Skip to content

Commit 135d7ab

Browse files
hazzlimhazzlim
committed
Address Review
1 parent 30b5939 commit 135d7ab

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
@@ -1098,6 +1098,8 @@ static bool optimizeVectorInductionWidthForTCAndVFUF(VPlan &Plan,
10981098
if (!Plan.getVectorLoopRegion())
10991099
return false;
11001100

1101+
if (!Plan.getTripCount()->isLiveIn())
1102+
return false;
11011103
auto *TC = dyn_cast_if_present<ConstantInt>(
11021104
Plan.getTripCount()->getUnderlyingValue());
11031105
if (!TC || !BestVF.isFixed())
@@ -1124,8 +1126,8 @@ static bool optimizeVectorInductionWidthForTCAndVFUF(VPlan &Plan,
11241126
auto *WideIV = dyn_cast<VPWidenIntOrFpInductionRecipe>(&Phi);
11251127

11261128
// Currently only handle canonical IVs as it is trivial to replace the start
1127-
// and stop values, and we only perform the optimisation when the IV is only
1128-
// used by the comparison controlling loop control-flow.
1129+
// and stop values, and we currently only perform the optimization when the
1130+
// IV has a single use.
11291131
if (!WideIV || !WideIV->isCanonical() ||
11301132
WideIV->hasMoreThanOneUniqueUser() ||
11311133
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
@@ -244,9 +244,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
244244
; DEFAULT-NEXT: br label %[[VECTOR_BODY:.*]]
245245
; DEFAULT: [[VECTOR_BODY]]:
246246
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
247-
; 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]] ]
247+
; 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]] ]
248248
; 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]] ]
249-
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
249+
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
250250
; DEFAULT-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
251251
; DEFAULT-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
252252
; DEFAULT-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -398,7 +398,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
398398
; DEFAULT-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
399399
; DEFAULT-NEXT: br label %[[PRED_STORE_CONTINUE36]]
400400
; DEFAULT: [[PRED_STORE_CONTINUE36]]:
401-
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
401+
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
402402
; DEFAULT-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
403403
; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
404404
; DEFAULT-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
@@ -439,9 +439,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
439439
; OPTSIZE-NEXT: br label %[[VECTOR_BODY:.*]]
440440
; OPTSIZE: [[VECTOR_BODY]]:
441441
; OPTSIZE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
442-
; 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]] ]
442+
; 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]] ]
443443
; 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]] ]
444-
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
444+
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
445445
; OPTSIZE-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
446446
; OPTSIZE-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
447447
; OPTSIZE-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -593,7 +593,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
593593
; OPTSIZE-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
594594
; OPTSIZE-NEXT: br label %[[PRED_STORE_CONTINUE36]]
595595
; OPTSIZE: [[PRED_STORE_CONTINUE36]]:
596-
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
596+
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
597597
; OPTSIZE-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
598598
; OPTSIZE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
599599
; 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
@@ -240,9 +240,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
240240
; DEFAULT-NEXT: br label %[[VECTOR_BODY:.*]]
241241
; DEFAULT: [[VECTOR_BODY]]:
242242
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
243-
; 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]] ]
243+
; 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]] ]
244244
; 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]] ]
245-
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
245+
; DEFAULT-NEXT: [[TMP0:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
246246
; DEFAULT-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
247247
; DEFAULT-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
248248
; DEFAULT-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -394,7 +394,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
394394
; DEFAULT-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
395395
; DEFAULT-NEXT: br label %[[PRED_STORE_CONTINUE36]]
396396
; DEFAULT: [[PRED_STORE_CONTINUE36]]:
397-
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
397+
; DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
398398
; DEFAULT-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
399399
; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
400400
; DEFAULT-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
@@ -435,9 +435,9 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
435435
; OPTSIZE-NEXT: br label %[[VECTOR_BODY:.*]]
436436
; OPTSIZE: [[VECTOR_BODY]]:
437437
; OPTSIZE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_STORE_CONTINUE36:.*]] ]
438-
; 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]] ]
438+
; 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]] ]
439439
; 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]] ]
440-
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i64> [[VEC_IND]], splat (i64 14)
440+
; OPTSIZE-NEXT: [[TMP72:%.*]] = icmp ule <16 x i8> [[VEC_IND]], splat (i8 14)
441441
; OPTSIZE-NEXT: [[TMP1:%.*]] = mul <16 x i8> [[BROADCAST_SPLAT]], [[VEC_IND1]]
442442
; OPTSIZE-NEXT: [[TMP2:%.*]] = lshr <16 x i8> [[VEC_IND1]], splat (i8 1)
443443
; OPTSIZE-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP2]], [[BROADCAST_SPLAT4]]
@@ -589,7 +589,7 @@ define void @tail_predicate_without_optsize(ptr %p, i8 %a, i8 %b, i8 %c, i32 %n)
589589
; OPTSIZE-NEXT: store i8 [[TMP71]], ptr [[TMP70]], align 1
590590
; OPTSIZE-NEXT: br label %[[PRED_STORE_CONTINUE36]]
591591
; OPTSIZE: [[PRED_STORE_CONTINUE36]]:
592-
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
592+
; OPTSIZE-NEXT: [[VEC_IND_NEXT]] = add <16 x i8> [[VEC_IND]], splat (i8 16)
593593
; OPTSIZE-NEXT: [[VEC_IND_NEXT2]] = add <16 x i8> [[VEC_IND1]], splat (i8 16)
594594
; OPTSIZE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
595595
; OPTSIZE-NEXT: br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]

0 commit comments

Comments
 (0)