RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-21 03:53:04 +02:00
Code Issues Projects Releases Wiki Activity

[Pipeliner] Fix in the pipeliner phi reuse code

Browse Source 
When the definition of a phi is used by a phi in the next iteration,
the pipeliner was assuming that the definition is processed first.
Because of the assumption, an incorrect phi name was used. This patch
has a check to see if the phi definition has been processed already.

Patch by Brendon Cahoon.

llvm-svn: 328510
This commit is contained in:
Krzysztof Parzyszek 2018-03-26 15:58:16 +00:00
parent d7c0f824c4
commit 72c5d3ff50
2 changed files with 46 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
lib/CodeGen/MachinePipeliner.cpp
View File

@ -2723,7 +2723,8 @@ void SwingSchedulerDAG::generateExistingPhis(
int LVNumStages = Schedule.getStagesForPhi(LoopVal); int LVNumStages = Schedule.getStagesForPhi(LoopVal);
int StageDiff = (StageScheduled - LoopValStage); int StageDiff = (StageScheduled - LoopValStage);
LVNumStages -= StageDiff; LVNumStages -= StageDiff;
if (LVNumStages > (int)np) { // Make sure the loop value Phi has been processed already.
if (LVNumStages > (int)np && VRMap[CurStageNum].count(LoopVal)) {
NewReg = PhiOp2; NewReg = PhiOp2;
unsigned ReuseStage = CurStageNum; unsigned ReuseStage = CurStageNum;
if (Schedule.isLoopCarried(this, *PhiInst)) if (Schedule.isLoopCarried(this, *PhiInst))

44
test/CodeGen/Hexagon/swp-reuse-phi-5.ll Normal file
View File

@ -0,0 +1,44 @@
; RUN: llc -march=hexagon -hexagon-bit=false < %s
; REQUIRES: asserts
; Fix for an undefined virtual register assert that was caused by an
; incorrect phi generated by the pipeliner. In this case, there is a
; phi that defines a value used by another phi in the next iteration.
; The pipeliner code for generating new phis was assuming that the
; definition is processed before the use, so an incorrect value was
; used.
; Function Attrs: nounwind
define void @f0() local_unnamed_addr #0 {
b0:
br label %b1
b1: ; preds = %b1, %b0
%v0 = phi i64 [ 0, %b0 ], [ %v5, %b1 ]
%v1 = phi i64 [ undef, %b0 ], [ %v9, %b1 ]
%v2 = phi i32 [ 0, %b0 ], [ %v10, %b1 ]
%v3 = phi i32 [ undef, %b0 ], [ %v4, %b1 ]
%v4 = phi i32 [ undef, %b0 ], [ %v8, %b1 ]
%v5 = tail call i64 @llvm.hexagon.M2.vdmacs.s0(i64 %v0, i64 %v1, i64 undef)
%v6 = tail call i64 @llvm.hexagon.A2.combinew(i32 %v3, i32 %v3)
%v7 = tail call i64 @llvm.hexagon.M2.vdmacs.s0(i64 undef, i64 %v6, i64 undef)
%v8 = tail call i32 @llvm.hexagon.A2.combine.ll(i32 undef, i32 undef)
%v9 = tail call i64 @llvm.hexagon.A2.combinew(i32 %v8, i32 %v4)
%v10 = add nuw nsw i32 %v2, 1
%v11 = icmp eq i32 %v10, undef
br i1 %v11, label %b2, label %b1
b2: ; preds = %b1
%v12 = lshr i64 %v7, 32
%v13 = trunc i64 %v12 to i32
store i32 %v13, i32* undef, align 4
%v14 = lshr i64 %v5, 32
ret void
}
declare i32 @llvm.hexagon.A2.combine.ll(i32, i32) #1
declare i64 @llvm.hexagon.A2.combinew(i32, i32) #1
declare i64 @llvm.hexagon.M2.vdmacs.s0(i64, i64, i64) #1
attributes #0 = { nounwind "target-cpu"="hexagonv65" }
attributes #1 = { nounwind readnone }