maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

Fix ScalarEvolutionExpander step scaling bug 

The expandAddRecExprLiterally function incorrectly transforms
`[Start + Step * X]` into `Step * [Start + X]` instead of the correct
transform of `[Step * X] + Start`.

This caused https://github.com/JuliaLang/julia/issues/14704#issuecomment-174126219
due to what appeared to be sufficiently complicated loop interactions.

Patch by Jameson Nash (jameson@juliacomputing.com).

Reviewers: sanjoy
Differential Revision: http://reviews.llvm.org/D16505

llvm-svn: 275239
This commit is contained in:
Keno Fischer 2016-07-13 01:28:12 +00:00
parent 835df56cb3
commit 1efc3b70c5
2 changed files with 55 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
llvm/lib/Analysis/ScalarEvolutionExpander.cpp
View File

@ -1288,6 +1288,13 @@ Value *SCEVExpander::expandAddRecExprLiterally(const SCEVAddRecExpr *S) {
if (!SE.dominates(Step, L->getHeader())) {
PostLoopScale = Step;
Step = SE.getConstant(Normalized->getType(), 1);
if (!Start->isZero()) {
// The normalization below assumes that Start is constant zero, so if
// it isn't re-associate Start to PostLoopOffset.
assert(!PostLoopOffset && "Start not-null but PostLoopOffset set?");
PostLoopOffset = Start;
Start = SE.getConstant(Normalized->getType(), 0);
}
Normalized =
cast<SCEVAddRecExpr>(SE.getAddRecExpr(
Start, Step, Normalized->getLoop(),

48
llvm/test/Analysis/ScalarEvolution/incorrect-offset-scaling.ll Normal file
View File

@ -0,0 +1,48 @@
; RUN: opt -S -loop-reduce < %s | FileCheck %s
target triple = "x86_64-unknown-unknown"
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
define void @incorrect_offset_scaling(i64, i64*) {
top:
br label %L
L: ; preds = %idxend.10, %idxend, %L2, %top
br i1 undef, label %L, label %L1
L1: ; preds = %L1.preheader, %L2
%r13 = phi i64 [ %r1, %L2 ], [ 1, %L ]
; CHECK: %lsr.iv = phi i64 [ 0, %L{{[^ ]+}} ], [ %lsr.iv.next, %L2 ]
; CHECK-NOT: %lsr.iv = phi i64 [ -1, %L{{[^ ]+}} ], [ %lsr.iv.next, %L2 ]
; CHECK: br
%r0 = add i64 %r13, -1
br label %idxend.8
L2: ; preds = %idxend.8
%r1 = add i64 %r13, 1
br i1 undef, label %L, label %L1
if6: ; preds = %idxend.8
%r2 = add i64 %0, -1
%r3 = load i64, i64* %1, align 8
; CHECK-NOT: %r2
; CHECK: %r3 = load i64
br label %ib
idxend.8: ; preds = %L1
br i1 undef, label %if6, label %L2
ib: ; preds = %if6
%r4 = mul i64 %r3, %r0
%r5 = add i64 %r2, %r4
%r6 = icmp ult i64 %r5, undef
; CHECK %2 = mul i64 %lsr.iv, %r3
; CHECK %3 = add i64 %1, -1
; CHECK %4 = add i64 %0, %r3
; CHECK %r6
%r7 = getelementptr i64, i64* undef, i64 %r5
store i64 1, i64* %r7, align 8
; CHECK %5 = mul i64 %lsr.iv, %r3
; CHECK %6 = add i64 %5, -1
br label %L
}