path: root/llvm/test/CodeGen/AArch64/machine-combiner.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-STD
; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 -enable-unsafe-fp-math < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UNSAFE

; Incremental updates of the instruction depths should be enough for this test
; case.
; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 -enable-unsafe-fp-math \
; RUN:     -machine-combiner-inc-threshold=0 -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UNSAFE

; Verify that the first two adds are independent regardless of how the inputs are
; commuted. The destination registers are used as source registers for the third add.

define float @reassociate_adds1(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds1:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s0, s2
; CHECK-STD-NEXT:    fadd s0, s0, s3
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds1:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s2, s3
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd float %x0, %x1
  %t1 = fadd float %t0, %x2
  %t2 = fadd float %t1, %x3
  ret float %t2
}

define float @reassociate_adds1_fast(float %x0, float %x1, float %x2, float %x3) {
; CHECK-LABEL: reassociate_adds1_fast:
; CHECK:       // %bb.0:
; CHECK-NEXT:    fadd s0, s0, s1
; CHECK-NEXT:    fadd s1, s2, s3
; CHECK-NEXT:    fadd s0, s0, s1
; CHECK-NEXT:    ret
  %t0 = fadd fast float %x0, %x1
  %t1 = fadd fast float %t0, %x2
  %t2 = fadd fast float %t1, %x3
  ret float %t2
}

define float @reassociate_adds1_reassoc(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds1_reassoc:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s0, s2
; CHECK-STD-NEXT:    fadd s0, s0, s3
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds1_reassoc:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s2, s3
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd reassoc float %x0, %x1
  %t1 = fadd reassoc float %t0, %x2
  %t2 = fadd reassoc float %t1, %x3
  ret float %t2
}

define float @reassociate_adds2(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds2:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s2, s0
; CHECK-STD-NEXT:    fadd s0, s0, s3
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds2:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s2, s3
; CHECK-UNSAFE-NEXT:    fadd s0, s1, s0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd float %x0, %x1
  %t1 = fadd float %x2, %t0
  %t2 = fadd float %t1, %x3
  ret float %t2
}

define float @reassociate_adds3(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds3:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s0, s2
; CHECK-STD-NEXT:    fadd s0, s3, s0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds3:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s3, s2
; CHECK-UNSAFE-NEXT:    fadd s0, s1, s0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd float %x0, %x1
  %t1 = fadd float %t0, %x2
  %t2 = fadd float %x3, %t1
  ret float %t2
}

define float @reassociate_adds4(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds4:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s2, s0
; CHECK-STD-NEXT:    fadd s0, s3, s0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds4:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s3, s2
; CHECK-UNSAFE-NEXT:    fadd s0, s1, s0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd float %x0, %x1
  %t1 = fadd float %x2, %t0
  %t2 = fadd float %x3, %t1
  ret float %t2
}

; Verify that we reassociate some of these ops. The optimal balanced tree of adds is not
; produced because that would cost more compile time.

define float @reassociate_adds5(float %x0, float %x1, float %x2, float %x3, float %x4, float %x5, float %x6, float %x7) {
; CHECK-STD-LABEL: reassociate_adds5:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s0, s2
; CHECK-STD-NEXT:    fadd s0, s0, s3
; CHECK-STD-NEXT:    fadd s0, s0, s4
; CHECK-STD-NEXT:    fadd s0, s0, s5
; CHECK-STD-NEXT:    fadd s0, s0, s6
; CHECK-STD-NEXT:    fadd s0, s0, s7
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds5:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s2, s3
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s4, s5
; CHECK-UNSAFE-NEXT:    fadd s1, s1, s6
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s0, s0, s7
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd float %x0, %x1
  %t1 = fadd float %t0, %x2
  %t2 = fadd float %t1, %x3
  %t3 = fadd float %t2, %x4
  %t4 = fadd float %t3, %x5
  %t5 = fadd float %t4, %x6
  %t6 = fadd float %t5, %x7
  ret float %t6
}

; Verify that we only need two associative operations to reassociate the operands.
; Also, we should reassociate such that the result of the high latency division
; is used by the final 'add' rather than reassociating the %x3 operand with the
; division. The latter reassociation would not improve anything.

define float @reassociate_adds6(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_adds6:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv s0, s0, s1
; CHECK-STD-NEXT:    fadd s0, s2, s0
; CHECK-STD-NEXT:    fadd s0, s3, s0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds6:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv s0, s0, s1
; CHECK-UNSAFE-NEXT:    fadd s1, s3, s2
; CHECK-UNSAFE-NEXT:    fadd s0, s1, s0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv float %x0, %x1
  %t1 = fadd float %x2, %t0
  %t2 = fadd float %x3, %t1
  ret float %t2
}

; Verify that scalar single-precision multiplies are reassociated.

define float @reassociate_muls1(float %x0, float %x1, float %x2, float %x3) {
; CHECK-STD-LABEL: reassociate_muls1:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv s0, s0, s1
; CHECK-STD-NEXT:    fmul s0, s2, s0
; CHECK-STD-NEXT:    fmul s0, s3, s0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls1:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv s0, s0, s1
; CHECK-UNSAFE-NEXT:    fmul s1, s3, s2
; CHECK-UNSAFE-NEXT:    fmul s0, s1, s0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv float %x0, %x1
  %t1 = fmul float %x2, %t0
  %t2 = fmul float %x3, %t1
  ret float %t2
}

; Verify that scalar double-precision adds are reassociated.

define double @reassociate_adds_double(double %x0, double %x1, double %x2, double %x3) {
; CHECK-STD-LABEL: reassociate_adds_double:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv d0, d0, d1
; CHECK-STD-NEXT:    fadd d0, d2, d0
; CHECK-STD-NEXT:    fadd d0, d3, d0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_double:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv d0, d0, d1
; CHECK-UNSAFE-NEXT:    fadd d1, d3, d2
; CHECK-UNSAFE-NEXT:    fadd d0, d1, d0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv double %x0, %x1
  %t1 = fadd double %x2, %t0
  %t2 = fadd double %x3, %t1
  ret double %t2
}

; Verify that scalar double-precision multiplies are reassociated.

define double @reassociate_muls_double(double %x0, double %x1, double %x2, double %x3) {
; CHECK-STD-LABEL: reassociate_muls_double:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv d0, d0, d1
; CHECK-STD-NEXT:    fmul d0, d2, d0
; CHECK-STD-NEXT:    fmul d0, d3, d0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_double:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv d0, d0, d1
; CHECK-UNSAFE-NEXT:    fmul d1, d3, d2
; CHECK-UNSAFE-NEXT:    fmul d0, d1, d0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv double %x0, %x1
  %t1 = fmul double %x2, %t0
  %t2 = fmul double %x3, %t1
  ret double %t2
}

; Verify that scalar half-precision adds are reassociated.

define half @reassociate_adds_half(half %x0, half %x1, half %x2, half %x3) {
; CHECK-STD-LABEL: reassociate_adds_half:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv h0, h0, h1
; CHECK-STD-NEXT:    fadd h0, h2, h0
; CHECK-STD-NEXT:    fadd h0, h3, h0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_half:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv h0, h0, h1
; CHECK-UNSAFE-NEXT:    fadd h1, h3, h2
; CHECK-UNSAFE-NEXT:    fadd h0, h1, h0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv half %x0, %x1
  %t1 = fadd half %x2, %t0
  %t2 = fadd half %x3, %t1
  ret half %t2
}

; Verify that scalar half-precision multiplies are reassociated.

define half @reassociate_muls_half(half %x0, half %x1, half %x2, half %x3) {
; CHECK-STD-LABEL: reassociate_muls_half:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fdiv h0, h0, h1
; CHECK-STD-NEXT:    fmul h0, h2, h0
; CHECK-STD-NEXT:    fmul h0, h3, h0
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_half:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fdiv h0, h0, h1
; CHECK-UNSAFE-NEXT:    fmul h1, h3, h2
; CHECK-UNSAFE-NEXT:    fmul h0, h1, h0
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fdiv half %x0, %x1
  %t1 = fmul half %x2, %t0
  %t2 = fmul half %x3, %t1
  ret half %t2
}

; Verify that scalar integer adds are reassociated.

define i32 @reassociate_adds_i32(i32 %x0, i32 %x1, i32 %x2, i32 %x3) {
; CHECK-LABEL: reassociate_adds_i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    udiv w8, w0, w1
; CHECK-NEXT:    add w9, w3, w2
; CHECK-NEXT:    add w0, w9, w8
; CHECK-NEXT:    ret
  %t0 = udiv i32 %x0, %x1
  %t1 = add i32 %x2, %t0
  %t2 = add i32 %x3, %t1
  ret i32 %t2
}

define i64 @reassociate_adds_i64(i64 %x0, i64 %x1, i64 %x2, i64 %x3) {
; CHECK-LABEL: reassociate_adds_i64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    udiv x8, x0, x1
; CHECK-NEXT:    add x9, x3, x2
; CHECK-NEXT:    add x0, x9, x8
; CHECK-NEXT:    ret
  %t0 = udiv i64 %x0, %x1
  %t1 = add i64 %x2, %t0
  %t2 = add i64 %x3, %t1
  ret i64 %t2
}

; Verify that scalar bitwise operations are reassociated.

define i32 @reassociate_ands_i32(i32 %x0, i32 %x1, i32 %x2, i32 %x3) {
; CHECK-LABEL: reassociate_ands_i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    and w8, w0, w1
; CHECK-NEXT:    and w9, w2, w3
; CHECK-NEXT:    and w0, w8, w9
; CHECK-NEXT:    ret
  %t0 = and i32 %x0, %x1
  %t1 = and i32 %t0, %x2
  %t2 = and i32 %t1, %x3
  ret i32 %t2
}

define i64 @reassociate_ors_i64(i64 %x0, i64 %x1, i64 %x2, i64 %x3) {
; CHECK-LABEL: reassociate_ors_i64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    orr x8, x0, x1
; CHECK-NEXT:    orr x9, x2, x3
; CHECK-NEXT:    orr x0, x8, x9
; CHECK-NEXT:    ret
  %t0 = or i64 %x0, %x1
  %t1 = or i64 %t0, %x2
  %t2 = or i64 %t1, %x3
  ret i64 %t2
}

define i32 @reassociate_xors_i32(i32 %x0, i32 %x1, i32 %x2, i32 %x3) {
; CHECK-LABEL: reassociate_xors_i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    eor w8, w0, w1
; CHECK-NEXT:    eor w9, w2, w3
; CHECK-NEXT:    eor w0, w8, w9
; CHECK-NEXT:    ret
  %t0 = xor i32 %x0, %x1
  %t1 = xor i32 %t0, %x2
  %t2 = xor i32 %t1, %x3
  ret i32 %t2
}

; Verify that we reassociate vector instructions too.

define <4 x float> @vector_reassociate_adds1(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
; CHECK-STD-LABEL: vector_reassociate_adds1:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v2.4s
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v3.4s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: vector_reassociate_adds1:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    fadd v1.4s, v2.4s, v3.4s
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x float> %x0, %x1
  %t1 = fadd <4 x float> %t0, %x2
  %t2 = fadd <4 x float> %t1, %x3
  ret <4 x float> %t2
}

define <4 x float> @vector_reassociate_adds2(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
; CHECK-STD-LABEL: vector_reassociate_adds2:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-STD-NEXT:    fadd v0.4s, v2.4s, v0.4s
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v3.4s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: vector_reassociate_adds2:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    fadd v1.4s, v2.4s, v3.4s
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v1.4s, v0.4s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x float> %x0, %x1
  %t1 = fadd <4 x float> %x2, %t0
  %t2 = fadd <4 x float> %t1, %x3
  ret <4 x float> %t2
}

define <4 x float> @vector_reassociate_adds3(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
; CHECK-STD-LABEL: vector_reassociate_adds3:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v2.4s
; CHECK-STD-NEXT:    fadd v0.4s, v3.4s, v0.4s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: vector_reassociate_adds3:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    fadd v1.4s, v3.4s, v2.4s
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v1.4s, v0.4s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x float> %x0, %x1
  %t1 = fadd <4 x float> %t0, %x2
  %t2 = fadd <4 x float> %x3, %t1
  ret <4 x float> %t2
}

define <4 x float> @vector_reassociate_adds4(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
; CHECK-STD-LABEL: vector_reassociate_adds4:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-STD-NEXT:    fadd v0.4s, v2.4s, v0.4s
; CHECK-STD-NEXT:    fadd v0.4s, v3.4s, v0.4s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: vector_reassociate_adds4:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    fadd v1.4s, v3.4s, v2.4s
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v1.4s, v0.4s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x float> %x0, %x1
  %t1 = fadd <4 x float> %x2, %t0
  %t2 = fadd <4 x float> %x3, %t1
  ret <4 x float> %t2
}

; Verify that 64-bit vector half-precision adds are reassociated.

define <4 x half> @reassociate_adds_v4f16(<4 x half> %x0, <4 x half> %x1, <4 x half> %x2, <4 x half> %x3) {
; CHECK-STD-LABEL: reassociate_adds_v4f16:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4h, v0.4h, v1.4h
; CHECK-STD-NEXT:    fadd v0.4h, v2.4h, v0.4h
; CHECK-STD-NEXT:    fadd v0.4h, v3.4h, v0.4h
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_v4f16:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4h, v0.4h, v1.4h
; CHECK-UNSAFE-NEXT:    fadd v1.4h, v3.4h, v2.4h
; CHECK-UNSAFE-NEXT:    fadd v0.4h, v1.4h, v0.4h
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x half> %x0, %x1
  %t1 = fadd <4 x half> %x2, %t0
  %t2 = fadd <4 x half> %x3, %t1
  ret <4 x half> %t2
}

; Verify that 128-bit vector half-precision multiplies are reassociated.

define <8 x half> @reassociate_muls_v8f16(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, <8 x half> %x3) {
; CHECK-STD-LABEL: reassociate_muls_v8f16:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.8h, v0.8h, v1.8h
; CHECK-STD-NEXT:    fmul v0.8h, v2.8h, v0.8h
; CHECK-STD-NEXT:    fmul v0.8h, v3.8h, v0.8h
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_v8f16:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.8h, v0.8h, v1.8h
; CHECK-UNSAFE-NEXT:    fmul v1.8h, v3.8h, v2.8h
; CHECK-UNSAFE-NEXT:    fmul v0.8h, v1.8h, v0.8h
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <8 x half> %x0, %x1
  %t1 = fmul <8 x half> %x2, %t0
  %t2 = fmul <8 x half> %x3, %t1
  ret <8 x half> %t2
}

; Verify that 128-bit vector single-precision multiplies are reassociated.

define <4 x float> @reassociate_muls_v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
; CHECK-STD-LABEL: reassociate_muls_v4f32:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-STD-NEXT:    fmul v0.4s, v2.4s, v0.4s
; CHECK-STD-NEXT:    fmul v0.4s, v3.4s, v0.4s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_v4f32:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.4s, v0.4s, v1.4s
; CHECK-UNSAFE-NEXT:    fmul v1.4s, v3.4s, v2.4s
; CHECK-UNSAFE-NEXT:    fmul v0.4s, v1.4s, v0.4s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <4 x float> %x0, %x1
  %t1 = fmul <4 x float> %x2, %t0
  %t2 = fmul <4 x float> %x3, %t1
  ret <4 x float> %t2
}

; Verify that 128-bit vector double-precision multiplies are reassociated.

define <2 x double> @reassociate_muls_v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, <2 x double> %x3) {
; CHECK-STD-LABEL: reassociate_muls_v2f64:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd v0.2d, v0.2d, v1.2d
; CHECK-STD-NEXT:    fmul v0.2d, v2.2d, v0.2d
; CHECK-STD-NEXT:    fmul v0.2d, v3.2d, v0.2d
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_v2f64:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd v0.2d, v0.2d, v1.2d
; CHECK-UNSAFE-NEXT:    fmul v1.2d, v3.2d, v2.2d
; CHECK-UNSAFE-NEXT:    fmul v0.2d, v1.2d, v0.2d
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd <2 x double> %x0, %x1
  %t1 = fmul <2 x double> %x2, %t0
  %t2 = fmul <2 x double> %x3, %t1
  ret <2 x double> %t2
}

; Verify that vector integer arithmetic operations are reassociated.

define <2 x i32> @reassociate_muls_v2i32(<2 x i32> %x0, <2 x i32> %x1, <2 x i32> %x2, <2 x i32> %x3) {
; CHECK-LABEL: reassociate_muls_v2i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    mul v0.2s, v0.2s, v1.2s
; CHECK-NEXT:    mul v1.2s, v3.2s, v2.2s
; CHECK-NEXT:    mul v0.2s, v1.2s, v0.2s
; CHECK-NEXT:    ret
  %t0 = mul <2 x i32> %x0, %x1
  %t1 = mul <2 x i32> %x2, %t0
  %t2 = mul <2 x i32> %x3, %t1
  ret <2 x i32> %t2
}

define <2 x i64> @reassociate_adds_v2i64(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, <2 x i64> %x3) {
; CHECK-LABEL: reassociate_adds_v2i64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    add v0.2d, v0.2d, v1.2d
; CHECK-NEXT:    add v1.2d, v3.2d, v2.2d
; CHECK-NEXT:    add v0.2d, v1.2d, v0.2d
; CHECK-NEXT:    ret
  %t0 = add <2 x i64> %x0, %x1
  %t1 = add <2 x i64> %x2, %t0
  %t2 = add <2 x i64> %x3, %t1
  ret <2 x i64> %t2
}

; Verify that vector bitwise operations are reassociated.

define <16 x i8> @reassociate_ands_v16i8(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, <16 x i8> %x3) {
; CHECK-LABEL: reassociate_ands_v16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
; CHECK-NEXT:    and v1.16b, v2.16b, v3.16b
; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-NEXT:    ret
  %t0 = or <16 x i8> %x0, %x1
  %t1 = and <16 x i8> %t0, %x2
  %t2 = and <16 x i8> %t1, %x3
  ret <16 x i8> %t2
}

define <4 x i16> @reassociate_ors_v4i16(<4 x i16> %x0, <4 x i16> %x1, <4 x i16> %x2, <4 x i16> %x3) {
; CHECK-LABEL: reassociate_ors_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    eor v0.8b, v0.8b, v1.8b
; CHECK-NEXT:    orr v1.8b, v2.8b, v3.8b
; CHECK-NEXT:    orr v0.8b, v0.8b, v1.8b
; CHECK-NEXT:    ret
  %t0 = xor <4 x i16> %x0, %x1
  %t1 = or <4 x i16> %t0, %x2
  %t2 = or <4 x i16> %t1, %x3
  ret <4 x i16> %t2
}

define <4 x i32> @reassociate_xors_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
; CHECK-LABEL: reassociate_xors_v4i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-NEXT:    // kill: def $q3 killed $q3 def $z3
; CHECK-NEXT:    // kill: def $q2 killed $q2 def $z2
; CHECK-NEXT:    eor3 z0.d, z0.d, z2.d, z3.d
; CHECK-NEXT:    // kill: def $q0 killed $q0 killed $z0
; CHECK-NEXT:    ret
  %t0 = and <4 x i32> %x0, %x1
  %t1 = xor <4 x i32> %t0, %x2
  %t2 = xor <4 x i32> %t1, %x3
  ret <4 x i32> %t2
}

; Verify that scalable vector FP arithmetic operations are reassociated.

define <vscale x 8 x half> @reassociate_adds_nxv4f16(<vscale x 8 x half> %x0, <vscale x 8 x half> %x1, <vscale x 8 x half> %x2, <vscale x 8 x half> %x3) {
; CHECK-STD-LABEL: reassociate_adds_nxv4f16:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd z0.h, z0.h, z1.h
; CHECK-STD-NEXT:    fadd z0.h, z2.h, z0.h
; CHECK-STD-NEXT:    fadd z0.h, z3.h, z0.h
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_nxv4f16:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd z0.h, z0.h, z1.h
; CHECK-UNSAFE-NEXT:    fadd z1.h, z3.h, z2.h
; CHECK-UNSAFE-NEXT:    fadd z0.h, z1.h, z0.h
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd reassoc <vscale x 8 x half> %x0, %x1
  %t1 = fadd reassoc <vscale x 8 x half> %x2, %t0
  %t2 = fadd reassoc <vscale x 8 x half> %x3, %t1
  ret <vscale x 8 x half> %t2
}

define <vscale x 4 x float> @reassociate_adds_nxv4f32(<vscale x 4 x float> %x0, <vscale x 4 x float> %x1, <vscale x 4 x float> %x2, <vscale x 4 x float> %x3) {
; CHECK-STD-LABEL: reassociate_adds_nxv4f32:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fadd z0.s, z0.s, z1.s
; CHECK-STD-NEXT:    fadd z0.s, z2.s, z0.s
; CHECK-STD-NEXT:    fadd z0.s, z3.s, z0.s
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_nxv4f32:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fadd z0.s, z0.s, z1.s
; CHECK-UNSAFE-NEXT:    fadd z1.s, z3.s, z2.s
; CHECK-UNSAFE-NEXT:    fadd z0.s, z1.s, z0.s
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fadd reassoc <vscale x 4 x float> %x0, %x1
  %t1 = fadd reassoc <vscale x 4 x float> %x2, %t0
  %t2 = fadd reassoc <vscale x 4 x float> %x3, %t1
  ret <vscale x 4 x float> %t2
}

define <vscale x 2 x double> @reassociate_muls_nxv2f64(<vscale x 2 x double> %x0, <vscale x 2 x double> %x1, <vscale x 2 x double> %x2, <vscale x 2 x double> %x3) {
; CHECK-STD-LABEL: reassociate_muls_nxv2f64:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    fmul z0.d, z0.d, z1.d
; CHECK-STD-NEXT:    fmul z0.d, z2.d, z0.d
; CHECK-STD-NEXT:    fmul z0.d, z3.d, z0.d
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_muls_nxv2f64:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    fmul z0.d, z0.d, z1.d
; CHECK-UNSAFE-NEXT:    fmul z1.d, z3.d, z2.d
; CHECK-UNSAFE-NEXT:    fmul z0.d, z1.d, z0.d
; CHECK-UNSAFE-NEXT:    ret
  %t0 = fmul reassoc <vscale x 2 x double> %x0, %x1
  %t1 = fmul reassoc <vscale x 2 x double> %x2, %t0
  %t2 = fmul reassoc <vscale x 2 x double> %x3, %t1
  ret <vscale x 2 x double> %t2
}

; Verify that scalable vector integer arithmetic operations are reassociated.

define <vscale x 16 x i8> @reassociate_muls_nxv16i8(<vscale x 16 x i8> %x0, <vscale x 16 x i8> %x1, <vscale x 16 x i8> %x2, <vscale x 16 x i8> %x3) {
; CHECK-LABEL: reassociate_muls_nxv16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    mul z0.b, z0.b, z1.b
; CHECK-NEXT:    mul z1.b, z3.b, z2.b
; CHECK-NEXT:    mul z0.b, z1.b, z0.b
; CHECK-NEXT:    ret
  %t0 = mul <vscale x 16 x i8> %x0, %x1
  %t1 = mul <vscale x 16 x i8> %x2, %t0
  %t2 = mul <vscale x 16 x i8> %x3, %t1
  ret <vscale x 16 x i8> %t2
}

define <vscale x 8 x i16> @reassociate_adds_nxv8i16(<vscale x 8 x i16> %x0, <vscale x 8 x i16> %x1, <vscale x 8 x i16> %x2, <vscale x 8 x i16> %x3) {
; CHECK-LABEL: reassociate_adds_nxv8i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    add z0.h, z0.h, z1.h
; CHECK-NEXT:    add z1.h, z3.h, z2.h
; CHECK-NEXT:    add z0.h, z1.h, z0.h
; CHECK-NEXT:    ret
  %t0 = add <vscale x 8 x i16> %x0, %x1
  %t1 = add <vscale x 8 x i16> %x2, %t0
  %t2 = add <vscale x 8 x i16> %x3, %t1
  ret <vscale x 8 x i16> %t2
}

define <vscale x 4 x i32> @reassociate_muls_nxv4i32(<vscale x 4 x i32> %x0, <vscale x 4 x i32> %x1, <vscale x 4 x i32> %x2, <vscale x 4 x i32> %x3) {
; CHECK-LABEL: reassociate_muls_nxv4i32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    mul z0.s, z0.s, z1.s
; CHECK-NEXT:    mul z1.s, z3.s, z2.s
; CHECK-NEXT:    mul z0.s, z1.s, z0.s
; CHECK-NEXT:    ret
  %t0 = mul <vscale x 4 x i32> %x0, %x1
  %t1 = mul <vscale x 4 x i32> %x2, %t0
  %t2 = mul <vscale x 4 x i32> %x3, %t1
  ret <vscale x 4 x i32> %t2
}

define <vscale x 2 x i64> @reassociate_adds_nxv2i64(<vscale x 2 x i64> %x0, <vscale x 2 x i64> %x1, <vscale x 2 x i64> %x2, <vscale x 2 x i64> %x3) {
; CHECK-LABEL: reassociate_adds_nxv2i64:
; CHECK:       // %bb.0:
; CHECK-NEXT:    add z0.d, z0.d, z1.d
; CHECK-NEXT:    add z1.d, z3.d, z2.d
; CHECK-NEXT:    add z0.d, z1.d, z0.d
; CHECK-NEXT:    ret
  %t0 = add <vscale x 2 x i64> %x0, %x1
  %t1 = add <vscale x 2 x i64> %x2, %t0
  %t2 = add <vscale x 2 x i64> %x3, %t1
  ret <vscale x 2 x i64> %t2
}

; Verify that scalable vector bitwise operations are reassociated.

define <vscale x 16 x i8> @reassociate_ands_nxv16i8(<vscale x 16 x i8> %x0, <vscale x 16 x i8> %x1, <vscale x 16 x i8> %x2, <vscale x 16 x i8> %x3) {
; CHECK-LABEL: reassociate_ands_nxv16i8:
; CHECK:       // %bb.0:
; CHECK-NEXT:    orr z0.d, z0.d, z1.d
; CHECK-NEXT:    and z1.d, z2.d, z3.d
; CHECK-NEXT:    and z0.d, z0.d, z1.d
; CHECK-NEXT:    ret
  %t0 = or <vscale x 16 x i8> %x0, %x1
  %t1 = and <vscale x 16 x i8> %t0, %x2
  %t2 = and <vscale x 16 x i8> %t1, %x3
  ret <vscale x 16 x i8> %t2
}

define <vscale x 8 x i16> @reassociate_ors_nxv8i16(<vscale x 8 x i16> %x0, <vscale x 8 x i16> %x1, <vscale x 8 x i16> %x2, <vscale x 8 x i16> %x3) {
; CHECK-LABEL: reassociate_ors_nxv8i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    eor z0.d, z0.d, z1.d
; CHECK-NEXT:    orr z1.d, z2.d, z3.d
; CHECK-NEXT:    orr z0.d, z0.d, z1.d
; CHECK-NEXT:    ret
  %t0 = xor <vscale x 8 x i16> %x0, %x1
  %t1 = or <vscale x 8 x i16> %t0, %x2
  %t2 = or <vscale x 8 x i16> %t1, %x3
  ret <vscale x 8 x i16> %t2
}

; PR25016: https://llvm.org/bugs/show_bug.cgi?id=25016
; Verify that reassociation is not happening needlessly or wrongly.

declare double @bar()

define double @reassociate_adds_from_calls() {
; CHECK-STD-LABEL: reassociate_adds_from_calls:
; CHECK-STD:       // %bb.0:
; CHECK-STD-NEXT:    str d10, [sp, #-32]! // 8-byte Folded Spill
; CHECK-STD-NEXT:    stp d9, d8, [sp, #8] // 16-byte Folded Spill
; CHECK-STD-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-STD-NEXT:    .cfi_def_cfa_offset 32
; CHECK-STD-NEXT:    .cfi_offset w30, -8
; CHECK-STD-NEXT:    .cfi_offset b8, -16
; CHECK-STD-NEXT:    .cfi_offset b9, -24
; CHECK-STD-NEXT:    .cfi_offset b10, -32
; CHECK-STD-NEXT:    bl bar
; CHECK-STD-NEXT:    fmov d8, d0
; CHECK-STD-NEXT:    bl bar
; CHECK-STD-NEXT:    fmov d9, d0
; CHECK-STD-NEXT:    bl bar
; CHECK-STD-NEXT:    fmov d10, d0
; CHECK-STD-NEXT:    bl bar
; CHECK-STD-NEXT:    fadd d1, d8, d9
; CHECK-STD-NEXT:    ldp d9, d8, [sp, #8] // 16-byte Folded Reload
; CHECK-STD-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-STD-NEXT:    fadd d1, d1, d10
; CHECK-STD-NEXT:    fadd d0, d1, d0
; CHECK-STD-NEXT:    ldr d10, [sp], #32 // 8-byte Folded Reload
; CHECK-STD-NEXT:    ret
;
; CHECK-UNSAFE-LABEL: reassociate_adds_from_calls:
; CHECK-UNSAFE:       // %bb.0:
; CHECK-UNSAFE-NEXT:    str d10, [sp, #-32]! // 8-byte Folded Spill
; CHECK-UNSAFE-NEXT:    stp d9, d8, [sp, #8] // 16-byte Folded Spill
; CHECK-UNSAFE-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-UNSAFE-NEXT:    .cfi_def_cfa_offset 32
; CHECK-UNSAFE-NEXT:    .cfi_offset w30, -8
; CHECK-UNSAFE-NEXT:    .cfi_offset b8, -16
; CHECK-UNSAFE-NEXT:    .cfi_offset b9, -24
; CHECK-UNSAFE-NEXT:    .cfi_offset b10, -32
; CHECK-UNSAFE-NEXT:    bl bar
; CHECK-UNSAFE-NEXT:    fmov d8, d0
; CHECK-UNSAFE-NEXT:    bl bar
; CHECK-UNSAFE-NEXT:    fmov d9, d0
; CHECK-UNSAFE-NEXT:    bl bar
; CHECK-UNSAFE-NEXT:    fmov d10, d0
; CHECK-UNSAFE-NEXT:    bl bar
; CHECK-UNSAFE-NEXT:    fadd d1, d8, d9
; CHECK-UNSAFE-NEXT:    ldp d9, d8, [sp, #8] // 16-byte Folded Reload
; CHECK-UNSAFE-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-UNSAFE-NEXT:    fadd d0, d10, d0
; CHECK-UNSAFE-NEXT:    fadd d0, d1, d0
; CHECK-UNSAFE-NEXT:    ldr d10, [sp], #32 // 8-byte Folded Reload
; CHECK-UNSAFE-NEXT:    ret
  %x0 = call double @bar()
  %x1 = call double @bar()
  %x2 = call double @bar()
  %x3 = call double @bar()
  %t0 = fadd double %x0, %x1
  %t1 = fadd double %t0, %x2
  %t2 = fadd double %t1, %x3
  ret double %t2
}

define double @already_reassociated() {
; CHECK-LABEL: already_reassociated:
; CHECK:       // %bb.0:
; CHECK-NEXT:    str d10, [sp, #-32]! // 8-byte Folded Spill
; CHECK-NEXT:    stp d9, d8, [sp, #8] // 16-byte Folded Spill
; CHECK-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -8
; CHECK-NEXT:    .cfi_offset b8, -16
; CHECK-NEXT:    .cfi_offset b9, -24
; CHECK-NEXT:    .cfi_offset b10, -32
; CHECK-NEXT:    bl bar
; CHECK-NEXT:    fmov d8, d0
; CHECK-NEXT:    bl bar
; CHECK-NEXT:    fmov d9, d0
; CHECK-NEXT:    bl bar
; CHECK-NEXT:    fmov d10, d0
; CHECK-NEXT:    bl bar
; CHECK-NEXT:    fadd d1, d8, d9
; CHECK-NEXT:    ldp d9, d8, [sp, #8] // 16-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-NEXT:    fadd d0, d10, d0
; CHECK-NEXT:    fadd d0, d1, d0
; CHECK-NEXT:    ldr d10, [sp], #32 // 8-byte Folded Reload
; CHECK-NEXT:    ret
  %x0 = call double @bar()
  %x1 = call double @bar()
  %x2 = call double @bar()
  %x3 = call double @bar()
  %t0 = fadd double %x0, %x1
  %t1 = fadd double %x2, %x3
  %t2 = fadd double %t0, %t1
  ret double %t2
}