test/Transforms/InstCombine/add-mask-neg.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -S -passes=instcombine | FileCheck %s

;
; Canonicalize ((X & -X) - 1) --> ((X - 1) & ~X)
;

define i32 @dec_mask_neg_i32(i32 %X) {
; CHECK-LABEL: @dec_mask_neg_i32(
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[X:%.*]], -1
; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[X]], -1
; CHECK-NEXT:    [[DEC:%.*]] = and i32 [[TMP1]], [[TMP2]]
; CHECK-NEXT:    ret i32 [[DEC]]
;
  %neg = sub i32 0, %X
  %mask = and i32 %neg, %X
  %dec = add i32 %mask, -1
  ret i32 %dec
}

define i32 @dec_mask_commute_neg_i32(i32 %A) {
; CHECK-LABEL: @dec_mask_commute_neg_i32(
; CHECK-NEXT:    [[X:%.*]] = sdiv i32 42, [[A:%.*]]
; CHECK-NEXT:    [[TMP1:%.*]] = add nsw i32 [[X]], -1
; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[X]], -1
; CHECK-NEXT:    [[DEC:%.*]] = and i32 [[TMP1]], [[TMP2]]
; CHECK-NEXT:    ret i32 [[DEC]]
;
  %X = sdiv i32 42, %A ; thwart complexity-based canonicalization
  %neg = sub i32 0, %X
  %mask = and i32 %X, %neg
  %dec = add i32 %mask, -1
  ret i32 %dec
}

define i32 @dec_commute_mask_neg_i32(i32 %X) {
; CHECK-LABEL: @dec_commute_mask_neg_i32(
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[X:%.*]], -1
; CHECK-NEXT:    [[TMP2:%.*]] = xor i32 [[X]], -1
; CHECK-NEXT:    [[DEC:%.*]] = and i32 [[TMP1]], [[TMP2]]
; CHECK-NEXT:    ret i32 [[DEC]]
;
  %neg = sub i32 0, %X
  %mask = and i32 %neg, %X
  %dec = add i32 -1, %mask
  ret i32 %dec
}

define i32 @dec_mask_neg_multiuse_i32(i32 %X) {
; CHECK-LABEL: @dec_mask_neg_multiuse_i32(
; CHECK-NEXT:    [[NEG:%.*]] = sub i32 0, [[X:%.*]]
; CHECK-NEXT:    [[MASK:%.*]] = and i32 [[NEG]], [[X]]
; CHECK-NEXT:    [[DEC:%.*]] = add i32 [[MASK]], -1
; CHECK-NEXT:    call void @use(i32 [[NEG]])
; CHECK-NEXT:    ret i32 [[DEC]]
;
  %neg = sub i32 0, %X
  %mask = and i32 %neg, %X
  %dec = add i32 %mask, -1
  call void @use(i32 %neg)
  ret i32 %dec
}

define i32 @dec_mask_multiuse_neg_i32(i32 %X) {
; CHECK-LABEL: @dec_mask_multiuse_neg_i32(
; CHECK-NEXT:    [[NEG:%.*]] = sub i32 0, [[X:%.*]]
; CHECK-NEXT:    [[MASK:%.*]] = and i32 [[NEG]], [[X]]
; CHECK-NEXT:    [[DEC:%.*]] = add i32 [[MASK]], -1
; CHECK-NEXT:    call void @use(i32 [[MASK]])
; CHECK-NEXT:    ret i32 [[DEC]]
;
  %neg = sub i32 0, %X
  %mask = and i32 %neg, %X
  %dec = add i32 %mask, -1
  call void @use(i32 %mask)
  ret i32 %dec
}

define <2 x i32> @dec_mask_neg_v2i32(<2 x i32> %X) {
; CHECK-LABEL: @dec_mask_neg_v2i32(
; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -1, i32 -1>
; CHECK-NEXT:    [[TMP2:%.*]] = xor <2 x i32> [[X]], <i32 -1, i32 -1>
; CHECK-NEXT:    [[DEC:%.*]] = and <2 x i32> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    ret <2 x i32> [[DEC]]
;
  %neg = sub <2 x i32> zeroinitializer, %X
  %mask = and <2 x i32> %neg, %X
  %dec = add <2 x i32> %mask, <i32 -1, i32 -1>
  ret <2 x i32> %dec
}

define <2 x i32> @dec_mask_neg_v2i32_undef(<2 x i32> %X) {
; CHECK-LABEL: @dec_mask_neg_v2i32_undef(
; CHECK-NEXT:    [[TMP1:%.*]] = add <2 x i32> [[X:%.*]], <i32 -1, i32 -1>
; CHECK-NEXT:    [[TMP2:%.*]] = xor <2 x i32> [[X]], <i32 -1, i32 -1>
; CHECK-NEXT:    [[DEC:%.*]] = and <2 x i32> [[TMP1]], [[TMP2]]
; CHECK-NEXT:    ret <2 x i32> [[DEC]]
;
  %neg = sub <2 x i32> zeroinitializer, %X
  %mask = and <2 x i32> %neg, %X
  %dec = add <2 x i32> %mask, <i32 -1, i32 undef>
  ret <2 x i32> %dec
}

define <2 x i32> @dec_mask_multiuse_neg_multiuse_v2i32(<2 x i32> %X) {
; CHECK-LABEL: @dec_mask_multiuse_neg_multiuse_v2i32(
; CHECK-NEXT:    [[NEG:%.*]] = sub <2 x i32> zeroinitializer, [[X:%.*]]
; CHECK-NEXT:    [[MASK:%.*]] = and <2 x i32> [[NEG]], [[X]]
; CHECK-NEXT:    [[DEC:%.*]] = add <2 x i32> [[MASK]], <i32 -1, i32 -1>
; CHECK-NEXT:    call void @usev(<2 x i32> [[NEG]])
; CHECK-NEXT:    call void @usev(<2 x i32> [[MASK]])
; CHECK-NEXT:    ret <2 x i32> [[DEC]]
;
  %neg = sub <2 x i32> zeroinitializer, %X
  %mask = and <2 x i32> %neg, %X
  %dec = add <2 x i32> %mask, <i32 -1, i32 -1>
  call void @usev(<2 x i32> %neg)
  call void @usev(<2 x i32> %mask)
  ret <2 x i32> %dec
}

declare void @use(i32)
declare void @usev(<2 x i32>)