llvm/test/CodeGen/AArch64/sink-fneg.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64 < %s | FileCheck %s


define void @shared_fneg_across_bbs(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_across_bbs:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v1.4s, v0.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB0_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v3.4s, v0.4s
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB0_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %r1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %y, <4 x float> %neg, <4 x float> %z)
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %r2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %neg, <4 x float> %b)
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define void @shared_fnegs_across_bbs(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fnegs_across_bbs:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmla v2.4s, v0.4s, v3.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB1_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmla v4.4s, v0.4s, v3.4s
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB1_2: // %exit
; CHECK-NEXT:    ret
entry:
  %negx = fneg <4 x float> %x
  %nega = fneg <4 x float> %a
  %r1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %nega, <4 x float> %negx, <4 x float> %z)
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %r2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %nega, <4 x float> %negx, <4 x float> %b)
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define <4 x float> @shared_fneg_with_other_users(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_with_other_users:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v1.4s, v0.4s
; CHECK-NEXT:    fneg v1.4s, v0.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB2_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v3.4s, v0.4s
; CHECK-NEXT:    mov v0.16b, v1.16b
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:    ret
; CHECK-NEXT:  .LBB2_2: // %other_use
; CHECK-NEXT:    sub sp, sp, #32
; CHECK-NEXT:    str x30, [sp, #16] // 8-byte Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    mov v0.16b, v1.16b
; CHECK-NEXT:    str q1, [sp] // 16-byte Spill
; CHECK-NEXT:    bl foo
; CHECK-NEXT:    ldr q0, [sp] // 16-byte Reload
; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Reload
; CHECK-NEXT:    add sp, sp, #32
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %r1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %y, <4 x float> %neg, <4 x float> %z)
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %other_use

use_bb:
  %r2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %neg, <4 x float> %b)
  store <4 x float> %r2, ptr %out2
  br label %exit

other_use:
  call void @foo(<4 x float> %neg)
  br label %exit

exit:
  ret <4 x float> %neg
}

define void @shared_fneg_across_bbs_fmuladd(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_across_bbs_fmuladd:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v1.4s, v0.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB3_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v3.4s, v0.4s
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB3_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %r1 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %y, <4 x float> %neg, <4 x float> %z)
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %r2 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %a, <4 x float> %neg, <4 x float> %b)
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define void @shared_fneg_across_bbs_fmul_fast(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_across_bbs_fmul_fast:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v0.4s, v1.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB4_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v0.4s, v3.4s
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB4_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %mul1 = fmul fast <4 x float> %y, %neg
  %r1 = fadd fast <4 x float> %mul1, %z
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %mul2 = fmul fast <4 x float> %a, %neg
  %r2 = fadd fast <4 x float> %mul2, %b
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define void @shared_fneg_across_bbs_fmul_contract(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_across_bbs_fmul_contract:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v0.4s, v1.4s
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB5_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v0.4s, v3.4s
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB5_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %mul1 = fmul contract <4 x float> %y, %neg
  %r1 = fadd contract <4 x float> %mul1, %z
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %mul2 = fmul contract <4 x float> %a, %neg
  %r2 = fadd contract <4 x float> %mul2, %b
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define void @shared_fneg_across_bbs_fmul_scalar(float %x, float %y, float %z, float %a, float %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_across_bbs_fmul_scalar:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmsub s1, s1, s0, s2
; CHECK-NEXT:    str s1, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB6_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmsub s0, s3, s0, s4
; CHECK-NEXT:    str s0, [x2]
; CHECK-NEXT:  .LBB6_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg float %x
  %mul1 = fmul fast float %y, %neg
  %r1 = fadd fast float %mul1, %z
  store float %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %mul2 = fmul fast float %a, %neg
  %r2 = fadd fast float %mul2, %b
  store float %r2, ptr %out2
  br label %exit

exit:
  ret void
}

define void @shared_fneg_splat_across_bbs_fmul(<4 x float> %x, <4 x float> %y, <4 x float> %z, <4 x float> %a, <4 x float> %b, i1 %cond, ptr %out1, ptr %out2) {
; CHECK-LABEL: shared_fneg_splat_across_bbs_fmul:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    fmls v2.4s, v0.4s, v1.s[0]
; CHECK-NEXT:    str q2, [x1]
; CHECK-NEXT:    tbz w0, #0, .LBB7_2
; CHECK-NEXT:  // %bb.1: // %use_bb
; CHECK-NEXT:    fmls v4.4s, v0.4s, v1.s[0]
; CHECK-NEXT:    str q4, [x2]
; CHECK-NEXT:  .LBB7_2: // %exit
; CHECK-NEXT:    ret
entry:
  %neg = fneg <4 x float> %x
  %splat = shufflevector <4 x float> %y, <4 x float> poison, <4 x i32> zeroinitializer
  %mul1 = fmul fast <4 x float> %splat, %neg
  %r1 = fadd fast <4 x float> %mul1, %z
  store <4 x float> %r1, ptr %out1
  br i1 %cond, label %use_bb, label %exit

use_bb:
  %mul2 = fmul fast <4 x float> %splat, %neg
  %r2 = fadd fast <4 x float> %mul2, %b
  store <4 x float> %r2, ptr %out2
  br label %exit

exit:
  ret void
}

declare void @foo(<4 x float>)