This is an archive of the discontinued LLVM Phabricator instance.

Differential D149426

[InstCombine] Fold `(cmp eq/ne (umax X, Y),0)` -> `(cmp eq/ne (or X, Y),0)`
ClosedPublic

Authored by goldstein.w.n on Apr 27 2023, 11:24 PM.

Details

Reviewers
nikic
spatel
Commits
rGdc13624e88d8: [InstCombine] Fold `(cmp eq/ne (umax X, Y),0)` -> `(cmp eq/ne (or X, Y),0)`
Summary

or is almost always preferable.

Diff Detail

Event Timeline

goldstein.w.n created this revision.Apr 27 2023, 11:24 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 27 2023, 11:24 PM
Herald added subscribers: StephenFan, hiraditya. · View Herald Transcript
goldstein.w.n requested review of this revision.Apr 27 2023, 11:24 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 27 2023, 11:24 PM
Herald added a subscriber: llvm-commits. · View Herald Transcript
Harbormaster completed remote builds in B228746: Diff 517810.Apr 27 2023, 11:25 PM
goldstein.w.n added a parent revision: D149425: [InstCombine] Don't fold `uadd.sat` to `or` if it increase instruction count.Apr 27 2023, 11:29 PM
nikic accepted this revision.Apr 28 2023, 12:28 AM

LGTM

This revision is now accepted and ready to land.Apr 28 2023, 12:28 AM
This revision was landed with ongoing or failed builds.Apr 29 2023, 10:39 AM
Closed by commit rGdc13624e88d8: [InstCombine] Fold `(cmp eq/ne (umax X, Y),0)` -> `(cmp eq/ne (or X, Y),0)` (authored by goldstein.w.n). · Explain Why
This revision was automatically updated to reflect the committed changes.
goldstein.w.n added a commit: rGdc13624e88d8: [InstCombine] Fold `(cmp eq/ne (umax X, Y),0)` -> `(cmp eq/ne (or X, Y),0)`.

Revision Contents

PathSize
llvm/
lib/
Transforms/
InstCombine/
2 lines
test/
Transforms/
InstCombine/
8 lines

Diff 517810

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 3,378 Lines • ▼ Show 20 Lines if (II->getArgOperand(0) == II->getArgOperand(1)) {
if (match(II->getArgOperand(2), m_APInt(RotAmtC))) if (match(II->getArgOperand(2), m_APInt(RotAmtC)))
return new ICmpInst(Pred, II->getArgOperand(0), return new ICmpInst(Pred, II->getArgOperand(0),
II->getIntrinsicID() == Intrinsic::fshl II->getIntrinsicID() == Intrinsic::fshl
? ConstantInt::get(Ty, C.rotr(*RotAmtC)) ? ConstantInt::get(Ty, C.rotr(*RotAmtC))
: ConstantInt::get(Ty, C.rotl(*RotAmtC))); : ConstantInt::get(Ty, C.rotl(*RotAmtC)));
} }
break; break;
case Intrinsic::umax:
case Intrinsic::uadd_sat: { case Intrinsic::uadd_sat: {
// uadd.sat(a, b) == 0 -> (a | b) == 0 // uadd.sat(a, b) == 0 -> (a | b) == 0
// umax(a, b) == 0 -> (a | b) == 0
if (C.isZero() && II->hasOneUse()) { if (C.isZero() && II->hasOneUse()) {
Value *Or = Builder.CreateOr(II->getArgOperand(0), II->getArgOperand(1)); Value *Or = Builder.CreateOr(II->getArgOperand(0), II->getArgOperand(1));
return new ICmpInst(Pred, Or, Constant::getNullValue(Ty)); return new ICmpInst(Pred, Or, Constant::getNullValue(Ty));
} }
break; break;
} }
case Intrinsic::usub_sat: { case Intrinsic::usub_sat: {
▲ Show 20 Lines • Show All 3,928 Lines • Show Last 20 Lines

llvm/test/Transforms/InstCombine/cmp-intrinsic.ll

Show First 20 Lines • Show All 806 Lines • ▼ Show 20 Lines
; ;
%y = call <2 x i8> @llvm.bitreverse.v2i8(<2 x i8> %x) %y = call <2 x i8> @llvm.bitreverse.v2i8(<2 x i8> %x)
%z = icmp eq <2 x i8> %y, <i8 1, i8 2> %z = icmp eq <2 x i8> %y, <i8 1, i8 2>
ret <2 x i1> %z ret <2 x i1> %z
} }
define i1 @umax_eq_zero(i8 %x, i8 %y) { define i1 @umax_eq_zero(i8 %x, i8 %y) {
; CHECK-LABEL: @umax_eq_zero( ; CHECK-LABEL: @umax_eq_zero(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 [[Y:%.*]]) ; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[M]], 0 ; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[R]] ; CHECK-NEXT: ret i1 [[R]]
; ;
%m = call i8 @llvm.umax.i8(i8 %x, i8 %y) %m = call i8 @llvm.umax.i8(i8 %x, i8 %y)
%r = icmp eq i8 %m, 0 %r = icmp eq i8 %m, 0
ret i1 %r ret i1 %r
} }
define i1 @umax_eq_1_fail(i8 %x, i8 %y) { define i1 @umax_eq_1_fail(i8 %x, i8 %y) {
Show All 15 Lines
; ;
%m = call i8 @llvm.umax.i8(i8 %x, i8 %y) %m = call i8 @llvm.umax.i8(i8 %x, i8 %y)
%r = icmp sle i8 %m, 0 %r = icmp sle i8 %m, 0
ret i1 %r ret i1 %r
} }
define i1 @umax_ne_zero(i8 %x, i8 %y) { define i1 @umax_ne_zero(i8 %x, i8 %y) {
; CHECK-LABEL: @umax_ne_zero( ; CHECK-LABEL: @umax_ne_zero(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 [[Y:%.*]]) ; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[M]], 0 ; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[R]] ; CHECK-NEXT: ret i1 [[R]]
; ;
%m = call i8 @llvm.umax.i8(i8 %x, i8 %y) %m = call i8 @llvm.umax.i8(i8 %x, i8 %y)
%r = icmp ne i8 %m, 0 %r = icmp ne i8 %m, 0
ret i1 %r ret i1 %r
} }
define i1 @umax_ne_zero_fail_multiuse(i8 %x, i8 %y) { define i1 @umax_ne_zero_fail_multiuse(i8 %x, i8 %y) {
Show All 26 Lines