Skip to content

Commit be753b2

Browse files
kmclaughlin-armkmclaughlin-arm
committed
Revert "[LV] Use lookThroughAnd with logical reductions"
Reverting patch due to buildbot failures. This reverts commit e22a599.
1 parent ca9b60f commit be753b2

File tree

4 files changed

+59
-303
lines changed

4 files changed

+59
-303
lines changed

‎llvm/lib/Analysis/IVDescriptors.cpp

+1-1
Original file line numberDiff line numberDiff line change
@@ -274,7 +274,7 @@ bool RecurrenceDescriptor::AddReductionVar(PHINode *Phi, RecurKind Kind,
274274
} else if (RecurrenceType->isIntegerTy()) {
275275
if (!isIntegerRecurrenceKind(Kind))
276276
return false;
277-
if (!isMinMaxRecurrenceKind(Kind))
277+
if (isArithmeticRecurrenceKind(Kind))
278278
Start = lookThroughAnd(Phi, RecurrenceType, VisitedInsts, CastInsts);
279279
} else {
280280
// Pointer min/max may exist, but it is not supported as a reduction op.

‎llvm/test/Transforms/LoopVectorize/reduction-inloop-pred.ll

+44-41
Original file line numberDiff line numberDiff line change
@@ -1541,69 +1541,72 @@ define i8 @reduction_and_trunc(i8* noalias nocapture %A) {
15411541
; CHECK: vector.body:
15421542
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE6:%.*]] ]
15431543
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE6]] ]
1544-
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ <i32 255, i32 -1, i32 -1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP29:%.*]], [[PRED_LOAD_CONTINUE6]] ]
1544+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 255, [[VECTOR_PH]] ], [ [[TMP34:%.*]], [[PRED_LOAD_CONTINUE6]] ]
15451545
; CHECK-NEXT: [[TMP0:%.*]] = icmp ult <4 x i32> [[VEC_IND]], <i32 257, i32 257, i32 257, i32 257>
1546-
; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i1> [[TMP0]], i32 0
1547-
; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
1546+
; CHECK-NEXT: [[TMP1:%.*]] = select <4 x i1> [[TMP0]], <4 x i32> <i32 255, i32 255, i32 255, i32 255>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
1547+
; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP1]])
1548+
; CHECK-NEXT: [[TMP3:%.*]] = and i32 [[TMP2]], [[VEC_PHI]]
1549+
; CHECK-NEXT: [[TMP4:%.*]] = extractelement <4 x i1> [[TMP0]], i32 0
1550+
; CHECK-NEXT: br i1 [[TMP4]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
15481551
; CHECK: pred.load.if:
1549-
; CHECK-NEXT: [[TMP2:%.*]] = sext i32 [[INDEX]] to i64
1550-
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP2]]
1551-
; CHECK-NEXT: [[TMP4:%.*]] = load i8, i8* [[TMP3]], align 4
1552-
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x i8> poison, i8 [[TMP4]], i32 0
1552+
; CHECK-NEXT: [[TMP5:%.*]] = sext i32 [[INDEX]] to i64
1553+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP5]]
1554+
; CHECK-NEXT: [[TMP7:%.*]] = load i8, i8* [[TMP6]], align 4
1555+
; CHECK-NEXT: [[TMP8:%.*]] = insertelement <4 x i8> poison, i8 [[TMP7]], i32 0
15531556
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]]
15541557
; CHECK: pred.load.continue:
1555-
; CHECK-NEXT: [[TMP6:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_LOAD_IF]] ]
1556-
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <4 x i1> [[TMP0]], i32 1
1557-
; CHECK-NEXT: br i1 [[TMP7]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]]
1558+
; CHECK-NEXT: [[TMP9:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_LOAD_IF]] ]
1559+
; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP0]], i32 1
1560+
; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]]
15581561
; CHECK: pred.load.if1:
1559-
; CHECK-NEXT: [[TMP8:%.*]] = or i32 [[INDEX]], 1
1560-
; CHECK-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
1561-
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP9]]
1562-
; CHECK-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 4
1563-
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i8> [[TMP6]], i8 [[TMP11]], i32 1
1562+
; CHECK-NEXT: [[TMP11:%.*]] = or i32 [[INDEX]], 1
1563+
; CHECK-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
1564+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP12]]
1565+
; CHECK-NEXT: [[TMP14:%.*]] = load i8, i8* [[TMP13]], align 4
1566+
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i8> [[TMP9]], i8 [[TMP14]], i32 1
15641567
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]]
15651568
; CHECK: pred.load.continue2:
1566-
; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x i8> [ [[TMP6]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP12]], [[PRED_LOAD_IF1]] ]
1567-
; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP0]], i32 2
1568-
; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]]
1569+
; CHECK-NEXT: [[TMP16:%.*]] = phi <4 x i8> [ [[TMP9]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP15]], [[PRED_LOAD_IF1]] ]
1570+
; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP0]], i32 2
1571+
; CHECK-NEXT: br i1 [[TMP17]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]]
15691572
; CHECK: pred.load.if3:
1570-
; CHECK-NEXT: [[TMP15:%.*]] = or i32 [[INDEX]], 2
1571-
; CHECK-NEXT: [[TMP16:%.*]] = sext i32 [[TMP15]] to i64
1572-
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP16]]
1573-
; CHECK-NEXT: [[TMP18:%.*]] = load i8, i8* [[TMP17]], align 4
1574-
; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP18]], i32 2
1573+
; CHECK-NEXT: [[TMP18:%.*]] = or i32 [[INDEX]], 2
1574+
; CHECK-NEXT: [[TMP19:%.*]] = sext i32 [[TMP18]] to i64
1575+
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP19]]
1576+
; CHECK-NEXT: [[TMP21:%.*]] = load i8, i8* [[TMP20]], align 4
1577+
; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i8> [[TMP16]], i8 [[TMP21]], i32 2
15751578
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]]
15761579
; CHECK: pred.load.continue4:
1577-
; CHECK-NEXT: [[TMP20:%.*]] = phi <4 x i8> [ [[TMP13]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP19]], [[PRED_LOAD_IF3]] ]
1578-
; CHECK-NEXT: [[TMP21:%.*]] = extractelement <4 x i1> [[TMP0]], i32 3
1579-
; CHECK-NEXT: br i1 [[TMP21]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6]]
1580+
; CHECK-NEXT: [[TMP23:%.*]] = phi <4 x i8> [ [[TMP16]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP22]], [[PRED_LOAD_IF3]] ]
1581+
; CHECK-NEXT: [[TMP24:%.*]] = extractelement <4 x i1> [[TMP0]], i32 3
1582+
; CHECK-NEXT: br i1 [[TMP24]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6]]
15801583
; CHECK: pred.load.if5:
1581-
; CHECK-NEXT: [[TMP22:%.*]] = or i32 [[INDEX]], 3
1582-
; CHECK-NEXT: [[TMP23:%.*]] = sext i32 [[TMP22]] to i64
1583-
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP23]]
1584-
; CHECK-NEXT: [[TMP25:%.*]] = load i8, i8* [[TMP24]], align 4
1585-
; CHECK-NEXT: [[TMP26:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP25]], i32 3
1584+
; CHECK-NEXT: [[TMP25:%.*]] = or i32 [[INDEX]], 3
1585+
; CHECK-NEXT: [[TMP26:%.*]] = sext i32 [[TMP25]] to i64
1586+
; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP26]]
1587+
; CHECK-NEXT: [[TMP28:%.*]] = load i8, i8* [[TMP27]], align 4
1588+
; CHECK-NEXT: [[TMP29:%.*]] = insertelement <4 x i8> [[TMP23]], i8 [[TMP28]], i32 3
15861589
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]]
15871590
; CHECK: pred.load.continue6:
1588-
; CHECK-NEXT: [[TMP27:%.*]] = phi <4 x i8> [ [[TMP20]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP26]], [[PRED_LOAD_IF5]] ]
1589-
; CHECK-NEXT: [[TMP28:%.*]] = zext <4 x i8> [[TMP27]] to <4 x i32>
1590-
; CHECK-NEXT: [[TMP29]] = and <4 x i32> [[VEC_PHI]], [[TMP28]]
1591+
; CHECK-NEXT: [[TMP30:%.*]] = phi <4 x i8> [ [[TMP23]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP29]], [[PRED_LOAD_IF5]] ]
1592+
; CHECK-NEXT: [[TMP31:%.*]] = zext <4 x i8> [[TMP30]] to <4 x i32>
1593+
; CHECK-NEXT: [[TMP32:%.*]] = select <4 x i1> [[TMP0]], <4 x i32> [[TMP31]], <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
1594+
; CHECK-NEXT: [[TMP33:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP32]])
1595+
; CHECK-NEXT: [[TMP34]] = and i32 [[TMP33]], [[TMP3]]
15911596
; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4
15921597
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
1593-
; CHECK-NEXT: [[TMP30:%.*]] = icmp eq i32 [[INDEX_NEXT]], 260
1594-
; CHECK-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
1598+
; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i32 [[INDEX_NEXT]], 260
1599+
; CHECK-NEXT: br i1 [[TMP35]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
15951600
; CHECK: middle.block:
1596-
; CHECK-NEXT: [[TMP31:%.*]] = select <4 x i1> [[TMP0]], <4 x i32> [[TMP29]], <4 x i32> [[VEC_PHI]]
1597-
; CHECK-NEXT: [[TMP32:%.*]] = trunc <4 x i32> [[TMP31]] to <4 x i8>
1598-
; CHECK-NEXT: [[TMP33:%.*]] = call i8 @llvm.vector.reduce.and.v4i8(<4 x i8> [[TMP32]])
15991601
; CHECK-NEXT: br i1 true, label [[DOT_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
16001602
; CHECK: scalar.ph:
16011603
; CHECK-NEXT: br label [[DOTLR_PH:%.*]]
16021604
; CHECK: .lr.ph:
16031605
; CHECK-NEXT: br i1 undef, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP33:![0-9]+]]
16041606
; CHECK: ._crit_edge:
1605-
; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i8 [ undef, [[DOTLR_PH]] ], [ [[TMP33]], [[MIDDLE_BLOCK]] ]
1606-
; CHECK-NEXT: ret i8 [[SUM_0_LCSSA]]
1607+
; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ undef, [[DOTLR_PH]] ], [ [[TMP34]], [[MIDDLE_BLOCK]] ]
1608+
; CHECK-NEXT: [[RET:%.*]] = trunc i32 [[SUM_0_LCSSA]] to i8
1609+
; CHECK-NEXT: ret i8 [[RET]]
16071610
;
16081611
entry:
16091612
br label %.lr.ph

‎llvm/test/Transforms/LoopVectorize/reduction-inloop.ll

+14-11
Original file line numberDiff line numberDiff line change
@@ -1052,25 +1052,28 @@ define i8 @reduction_and_trunc(i8* noalias nocapture %A) {
10521052
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
10531053
; CHECK: vector.body:
10541054
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1055-
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i8> [ <i8 -1, i8 -1, i8 -1, i8 -1>, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ]
1056-
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[INDEX]] to i64
1057-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP0]]
1058-
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to <4 x i8>*
1059-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP2]], align 4
1060-
; CHECK-NEXT: [[TMP3]] = and <4 x i8> [[VEC_PHI]], [[WIDE_LOAD]]
1055+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 255, [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[VECTOR_BODY]] ]
1056+
; CHECK-NEXT: [[TMP0:%.*]] = and i32 [[VEC_PHI]], 255
1057+
; CHECK-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
1058+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP1]]
1059+
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>*
1060+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP3]], align 4
1061+
; CHECK-NEXT: [[TMP4:%.*]] = zext <4 x i8> [[WIDE_LOAD]] to <4 x i32>
1062+
; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP4]])
1063+
; CHECK-NEXT: [[TMP6]] = and i32 [[TMP5]], [[TMP0]]
10611064
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
1062-
; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
1063-
; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
1065+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
1066+
; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
10641067
; CHECK: middle.block:
1065-
; CHECK-NEXT: [[TMP5:%.*]] = call i8 @llvm.vector.reduce.and.v4i8(<4 x i8> [[TMP3]])
10661068
; CHECK-NEXT: br i1 true, label [[DOT_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
10671069
; CHECK: scalar.ph:
10681070
; CHECK-NEXT: br label [[DOTLR_PH:%.*]]
10691071
; CHECK: .lr.ph:
10701072
; CHECK-NEXT: br i1 undef, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP39:![0-9]+]]
10711073
; CHECK: ._crit_edge:
1072-
; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i8 [ undef, [[DOTLR_PH]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ]
1073-
; CHECK-NEXT: ret i8 [[SUM_0_LCSSA]]
1074+
; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ undef, [[DOTLR_PH]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ]
1075+
; CHECK-NEXT: [[RET:%.*]] = trunc i32 [[SUM_0_LCSSA]] to i8
1076+
; CHECK-NEXT: ret i8 [[RET]]
10741077
;
10751078
entry:
10761079
br label %.lr.ph

0 commit comments

Comments
 (0)