RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00
Code Issues Projects Releases Wiki Activity

[LoopUtils, LV] Fix PR26734

Browse Source 
The vectorization of first-order recurrences (r261346) caused PR26734. When
detecting these recurrences, we need to ensure that the previous value is
actually defined inside the loop. This patch includes the fix and test case.

llvm-svn: 262624
This commit is contained in:
Matthew Simpson 2016-03-03 16:12:01 +00:00
parent 0d79c5acf8
commit a066e2421a
2 changed files with 50 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Transforms/Utils/LoopUtils.cpp
View File

@ -543,7 +543,7 @@ bool RecurrenceDescriptor::isFirstOrderRecurrence(PHINode *Phi, Loop *TheLoop,
// Get the previous value. The previous value comes from the latch edge while
// the initial value comes form the preheader edge.
auto *Previous = dyn_cast<Instruction>(Phi->getIncomingValueForBlock(Latch));
if (!Previous)
if (!Previous || !TheLoop->contains(Previous))
return false;
// Ensure every user of the phi node is dominated by the previous value. The

49
test/Transforms/LoopVectorize/AArch64/first-order-recurrence.ll
View File

@ -207,3 +207,52 @@ for.end.loopexit:
for.end:
ret void
}
; CHECK-LABEL: @PR26734
;
; void PR26734(short *a, int *b, int *c, int d, short *e) {
; for (; d != 21; d++) {
; *b &= *c;
; *e = *a - 6;
; *c = *e;
; }
; }
;
; CHECK-NOT: vector.ph:
;
define void @PR26734(i16* %a, i32* %b, i32* %c, i32 %d, i16* %e) {
entry:
%cmp4 = icmp eq i32 %d, 21
br i1 %cmp4, label %entry.for.end_crit_edge, label %for.body.lr.ph
entry.for.end_crit_edge:
%.pre = load i32, i32* %b, align 4
br label %for.end
for.body.lr.ph:
%0 = load i16, i16* %a, align 2
%sub = add i16 %0, -6
%conv2 = sext i16 %sub to i32
%c.promoted = load i32, i32* %c, align 4
%b.promoted = load i32, i32* %b, align 4
br label %for.body
for.body:
%inc7 = phi i32 [ %d, %for.body.lr.ph ], [ %inc, %for.body ]
%and6 = phi i32 [ %b.promoted, %for.body.lr.ph ], [ %and, %for.body ]
%conv25 = phi i32 [ %c.promoted, %for.body.lr.ph ], [ %conv2, %for.body ]
%and = and i32 %and6, %conv25
%inc = add nsw i32 %inc7, 1
%cmp = icmp eq i32 %inc, 21
br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
for.cond.for.end_crit_edge:
%and.lcssa = phi i32 [ %and, %for.body ]
store i32 %conv2, i32* %c, align 4
store i32 %and.lcssa, i32* %b, align 4
store i16 %sub, i16* %e, align 2
br label %for.end
for.end:
ret void
}