP_SRC_2]], align 4
; POW2-ONLY-NEXT:    [[FADD_2:%.*]] = fadd float [[L_SRC_2]], 1.000000e+01
; POW2-ONLY-NEXT:    [[TMP0:%.*]] = load <2 x float>, ptr [[GEP_SRC_0]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = fadd <2 x float> [[TMP0]], splat (float 1.000000e+01)
; POW2-ONLY-NEXT:    store <2 x float> [[TMP1]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr float, ptr [[DST]], i32 2
; POW2-ONLY-NEXT:    store float [[FADD_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %gep.src.0 = getelementptr inbounds float, ptr %src, i32 0
  %l.src.0 = load float , ptr %gep.src.0, align 4
  %fadd.0 = fadd float %l.src.0, 10.0

  %gep.src.1 = getelementptr inbounds float , ptr %src, i32 1
  %l.src.1 = load float, ptr %gep.src.1, align 4
  %fadd.1 = fadd float %l.src.1, 10.0

  %gep.src.2 = getelementptr inbounds float, ptr %src, i32 2
  %l.src.2 = load float, ptr %gep.src.2, align 4
  %fadd.2 = fadd float %l.src.2, 10.0

  store float %fadd.0, ptr %dst

  %dst.1 = getelementptr float, ptr %dst, i32 1
  store float %fadd.1, ptr %dst.1

  %dst.2 = getelementptr float, ptr %dst, i32 2
  store float %fadd.2, ptr %dst.2

  ret void
}

define void @phi_store3(ptr %dst) {
; NON-POW2-LABEL: @phi_store3(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    br label [[EXIT:%.*]]
; NON-POW2:       invoke.cont8.loopexit:
; NON-POW2-NEXT:    br label [[EXIT]]
; NON-POW2:       exit:
; NON-POW2-NEXT:    [[TMP0:%.*]] = phi <3 x i32> [ <i32 1, i32 2, i32 3>, [[ENTRY:%.*]] ], [ poison, [[INVOKE_CONT8_LOOPEXIT:%.*]] ]
; NON-POW2-NEXT:    store <3 x i32> [[TMP0]], ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @phi_store3(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    br label [[EXIT:%.*]]
; POW2-ONLY:       invoke.cont8.loopexit:
; POW2-ONLY-NEXT:    br label [[EXIT]]
; POW2-ONLY:       exit:
; POW2-ONLY-NEXT:    [[P_2:%.*]] = phi i32 [ 3, [[ENTRY:%.*]] ], [ 0, [[INVOKE_CONT8_LOOPEXIT:%.*]] ]
; POW2-ONLY-NEXT:    [[TMP0:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[ENTRY]] ], [ poison, [[INVOKE_CONT8_LOOPEXIT]] ]
; POW2-ONLY-NEXT:    [[DST_2:%.*]] = getelementptr i32, ptr [[DST:%.*]], i32 2
; POW2-ONLY-NEXT:    store <2 x i32> [[TMP0]], ptr [[DST]], align 4
; POW2-ONLY-NEXT:    store i32 [[P_2]], ptr [[DST_2]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  br label %exit

invoke.cont8.loopexit:                            ; No predecessors!
  br label %exit

exit:
  %p.0 = phi i32 [ 1, %entry ], [ 0, %invoke.cont8.loopexit ]
  %p.1 = phi i32 [ 2, %entry ], [ 0, %invoke.cont8.loopexit ]
  %p.2 = phi i32 [ 3, %entry ], [ 0, %invoke.cont8.loopexit ]

  %dst.1 = getelementptr i32, ptr %dst, i32 1
  %dst.2 = getelementptr i32, ptr %dst, i32 2

  store i32 %p.0, ptr %dst, align 4
  store i32 %p.1, ptr %dst.1, align 4
  store i32 %p.2, ptr %dst.2, align 4
  ret void
}

define void @store_try_reorder(ptr %dst) {
; NON-POW2-LABEL: @store_try_reorder(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    store <3 x i32> zeroinitializer, ptr [[DST:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @store_try_reorder(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[ADD:%.*]] = add i32 0, 0
; POW2-ONLY-NEXT:    store i32 [[ADD]], ptr [[DST:%.*]], align 4
; POW2-ONLY-NEXT:    [[ARRAYIDX_I1887:%.*]] = getelementptr i32, ptr [[DST]], i64 1
; POW2-ONLY-NEXT:    store <2 x i32> zeroinitializer, ptr [[ARRAYIDX_I1887]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %add = add i32 0, 0
  store i32 %add, ptr %dst, align 4
  %add207 = sub i32 0, 0
  %arrayidx.i1887 = getelementptr i32, ptr %dst, i64 1
  store i32 %add207, ptr %arrayidx.i1887, align 4
  %add216 = sub i32 0, 0
  %arrayidx.i1891 = getelementptr i32, ptr %dst, i64 2
  store i32 %add216, ptr %arrayidx.i1891, align 4
  ret void
}

define void @vec3_fpext_cost(ptr %Colour, float %0) {
; NON-POW2-LABEL: @vec3_fpext_cost(
; NON-POW2-NEXT:  entry:
; NON-POW2-NEXT:    [[TMP1:%.*]] = insertelement <3 x float> poison, float [[TMP0:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP2:%.*]] = shufflevector <3 x float> [[TMP1]], <3 x float> poison, <3 x i32> zeroinitializer
; NON-POW2-NEXT:    [[TMP3:%.*]] = fpext <3 x float> [[TMP2]] to <3 x double>
; NON-POW2-NEXT:    [[TMP4:%.*]] = call <3 x double> @llvm.fmuladd.v3f64(<3 x double> [[TMP3]], <3 x double> zeroinitializer, <3 x double> zeroinitializer)
; NON-POW2-NEXT:    [[TMP5:%.*]] = fptrunc <3 x double> [[TMP4]] to <3 x float>
; NON-POW2-NEXT:    store <3 x float> [[TMP5]], ptr [[COLOUR:%.*]], align 4
; NON-POW2-NEXT:    ret void
;
; POW2-ONLY-LABEL: @vec3_fpext_cost(
; POW2-ONLY-NEXT:  entry:
; POW2-ONLY-NEXT:    [[ARRAYIDX80:%.*]] = getelementptr float, ptr [[COLOUR:%.*]], i64 2
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> poison, float [[TMP0:%.*]], i32 0
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <2 x i32> zeroinitializer
; POW2-ONLY-NEXT:    [[TMP3:%.*]] = fpext <2 x float> [[TMP2]] to <2 x double>
; POW2-ONLY-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP3]], <2 x double> zeroinitializer, <2 x double> zeroinitializer)
; POW2-ONLY-NEXT:    [[TMP5:%.*]] = fptrunc <2 x double> [[TMP4]] to <2 x float>
; POW2-ONLY-NEXT:    store <2 x float> [[TMP5]], ptr [[COLOUR]], align 4
; POW2-ONLY-NEXT:    [[CONV78:%.*]] = fpext float [[TMP0]] to double
; POW2-ONLY-NEXT:    [[TMP6:%.*]] = call double @llvm.fmuladd.f64(double [[CONV78]], double 0.000000e+00, double 0.000000e+00)
; POW2-ONLY-NEXT:    [[CONV82:%.*]] = fptrunc double [[TMP6]] to float
; POW2-ONLY-NEXT:    store float [[CONV82]], ptr [[ARRAYIDX80]], align 4
; POW2-ONLY-NEXT:    ret void
;
entry:
  %arrayidx72 = getelementptr float, ptr %Colour, i64 1
  %arrayidx80 = getelementptr float, ptr %Colour, i64 2
  %conv62 = fpext float %0 to double
  %1 = call double @llvm.fmuladd.f64(double %conv62, double 0.000000e+00, double 0.000000e+00)
  %conv66 = fptrunc double %1 to float
  store float %conv66, ptr %Colour, align 4
  %conv70 = fpext float %0 to double
  %2 = call double @llvm.fmuladd.f64(double %conv70, double 0.000000e+00, double 0.000000e+00)
  %conv74 = fptrunc double %2 to float
  store float %conv74, ptr %arrayidx72, align 4
  %conv78 = fpext float %0 to double
  %3 = call double @llvm.fmuladd.f64(double %conv78, double 0.000000e+00, double 0.000000e+00)
  %conv82 = fptrunc double %3 to float
  store float %conv82, ptr %arrayidx80, align 4
  ret void
}

define void @fpext_scatter(ptr %dst, double %conv) {
; CHECK-LABEL: @fpext_scatter(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[CONV25:%.*]] = fptrunc double [[CONV:%.*]] to float
; CHECK-NEXT:    [[LENGTHS:%.*]] = getelementptr float, ptr [[DST:%.*]], i64 0
; CHECK-NEXT:    store float [[CONV25]], ptr [[LENGTHS]], align 4
; CHECK-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr float, ptr [[DST]], i64 1
; CHECK-NEXT:    store float [[CONV25]], ptr [[ARRAYIDX32]], align 4
; CHECK-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr float, ptr [[DST]], i64 2
; CHECK-NEXT:    store float [[CONV25]], ptr [[ARRAYIDX37]], align 4
; CHECK-NEXT:    ret void
;
entry:
  %conv25 = fptrunc double %conv to float
  %Lengths = getelementptr float, ptr %dst, i64 0
  store float %conv25, ptr %Lengths, align 4
  %arrayidx32 = getelementptr float, ptr %dst, i64 1
  store float %conv25, ptr %arrayidx32, align 4
  %arrayidx37 = getelementptr float, ptr %dst, i64 2
  store float %conv25, ptr %arrayidx37, align 4
  ret void
}

define i32 @reduce_add(ptr %src) {
; CHECK-LABEL: @reduce_add(
; CHECK-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; CHECK-NEXT:    [[L_SRC_0:%.*]] = load i32, ptr [[GEP_SRC_0]], align 4
; CHECK-NEXT:    [[GEP_SRC_1:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 1
; CHECK-NEXT:    [[L_SRC_1:%.*]] = load i32, ptr [[GEP_SRC_1]], align 4
; CHECK-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 2
; CHECK-NEXT:    [[L_SRC_2:%.*]] = load i32, ptr [[GEP_SRC_2]], align 4
; CHECK-NEXT:    [[ADD_0:%.*]] = add i32 [[L_SRC_0]], [[L_SRC_1]]
; CHECK-NEXT:    [[ADD_1:%.*]] = add i32 [[ADD_0]], [[L_SRC_2]]
; CHECK-NEXT:    ret i32 [[ADD_1]]
;
  %gep.src.0 = getelementptr inbounds i32, ptr %src, i32 0
  %l.src.0 = load i32, ptr %gep.src.0, align 4
  %gep.src.1 = getelementptr inbounds i32, ptr %src, i32 1
  %l.src.1 = load i32, ptr %gep.src.1, align 4
  %gep.src.2 = getelementptr inbounds i32, ptr %src, i32 2
  %l.src.2 = load i32, ptr %gep.src.2, align 4

  %add.0 = add i32 %l.src.0, %l.src.1
  %add.1 = add i32 %add.0, %l.src.2
  ret i32 %add.1
}

define float @reduce_fadd(ptr %src) {
; NON-POW2-LABEL: @reduce_fadd(
; NON-POW2-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds float, ptr [[SRC:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x float>, ptr [[GEP_SRC_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = call fast float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP1]])
; NON-POW2-NEXT:    ret float [[TMP2]]
;
; POW2-ONLY-LABEL: @reduce_fadd(
; POW2-ONLY-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds float, ptr [[SRC:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_SRC_0:%.*]] = load float, ptr [[GEP_SRC_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_1:%.*]] = getelementptr inbounds float, ptr [[SRC]], i32 1
; POW2-ONLY-NEXT:    [[L_SRC_1:%.*]] = load float, ptr [[GEP_SRC_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds float, ptr [[SRC]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2:%.*]] = load float, ptr [[GEP_SRC_2]], align 4
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = fadd fast float [[L_SRC_0]], [[L_SRC_1]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = fadd fast float [[ADD_0]], [[L_SRC_2]]
; POW2-ONLY-NEXT:    ret float [[ADD_1]]
;
  %gep.src.0 = getelementptr inbounds float, ptr %src, i32 0
  %l.src.0 = load float, ptr %gep.src.0, align 4
  %gep.src.1 = getelementptr inbounds float, ptr %src, i32 1
  %l.src.1 = load float, ptr %gep.src.1, align 4
  %gep.src.2 = getelementptr inbounds float, ptr %src, i32 2
  %l.src.2 = load float, ptr %gep.src.2, align 4

  %add.0 = fadd fast float %l.src.0, %l.src.1
  %add.1 = fadd fast float %add.0, %l.src.2
  ret float %add.1
}

define i32 @reduce_add_after_mul(ptr %src) {
; NON-POW2-LABEL: @reduce_add_after_mul(
; NON-POW2-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x i32>, ptr [[GEP_SRC_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = mul nsw <3 x i32> [[TMP1]], splat (i32 10)
; NON-POW2-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v3i32(<3 x i32> [[TMP2]])
; NON-POW2-NEXT:    ret i32 [[TMP3]]
;
; POW2-ONLY-LABEL: @reduce_add_after_mul(
; POW2-ONLY-NEXT:    [[GEP_SRC_0:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_SRC_0:%.*]] = load i32, ptr [[GEP_SRC_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_1:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 1
; POW2-ONLY-NEXT:    [[L_SRC_1:%.*]] = load i32, ptr [[GEP_SRC_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 2
; POW2-ONLY-NEXT:    [[L_SRC_2:%.*]] = load i32, ptr [[GEP_SRC_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_0:%.*]] = mul nsw i32 [[L_SRC_0]], 10
; POW2-ONLY-NEXT:    [[MUL_1:%.*]] = mul nsw i32 [[L_SRC_1]], 10
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_SRC_2]], 10
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = add i32 [[MUL_0]], [[MUL_1]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = add i32 [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret i32 [[ADD_1]]
;
  %gep.src.0 = getelementptr inbounds i32, ptr %src, i32 0
  %l.src.0 = load i32, ptr %gep.src.0, align 4
  %gep.src.1 = getelementptr inbounds i32, ptr %src, i32 1
  %l.src.1 = load i32, ptr %gep.src.1, align 4
  %gep.src.2 = getelementptr inbounds i32, ptr %src, i32 2
  %l.src.2 = load i32, ptr %gep.src.2, align 4

  %mul.0 = mul nsw i32 %l.src.0, 10
  %mul.1 = mul nsw i32 %l.src.1, 10
  %mul.2 = mul nsw i32 %l.src.2, 10

  %add.0 = add i32 %mul.0, %mul.1
  %add.1 = add i32 %add.0, %mul.2
  ret i32 %add.1
}

define i32 @dot_product_i32(ptr %a, ptr %b) {
; NON-POW2-LABEL: @dot_product_i32(
; NON-POW2-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x i32>, ptr [[GEP_A_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = load <3 x i32>, ptr [[GEP_B_0]], align 4
; NON-POW2-NEXT:    [[TMP3:%.*]] = mul nsw <3 x i32> [[TMP1]], [[TMP2]]
; NON-POW2-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.v3i32(<3 x i32> [[TMP3]])
; NON-POW2-NEXT:    ret i32 [[TMP4]]
;
; POW2-ONLY-LABEL: @dot_product_i32(
; POW2-ONLY-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_A_0:%.*]] = load i32, ptr [[GEP_A_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_A_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 1
; POW2-ONLY-NEXT:    [[L_A_1:%.*]] = load i32, ptr [[GEP_A_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 2
; POW2-ONLY-NEXT:    [[L_A_2:%.*]] = load i32, ptr [[GEP_A_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_B_0:%.*]] = load i32, ptr [[GEP_B_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_1:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 1
; POW2-ONLY-NEXT:    [[L_B_1:%.*]] = load i32, ptr [[GEP_B_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_2:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 2
; POW2-ONLY-NEXT:    [[L_B_2:%.*]] = load i32, ptr [[GEP_B_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_0:%.*]] = mul nsw i32 [[L_A_0]], [[L_B_0]]
; POW2-ONLY-NEXT:    [[MUL_1:%.*]] = mul nsw i32 [[L_A_1]], [[L_B_1]]
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_A_2]], [[L_B_2]]
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = add i32 [[MUL_0]], [[MUL_1]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = add i32 [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret i32 [[ADD_1]]
;
  %gep.a.0 = getelementptr inbounds i32, ptr %a, i32 0
  %l.a.0 = load i32, ptr %gep.a.0, align 4
  %gep.a.1 = getelementptr inbounds i32, ptr %a, i32 1
  %l.a.1 = load i32, ptr %gep.a.1, align 4
  %gep.a.2 = getelementptr inbounds i32, ptr %a, i32 2
  %l.a.2 = load i32, ptr %gep.a.2, align 4

  %gep.b.0 = getelementptr inbounds i32, ptr %b, i32 0
  %l.b.0 = load i32, ptr %gep.b.0, align 4
  %gep.b.1 = getelementptr inbounds i32, ptr %b, i32 1
  %l.b.1 = load i32, ptr %gep.b.1, align 4
  %gep.b.2 = getelementptr inbounds i32, ptr %b, i32 2
  %l.b.2 = load i32, ptr %gep.b.2, align 4

  %mul.0 = mul nsw i32 %l.a.0, %l.b.0
  %mul.1 = mul nsw i32 %l.a.1, %l.b.1
  %mul.2 = mul nsw i32 %l.a.2, %l.b.2

  %add.0 = add i32 %mul.0, %mul.1
  %add.1 = add i32 %add.0, %mul.2
  ret i32 %add.1
}

; Same as above, except the reduction order has been perturbed.  This
; is checking for our ability to reorder.
define i32 @dot_product_i32_reorder(ptr %a, ptr %b) {
; NON-POW2-LABEL: @dot_product_i32_reorder(
; NON-POW2-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x i32>, ptr [[GEP_A_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = load <3 x i32>, ptr [[GEP_B_0]], align 4
; NON-POW2-NEXT:    [[TMP3:%.*]] = mul nsw <3 x i32> [[TMP1]], [[TMP2]]
; NON-POW2-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.v3i32(<3 x i32> [[TMP3]])
; NON-POW2-NEXT:    ret i32 [[TMP4]]
;
; POW2-ONLY-LABEL: @dot_product_i32_reorder(
; POW2-ONLY-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_A_0:%.*]] = load i32, ptr [[GEP_A_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_A_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 1
; POW2-ONLY-NEXT:    [[L_A_1:%.*]] = load i32, ptr [[GEP_A_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 2
; POW2-ONLY-NEXT:    [[L_A_2:%.*]] = load i32, ptr [[GEP_A_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 0
; POW2-ONLY-NEXT:    [[L_B_0:%.*]] = load i32, ptr [[GEP_B_0]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_1:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 1
; POW2-ONLY-NEXT:    [[L_B_1:%.*]] = load i32, ptr [[GEP_B_1]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_2:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 2
; POW2-ONLY-NEXT:    [[L_B_2:%.*]] = load i32, ptr [[GEP_B_2]], align 4
; POW2-ONLY-NEXT:    [[MUL_0:%.*]] = mul nsw i32 [[L_A_0]], [[L_B_0]]
; POW2-ONLY-NEXT:    [[MUL_1:%.*]] = mul nsw i32 [[L_A_1]], [[L_B_1]]
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = mul nsw i32 [[L_A_2]], [[L_B_2]]
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = add i32 [[MUL_1]], [[MUL_0]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = add i32 [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret i32 [[ADD_1]]
;
  %gep.a.0 = getelementptr inbounds i32, ptr %a, i32 0
  %l.a.0 = load i32, ptr %gep.a.0, align 4
  %gep.a.1 = getelementptr inbounds i32, ptr %a, i32 1
  %l.a.1 = load i32, ptr %gep.a.1, align 4
  %gep.a.2 = getelementptr inbounds i32, ptr %a, i32 2
  %l.a.2 = load i32, ptr %gep.a.2, align 4

  %gep.b.0 = getelementptr inbounds i32, ptr %b, i32 0
  %l.b.0 = load i32, ptr %gep.b.0, align 4
  %gep.b.1 = getelementptr inbounds i32, ptr %b, i32 1
  %l.b.1 = load i32, ptr %gep.b.1, align 4
  %gep.b.2 = getelementptr inbounds i32, ptr %b, i32 2
  %l.b.2 = load i32, ptr %gep.b.2, align 4

  %mul.0 = mul nsw i32 %l.a.0, %l.b.0
  %mul.1 = mul nsw i32 %l.a.1, %l.b.1
  %mul.2 = mul nsw i32 %l.a.2, %l.b.2

  %add.0 = add i32 %mul.1, %mul.0
  %add.1 = add i32 %add.0, %mul.2
  ret i32 %add.1
}

define float @dot_product_fp32(ptr %a, ptr %b) {
; NON-POW2-LABEL: @dot_product_fp32(
; NON-POW2-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x float>, ptr [[GEP_A_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = load <3 x float>, ptr [[GEP_B_0]], align 4
; NON-POW2-NEXT:    [[TMP3:%.*]] = fmul fast <3 x float> [[TMP1]], [[TMP2]]
; NON-POW2-NEXT:    [[TMP4:%.*]] = call fast float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP3]])
; NON-POW2-NEXT:    ret float [[TMP4]]
;
; POW2-ONLY-LABEL: @dot_product_fp32(
; POW2-ONLY-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds float, ptr [[A]], i32 2
; POW2-ONLY-NEXT:    [[L_A_2:%.*]] = load float, ptr [[GEP_A_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_B_2:%.*]] = getelementptr inbounds float, ptr [[B]], i32 2
; POW2-ONLY-NEXT:    [[L_B_2:%.*]] = load float, ptr [[GEP_B_2]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[GEP_A_0]], align 4
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[GEP_B_0]], align 4
; POW2-ONLY-NEXT:    [[TMP3:%.*]] = fmul fast <2 x float> [[TMP1]], [[TMP2]]
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = fmul fast float [[L_A_2]], [[L_B_2]]
; POW2-ONLY-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; POW2-ONLY-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = fadd fast float [[TMP4]], [[TMP5]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = fadd fast float [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret float [[ADD_1]]
;
  %gep.a.0 = getelementptr inbounds float, ptr %a, i32 0
  %l.a.0 = load float, ptr %gep.a.0, align 4
  %gep.a.1 = getelementptr inbounds float, ptr %a, i32 1
  %l.a.1 = load float, ptr %gep.a.1, align 4
  %gep.a.2 = getelementptr inbounds float, ptr %a, i32 2
  %l.a.2 = load float, ptr %gep.a.2, align 4

  %gep.b.0 = getelementptr inbounds float, ptr %b, i32 0
  %l.b.0 = load float, ptr %gep.b.0, align 4
  %gep.b.1 = getelementptr inbounds float, ptr %b, i32 1
  %l.b.1 = load float, ptr %gep.b.1, align 4
  %gep.b.2 = getelementptr inbounds float, ptr %b, i32 2
  %l.b.2 = load float, ptr %gep.b.2, align 4

  %mul.0 = fmul fast float %l.a.0, %l.b.0
  %mul.1 = fmul fast float %l.a.1, %l.b.1
  %mul.2 = fmul fast float %l.a.2, %l.b.2

  %add.0 = fadd fast float %mul.0, %mul.1
  %add.1 = fadd fast float %add.0, %mul.2
  ret float %add.1
}

; Same as above, except the reduction order has been perturbed.  This
; is checking for our ability to reorder.
define float @dot_product_fp32_reorder(ptr %a, ptr %b) {
; NON-POW2-LABEL: @dot_product_fp32_reorder(
; NON-POW2-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x float>, ptr [[GEP_A_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = load <3 x float>, ptr [[GEP_B_0]], align 4
; NON-POW2-NEXT:    [[TMP3:%.*]] = fmul fast <3 x float> [[TMP1]], [[TMP2]]
; NON-POW2-NEXT:    [[TMP4:%.*]] = call fast float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP3]])
; NON-POW2-NEXT:    ret float [[TMP4]]
;
; POW2-ONLY-LABEL: @dot_product_fp32_reorder(
; POW2-ONLY-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds float, ptr [[A]], i32 2
; POW2-ONLY-NEXT:    [[L_A_2:%.*]] = load float, ptr [[GEP_A_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_B_2:%.*]] = getelementptr inbounds float, ptr [[B]], i32 2
; POW2-ONLY-NEXT:    [[L_B_2:%.*]] = load float, ptr [[GEP_B_2]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = load <2 x float>, ptr [[GEP_A_0]], align 4
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = load <2 x float>, ptr [[GEP_B_0]], align 4
; POW2-ONLY-NEXT:    [[TMP3:%.*]] = fmul fast <2 x float> [[TMP1]], [[TMP2]]
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = fmul fast float [[L_A_2]], [[L_B_2]]
; POW2-ONLY-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP3]], i32 0
; POW2-ONLY-NEXT:    [[TMP5:%.*]] = extractelement <2 x float> [[TMP3]], i32 1
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = fadd fast float [[TMP5]], [[TMP4]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = fadd fast float [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret float [[ADD_1]]
;
  %gep.a.0 = getelementptr inbounds float, ptr %a, i32 0
  %l.a.0 = load float, ptr %gep.a.0, align 4
  %gep.a.1 = getelementptr inbounds float, ptr %a, i32 1
  %l.a.1 = load float, ptr %gep.a.1, align 4
  %gep.a.2 = getelementptr inbounds float, ptr %a, i32 2
  %l.a.2 = load float, ptr %gep.a.2, align 4

  %gep.b.0 = getelementptr inbounds float, ptr %b, i32 0
  %l.b.0 = load float, ptr %gep.b.0, align 4
  %gep.b.1 = getelementptr inbounds float, ptr %b, i32 1
  %l.b.1 = load float, ptr %gep.b.1, align 4
  %gep.b.2 = getelementptr inbounds float, ptr %b, i32 2
  %l.b.2 = load float, ptr %gep.b.2, align 4

  %mul.0 = fmul fast float %l.a.0, %l.b.0
  %mul.1 = fmul fast float %l.a.1, %l.b.1
  %mul.2 = fmul fast float %l.a.2, %l.b.2

  %add.0 = fadd fast float %mul.1, %mul.0
  %add.1 = fadd fast float %add.0, %mul.2
  ret float %add.1
}


define double @dot_product_fp64(ptr %a, ptr %b) {
; NON-POW2-LABEL: @dot_product_fp64(
; NON-POW2-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds double, ptr [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds double, ptr [[B:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP1:%.*]] = load <3 x double>, ptr [[GEP_A_0]], align 4
; NON-POW2-NEXT:    [[TMP2:%.*]] = load <3 x double>, ptr [[GEP_B_0]], align 4
; NON-POW2-NEXT:    [[TMP3:%.*]] = fmul fast <3 x double> [[TMP1]], [[TMP2]]
; NON-POW2-NEXT:    [[TMP4:%.*]] = call fast double @llvm.vector.reduce.fadd.v3f64(double 0.000000e+00, <3 x double> [[TMP3]])
; NON-POW2-NEXT:    ret double [[TMP4]]
;
; POW2-ONLY-LABEL: @dot_product_fp64(
; POW2-ONLY-NEXT:    [[GEP_A_0:%.*]] = getelementptr inbounds double, ptr [[A:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_A_2:%.*]] = getelementptr inbounds double, ptr [[A]], i32 2
; POW2-ONLY-NEXT:    [[L_A_2:%.*]] = load double, ptr [[GEP_A_2]], align 4
; POW2-ONLY-NEXT:    [[GEP_B_0:%.*]] = getelementptr inbounds double, ptr [[B:%.*]], i32 0
; POW2-ONLY-NEXT:    [[GEP_B_2:%.*]] = getelementptr inbounds double, ptr [[B]], i32 2
; POW2-ONLY-NEXT:    [[L_B_2:%.*]] = load double, ptr [[GEP_B_2]], align 4
; POW2-ONLY-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[GEP_A_0]], align 4
; POW2-ONLY-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[GEP_B_0]], align 4
; POW2-ONLY-NEXT:    [[TMP3:%.*]] = fmul fast <2 x double> [[TMP1]], [[TMP2]]
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = fmul fast double [[L_A_2]], [[L_B_2]]
; POW2-ONLY-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
; POW2-ONLY-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = fadd fast double [[TMP4]], [[TMP5]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = fadd fast double [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret double [[ADD_1]]
;
  %gep.a.0 = getelementptr inbounds double, ptr %a, i32 0
  %l.a.0 = load double, ptr %gep.a.0, align 4
  %gep.a.1 = getelementptr inbounds double, ptr %a, i32 1
  %l.a.1 = load double, ptr %gep.a.1, align 4
  %gep.a.2 = getelementptr inbounds double, ptr %a, i32 2
  %l.a.2 = load double, ptr %gep.a.2, align 4

  %gep.b.0 = getelementptr inbounds double, ptr %b, i32 0
  %l.b.0 = load double, ptr %gep.b.0, align 4
  %gep.b.1 = getelementptr inbounds double, ptr %b, i32 1
  %l.b.1 = load double, ptr %gep.b.1, align 4
  %gep.b.2 = getelementptr inbounds double, ptr %b, i32 2
  %l.b.2 = load double, ptr %gep.b.2, align 4

  %mul.0 = fmul fast double %l.a.0, %l.b.0
  %mul.1 = fmul fast double %l.a.1, %l.b.1
  %mul.2 = fmul fast double %l.a.2, %l.b.2

  %add.0 = fadd fast double %mul.0, %mul.1
  %add.1 = fadd fast double %add.0, %mul.2
  ret double %add.1
}

;; Covers a case where SLP would previous crash due to a
;; missing bailout in TryToFindDuplicates for the case
;; where a VL=3 list was vectorized directly (without
;; a root instruction such as a store or reduce).
define double @no_root_reshuffle(ptr  %ptr) {
; CHECK-LABEL: @no_root_reshuffle(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = load double, ptr [[PTR:%.*]], align 8
; CHECK-NEXT:    [[MUL:%.*]] = fmul fast double [[TMP0]], [[TMP0]]
; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 8
; CHECK-NEXT:    [[TMP1:%.*]] = load double, ptr [[ARRAYIDX2]], align 8
; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 16
; CHECK-NEXT:    [[TMP2:%.*]] = load double, ptr [[ARRAYIDX3]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = fmul fast double [[TMP2]], [[TMP2]]
; CHECK-NEXT:    [[MUL6:%.*]] = fmul fast double [[TMP3]], [[TMP1]]
; CHECK-NEXT:    [[ADD:%.*]] = fadd fast double [[MUL6]], [[MUL]]
; CHECK-NEXT:    ret double [[ADD]]
;
entry:
  %0 = load double, ptr %ptr, align 8
  %mul = fmul fast double %0, %0
  %arrayidx2 = getelementptr inbounds i8, ptr %ptr, i64 8
  %1 = load double, ptr %arrayidx2, align 8
  %arrayidx3 = getelementptr inbounds i8, ptr %ptr, i64 16
  %2 = load double, ptr %arrayidx3, align 8
  %3 = fmul fast double %2, %2
  %mul6 = fmul fast double %3, %1
  %add = fadd fast double %mul6, %mul
  ret double %add
}

define float @reduce_fadd_after_fmul_of_buildvec(float %a, float %b, float %c) {
; NON-POW2-LABEL: @reduce_fadd_after_fmul_of_buildvec(
; NON-POW2-NEXT:    [[TMP1:%.*]] = insertelement <3 x float> poison, float [[A:%.*]], i32 0
; NON-POW2-NEXT:    [[TMP2:%.*]] = insertelement <3 x float> [[TMP1]], float [[B:%.*]], i32 1
; NON-POW2-NEXT:    [[TMP3:%.*]] = insertelement <3 x float> [[TMP2]], float [[C:%.*]], i32 2
; NON-POW2-NEXT:    [[TMP4:%.*]] = fmul fast <3 x float> [[TMP3]], splat (float 1.000000e+01)
; NON-POW2-NEXT:    [[TMP5:%.*]] = call fast float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP4]])
; NON-POW2-NEXT:    ret float [[TMP5]]
;
; POW2-ONLY-LABEL: @reduce_fadd_after_fmul_of_buildvec(
; POW2-ONLY-NEXT:    [[MUL_0:%.*]] = fmul fast float [[A:%.*]], 1.000000e+01
; POW2-ONLY-NEXT:    [[MUL_1:%.*]] = fmul fast float [[B:%.*]], 1.000000e+01
; POW2-ONLY-NEXT:    [[MUL_2:%.*]] = fmul fast float [[C:%.*]], 1.000000e+01
; POW2-ONLY-NEXT:    [[ADD_0:%.*]] = fadd fast float [[MUL_0]], [[MUL_1]]
; POW2-ONLY-NEXT:    [[ADD_1:%.*]] = fadd fast float [[ADD_0]], [[MUL_2]]
; POW2-ONLY-NEXT:    ret float [[ADD_1]]
;
  %mul.0 = fmul fast float %a, 10.0
  %mul.1 = fmul fast float %b, 10.0
  %mul.2 = fmul fast float %c, 10.0

  %add.0 = fadd fast float %mul.0, %mul.1
  %add.1 = fadd fast float %add.0, %mul.2
  ret float %add.1
}


declare float @llvm.fmuladd.f32(float, float, float)

declare double @llvm.fmuladd.f64(double, double, double)