Skip to content

Commit e2bead5

Browse files
hazzlimhazzlim
committed
Address Review
1 parent 8347cb4 commit e2bead5

File tree

3 files changed

+16
-14
lines changed

3 files changed

+16
-14
lines changed

‎llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp‎

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1017,6 +1017,8 @@ static bool optimizeVectorInductionWidthForTCAndVFUF(VPlan &Plan,
10171017
if (!Plan.getVectorLoopRegion())
10181018
return false;
10191019

1020+
if (!Plan.getTripCount()->isLiveIn())
1021+
return false;
10201022
auto *TC = dyn_cast_if_present<ConstantInt>(
10211023
Plan.getTripCount()->getUnderlyingValue());
10221024
if (!TC || !BestVF.isFixed())
@@ -1043,8 +1045,8 @@ static bool optimizeVectorInductionWidthForTCAndVFUF(VPlan &Plan,
10431045
auto *WideIV = dyn_cast<VPWidenIntOrFpInductionRecipe>(&Phi);
10441046

10451047
// Currently only handle canonical IVs as it is trivial to replace the start
1046-
// and stop values, and we only perform the optimisation when the IV is only
1047-
// used by the comparison controlling loop control-flow.
1048+
// and stop values, and we currently only perform the optimization when the
1049+
// IV has a single use.
10481050
if (!WideIV || !WideIV->isCanonical() ||
10491051
WideIV->hasMoreThanOneUniqueUser() ||
10501052
NewIVTy == WideIV->getScalarType())

‎llvm/test/Transforms/LoopVectorize/AArch64/optsize_minsize.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -245,9 +245,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
245245
; DEFAULT-NEXT: br label %[[VECTOR_BODY:.*]]
246246
; DEFAULT: [[VECTOR_BODY]]:
247247
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
248-
; DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <16 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
248+
; DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
249249
; DEFAULT-NEXT: [[VEC_IND1:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[PRED_STORE_CONTINUE36]] ]
250-
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
250+
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
251251
; DEFAULT-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
252252
; DEFAULT-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
253253
; DEFAULT-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -399,7 +399,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
399399
; DEFAULT-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
400400
; DEFAULT-NEXT: br label %[[PRED_STORE_CONTINUE36]]
401401
; DEFAULT: [[PRED_STORE_CONTINUE36]]:
402-
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
402+
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
403403
; DEFAULT-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
404404
; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
405405
; DEFAULT-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
@@ -440,9 +440,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
440440
; OPTSIZE-NEXT: br label %[[VECTOR_BODY:.*]]
441441
; OPTSIZE: [[VECTOR_BODY]]:
442442
; OPTSIZE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
443-
; OPTSIZE-NEXT: [[VEC_IND:%.*]] = phi <16 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
443+
; OPTSIZE-NEXT: [[VEC_IND:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
444444
; OPTSIZE-NEXT: [[VEC_IND1:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[PRED_STORE_CONTINUE36]] ]
445-
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
445+
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
446446
; OPTSIZE-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
447447
; OPTSIZE-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
448448
; OPTSIZE-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -594,7 +594,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
594594
; OPTSIZE-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
595595
; OPTSIZE-NEXT: br label %[[PRED_STORE_CONTINUE36]]
596596
; OPTSIZE: [[PRED_STORE_CONTINUE36]]:
597-
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
597+
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
598598
; OPTSIZE-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
599599
; OPTSIZE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
600600
; OPTSIZE-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]

‎llvm/test/Transforms/LoopVectorize/ARM/optsize_minsize.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -241,9 +241,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
241241
; DEFAULT-NEXT: br label %[[VECTOR_BODY:.*]]
242242
; DEFAULT: [[VECTOR_BODY]]:
243243
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
244-
; DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <16 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
244+
; DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
245245
; DEFAULT-NEXT: [[VEC_IND1:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[PRED_STORE_CONTINUE36]] ]
246-
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
246+
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
247247
; DEFAULT-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
248248
; DEFAULT-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
249249
; DEFAULT-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -395,7 +395,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
395395
; DEFAULT-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
396396
; DEFAULT-NEXT: br label %[[PRED_STORE_CONTINUE36]]
397397
; DEFAULT: [[PRED_STORE_CONTINUE36]]:
398-
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
398+
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
399399
; DEFAULT-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
400400
; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
401401
; DEFAULT-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
@@ -436,9 +436,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
436436
; OPTSIZE-NEXT: br label %[[VECTOR_BODY:.*]]
437437
; OPTSIZE: [[VECTOR_BODY]]:
438438
; OPTSIZE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
439-
; OPTSIZE-NEXT: [[VEC_IND:%.*]] = phi <16 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
439+
; OPTSIZE-NEXT: [[VEC_IND:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_STORE_CONTINUE36]] ]
440440
; OPTSIZE-NEXT: [[VEC_IND1:%.*]] = phi <16 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], %[[PRED_STORE_CONTINUE36]] ]
441-
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
441+
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
442442
; OPTSIZE-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
443443
; OPTSIZE-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
444444
; OPTSIZE-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -590,7 +590,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
590590
; OPTSIZE-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
591591
; OPTSIZE-NEXT: br label %[[PRED_STORE_CONTINUE36]]
592592
; OPTSIZE: [[PRED_STORE_CONTINUE36]]:
593-
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
593+
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
594594
; OPTSIZE-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
595595
; OPTSIZE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
596596
; OPTSIZE-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]

0 commit comments

Comments
 (0)