replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[InstCombine] remove unneeded one-use checks for icmp fold 

This fold and several others were added in:
rL125734
...with no explanation for the one-use checks other than the code
comments about register pressure.

Given that this is IR canonicalization, we shouldn't be worried
about register pressure though; the backend should be able to
adjust for that as needed.

There are similar checks as noted with the TODO comments. I'm
hoping to remove those restrictions too, but if any of these
does cause a regression, it should be easier to correct by making
small, individual commits.

This is part of solving PR43310 the theoretically right way:
https://bugs.llvm.org/show_bug.cgi?id=43310
...ie, if we don't cripple basic transforms, then we won't
need to add special-case code to detect larger patterns.

llvm-svn: 371940
This commit is contained in:
Sanjay Patel 2019-09-15 20:56:34 +00:00
parent c77ad16f8e
commit 3daf168fa9
2 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
View File

@ -3695,6 +3695,7 @@ Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) {
C == Op0 ? D : C);
// icmp (X+Y), (X+Z) -> icmp Y, Z for equalities or if there is no overflow.
// TODO: The one-use checks should not be necessary.
if (A && C && (A == C || A == D || B == C || B == D) && NoOp0WrapProblem &&
NoOp1WrapProblem &&
// Try not to increase register pressure.
@ -3842,11 +3843,11 @@ Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) {
return new ICmpInst(Pred, C, D);
// icmp (Y-X), (Z-X) -> icmp Y, Z for equalities or if there is no overflow.
if (B && D && B == D && NoOp0WrapProblem && NoOp1WrapProblem &&
// Try not to increase register pressure.
BO0->hasOneUse() && BO1->hasOneUse())
if (B && D && B == D && NoOp0WrapProblem && NoOp1WrapProblem)
return new ICmpInst(Pred, A, C);
// icmp (X-Y), (X-Z) -> icmp Z, Y for equalities or if there is no overflow.
// TODO: The one-use checks should not be necessary.
if (A && C && A == C && NoOp0WrapProblem && NoOp1WrapProblem &&
// Try not to increase register pressure.
BO0->hasOneUse() && BO1->hasOneUse())

4
llvm/test/Transforms/InstCombine/icmp.ll
View File

@ -562,7 +562,7 @@ define i1 @test27_extra_uses(i32 %x, i32 %y, i32 %z) {
; CHECK-NEXT: call void @foo(i32 [[LHS]])
; CHECK-NEXT: [[RHS:%.*]] = sub nsw i32 [[Y:%.*]], [[Z]]
; CHECK-NEXT: call void @foo(i32 [[RHS]])
; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[LHS]], [[RHS]]
; CHECK-NEXT: [[C:%.*]] = icmp sgt i32 [[X]], [[Y]]
; CHECK-NEXT: ret i1 [[C]]
;
%lhs = sub nsw i32 %x, %z
@ -591,7 +591,7 @@ define i1 @test28_extra_uses(i32 %x, i32 %y, i32 %z) {
; CHECK-NEXT: call void @foo(i32 [[LHS]])
; CHECK-NEXT: [[RHS:%.*]] = sub nuw i32 [[Y:%.*]], [[Z]]
; CHECK-NEXT: call void @foo(i32 [[RHS]])
; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[LHS]], [[RHS]]
; CHECK-NEXT: [[C:%.*]] = icmp ugt i32 [[X]], [[Y]]
; CHECK-NEXT: ret i1 [[C]]
;
%lhs = sub nuw i32 %x, %z