[InstCombine] cttz(-x) -> cttz(x)

Summary: Signedness does not change number of trailing zeros.

Reviewers: spatel, lebedev.ri, nikic

Reviewed By: spatel

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63534

llvm-svn: 363951
This commit is contained in:
David Bolvansky 2019-06-20 17:04:14 +00:00
parent 85fec937ad
commit e76e0a0612
2 changed files with 32 additions and 16 deletions

View File

@ -1194,6 +1194,12 @@ static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombiner &IC) {
return CallInst::Create(F, {X, II.getArgOperand(1)});
}
// cttz(-x) -> cttz(x)
if (IsTZ && match(Op0, m_Neg(m_Value(X)))) {
II.setOperand(0, X);
return ⅈ
}
KnownBits Known = IC.computeKnownBits(Op0, 0, &II);
// Create a mask for bits above (ctlz) or below (cttz) the first known one.

View File

@ -3,44 +3,53 @@
define i32 @cttz_neg_value(i32 %x) {
; CHECK-LABEL: @cttz_neg_value(
; CHECK-NEXT: [[A:%.*]] = sub nsw i32 0, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = call i32 @llvm.cttz.i32(i32 [[A]], i1 false), !range !0
; CHECK-NEXT: [[B:%.*]] = call i32 @llvm.cttz.i32(i32 [[X:%.*]], i1 false), !range !0
; CHECK-NEXT: ret i32 [[B]]
;
%a = sub nsw i32 0, %x
%a = sub i32 0, %x
%b = tail call i32 @llvm.cttz.i32(i32 %a)
ret i32 %b
}
define i32 @cttz_neg_value_multiuse(i32 %x) {
; CHECK-LABEL: @cttz_neg_value_multiuse(
; CHECK-NEXT: [[A:%.*]] = sub i32 0, [[X:%.*]]
; CHECK-NEXT: call void @use(i32 [[A]])
; CHECK-NEXT: [[B:%.*]] = call i32 @llvm.cttz.i32(i32 [[X]], i1 false), !range !0
; CHECK-NEXT: ret i32 [[B]]
;
%a = sub i32 0, %x
call void @use(i32 %a)
%b = tail call i32 @llvm.cttz.i32(i32 %a)
ret i32 %b
}
define i64 @cttz_neg_value_64(i64 %x) {
; CHECK-LABEL: @cttz_neg_value_64(
; CHECK-NEXT: [[A:%.*]] = sub nsw i64 0, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = tail call i64 @llvm.cttz.i64(i64 [[A]], i1 true), !range !1
; CHECK-NEXT: [[B:%.*]] = tail call i64 @llvm.cttz.i64(i64 [[X:%.*]], i1 true), !range !1
; CHECK-NEXT: ret i64 [[B]]
;
%a = sub nsw i64 0, %x
%a = sub i64 0, %x
%b = tail call i64 @llvm.cttz.i64(i64 %a, i1 true)
ret i64 %b
}
define i64 @cttz_neg_value2_64(i64 %x) {
; CHECK-LABEL: @cttz_neg_value2_64(
; CHECK-NEXT: [[A:%.*]] = sub nsw i64 0, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = tail call i64 @llvm.cttz.i64(i64 [[A]], i1 false), !range !1
; CHECK-NEXT: [[B:%.*]] = tail call i64 @llvm.cttz.i64(i64 [[X:%.*]], i1 false), !range !1
; CHECK-NEXT: ret i64 [[B]]
;
%a = sub nsw i64 0, %x
%a = sub i64 0, %x
%b = tail call i64 @llvm.cttz.i64(i64 %a, i1 false)
ret i64 %b
}
define <2 x i64> @cttz_neg_value_vec(<2 x i64> %x) {
; CHECK-LABEL: @cttz_neg_value_vec(
; CHECK-NEXT: [[A:%.*]] = sub nsw <2 x i64> zeroinitializer, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> [[A]], i1 false)
; CHECK-NEXT: [[B:%.*]] = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> [[X:%.*]], i1 false)
; CHECK-NEXT: ret <2 x i64> [[B]]
;
%a = sub nsw <2 x i64> zeroinitializer, %x
%a = sub <2 x i64> zeroinitializer, %x
%b = tail call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a)
ret <2 x i64> %b
}
@ -49,26 +58,27 @@ define <2 x i64> @cttz_neg_value_vec(<2 x i64> %x) {
define i32 @cttz_nonneg_value(i32 %x) {
; CHECK-LABEL: @cttz_nonneg_value(
; CHECK-NEXT: [[A:%.*]] = sub nsw i32 1, [[X:%.*]]
; CHECK-NEXT: [[A:%.*]] = sub i32 1, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = call i32 @llvm.cttz.i32(i32 [[A]], i1 false), !range !0
; CHECK-NEXT: ret i32 [[B]]
;
%a = sub nsw i32 1, %x
%a = sub i32 1, %x
%b = tail call i32 @llvm.cttz.i32(i32 %a)
ret i32 %b
}
define <2 x i64> @cttz_nonneg_value_vec(<2 x i64> %x) {
; CHECK-LABEL: @cttz_nonneg_value_vec(
; CHECK-NEXT: [[A:%.*]] = sub nsw <2 x i64> <i64 1, i64 0>, [[X:%.*]]
; CHECK-NEXT: [[A:%.*]] = sub <2 x i64> <i64 1, i64 0>, [[X:%.*]]
; CHECK-NEXT: [[B:%.*]] = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> [[A]], i1 false)
; CHECK-NEXT: ret <2 x i64> [[B]]
;
%a = sub nsw <2 x i64> <i64 1, i64 0>, %x
%a = sub <2 x i64> <i64 1, i64 0>, %x
%b = tail call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a)
ret <2 x i64> %b
}
declare void @use(i32 %a)
declare i32 @llvm.cttz.i32(i32)
declare i64 @llvm.cttz.i64(i64, i1)
declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>)