Skip to content

Commit e58cb96

Browse files
fhahnfhahn
committed
[VPlan] Compute induction end values in VPlan.
Use createDerivedIV to compute IV end values directly in VPlan, instead of creating them up-front. This allows updating IV users outside the loop as follow-up. Depends on #110004 and #109975.
1 parent 94c6dd6 commit e58cb96

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

54 files changed

+469
-425
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -233,8 +233,8 @@ class VPBuilder {
233233

234234
VPDerivedIVRecipe *createDerivedIV(InductionDescriptor::InductionKind Kind,
235235
FPMathOperator *FPBinOp, VPValue *Start,
236-
VPCanonicalIVPHIRecipe *CanonicalIV,
237-
VPValue *Step, const Twine &Name = "") {
236+
VPValue *CanonicalIV, VPValue *Step,
237+
const Twine &Name = "") {
238238
return tryInsertInstruction(
239239
new VPDerivedIVRecipe(Kind, FPBinOp, Start, CanonicalIV, Step, Name));
240240
}

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

Lines changed: 130 additions & 104 deletions
Large diffs are not rendered by default.

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -64,6 +64,7 @@ bool VPRecipeBase::mayWriteToMemory() const {
6464
case VPInstruction::FirstOrderRecurrenceSplice:
6565
case VPInstruction::LogicalAnd:
6666
case VPInstruction::PtrAdd:
67+
case VPInstruction::ResumePhi:
6768
return false;
6869
default:
6970
return true;

‎llvm/test/Transforms/LoopVectorize/AArch64/clamped-trip-count.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -13,9 +13,9 @@ define void @clamped_tc_8(ptr nocapture %dst, i32 %n, i64 %val) vscale_range(1,1
1313
; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 8, [[TMP4]]
1414
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
1515
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
16-
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[N_VEC]]
1716
; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
1817
; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 8
18+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[N_VEC]]
1919
; CHECK-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i64(i64 0, i64 8)
2020
; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64()
2121
; CHECK-NEXT: [[TMP9:%.*]] = add <vscale x 8 x i64> [[TMP8]], zeroinitializer
@@ -100,9 +100,9 @@ define void @clamped_tc_max_8(ptr nocapture %dst, i32 %n, i64 %val) vscale_range
100100
; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[WIDE_TRIP_COUNT]], [[TMP4]]
101101
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
102102
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
103-
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[N_VEC]]
104103
; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
105104
; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 8
105+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[N_VEC]]
106106
; CHECK-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i64(i64 0, i64 [[WIDE_TRIP_COUNT]])
107107
; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64()
108108
; CHECK-NEXT: [[TMP9:%.*]] = add <vscale x 8 x i64> [[TMP8]], zeroinitializer

‎llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -821,11 +821,11 @@ define void @multiple_exit_conditions(ptr %src, ptr noalias %dst) #1 {
821821
; PRED-NEXT: [[N_RND_UP:%.*]] = add i64 257, [[TMP2]]
822822
; PRED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
823823
; PRED-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
824+
; PRED-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
825+
; PRED-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 2
824826
; PRED-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 8
825827
; PRED-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP3]]
826828
; PRED-NEXT: [[IND_END1:%.*]] = mul i64 [[N_VEC]], 2
827-
; PRED-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
828-
; PRED-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 2
829829
; PRED-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
830830
; PRED-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 2
831831
; PRED-NEXT: [[TMP8:%.*]] = sub i64 257, [[TMP7]]

‎llvm/test/Transforms/LoopVectorize/AArch64/fixed-order-recurrence.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -48,8 +48,8 @@ define void @firstorderrec(ptr nocapture noundef readonly %x, ptr noalias nocapt
4848
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
4949
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
5050
; CHECK: scalar.ph:
51-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 1, [[FOR_BODY_PREHEADER]] ]
5251
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE]], [[FOR_BODY_PREHEADER]] ]
52+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 1, [[FOR_BODY_PREHEADER]] ]
5353
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
5454
; CHECK: for.cond.cleanup.loopexit:
5555
; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]
@@ -154,10 +154,10 @@ define void @thirdorderrec(ptr nocapture noundef readonly %x, ptr noalias nocapt
154154
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
155155
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
156156
; CHECK: scalar.ph:
157-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 3, [[FOR_BODY_PREHEADER]] ]
158157
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE45]], [[FOR_BODY_PREHEADER]] ]
159158
; CHECK-NEXT: [[SCALAR_RECUR_INIT7:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT6]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE44]], [[FOR_BODY_PREHEADER]] ]
160159
; CHECK-NEXT: [[SCALAR_RECUR_INIT10:%.*]] = phi i8 [ [[VECTOR_RECUR_EXTRACT9]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE]], [[FOR_BODY_PREHEADER]] ]
160+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 3, [[FOR_BODY_PREHEADER]] ]
161161
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
162162
; CHECK: for.cond.cleanup.loopexit:
163163
; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]

‎llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -821,8 +821,8 @@ define void @exit_cond_zext_iv(ptr %dst, i64 %N) {
821821
; PRED-NEXT: [[N_RND_UP:%.*]] = add i64 [[UMAX1]], 1
822822
; PRED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], 2
823823
; PRED-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
824-
; PRED-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC]] to i32
825824
; PRED-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[UMAX1]], 1
825+
; PRED-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC]] to i32
826826
; PRED-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0
827827
; PRED-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT2]], <2 x i64> poison, <2 x i32> zeroinitializer
828828
; PRED-NEXT: br label [[VECTOR_BODY:%.*]]

‎llvm/test/Transforms/LoopVectorize/AArch64/induction-costs.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -333,10 +333,10 @@ define i64 @test_ptr_ivs_and_widened_ivs(ptr %src, i32 %N) {
333333
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
334334
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
335335
; CHECK: scalar.ph:
336-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[SRC]], [[ENTRY:%.*]] ]
336+
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
337+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[SRC]], [[ENTRY]] ]
337338
; CHECK-NEXT: [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
338339
; CHECK-NEXT: [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[IND_END3]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
339-
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
340340
; CHECK-NEXT: br label [[LOOP:%.*]]
341341
; CHECK: loop:
342342
; CHECK-NEXT: [[P:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[SHL:%.*]], [[LOOP]] ]

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

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -105,9 +105,9 @@ define void @low_vf_ic_is_better(ptr nocapture noundef %p, i32 %tc, i16 noundef
105105
; CHECK-VS1-NEXT: [[TMP29:%.*]] = mul i64 [[TMP28]], 8
106106
; CHECK-VS1-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP3]], [[TMP29]]
107107
; CHECK-VS1-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF2]]
108-
; CHECK-VS1-NEXT: [[IND_END:%.*]] = add i64 [[TMP0]], [[N_VEC3]]
109108
; CHECK-VS1-NEXT: [[TMP30:%.*]] = call i64 @llvm.vscale.i64()
110109
; CHECK-VS1-NEXT: [[TMP31:%.*]] = mul i64 [[TMP30]], 8
110+
; CHECK-VS1-NEXT: [[TMP39:%.*]] = add i64 [[TMP0]], [[N_VEC3]]
111111
; CHECK-VS1-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <vscale x 8 x i8> poison, i8 [[CONV]], i64 0
112112
; CHECK-VS1-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <vscale x 8 x i8> [[BROADCAST_SPLATINSERT7]], <vscale x 8 x i8> poison, <vscale x 8 x i32> zeroinitializer
113113
; CHECK-VS1-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
@@ -127,7 +127,7 @@ define void @low_vf_ic_is_better(ptr nocapture noundef %p, i32 %tc, i16 noundef
127127
; CHECK-VS1-NEXT: [[CMP_N10:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC3]]
128128
; CHECK-VS1-NEXT: br i1 [[CMP_N10]], label %[[WHILE_END_LOOPEXIT]], label %[[VEC_EPILOG_SCALAR_PH]]
129129
; CHECK-VS1: [[VEC_EPILOG_SCALAR_PH]]:
130-
; CHECK-VS1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[TMP0]], %[[ITER_CHECK]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
130+
; CHECK-VS1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP39]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[TMP0]], %[[ITER_CHECK]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
131131
; CHECK-VS1-NEXT: br label %[[WHILE_BODY:.*]]
132132
; CHECK-VS1: [[WHILE_BODY]]:
133133
; CHECK-VS1-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[WHILE_BODY]] ]
@@ -213,9 +213,9 @@ define void @low_vf_ic_is_better(ptr nocapture noundef %p, i32 %tc, i16 noundef
213213
; CHECK-VS2-NEXT: [[TMP29:%.*]] = mul i64 [[TMP28]], 4
214214
; CHECK-VS2-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP3]], [[TMP29]]
215215
; CHECK-VS2-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF2]]
216-
; CHECK-VS2-NEXT: [[IND_END:%.*]] = add i64 [[TMP0]], [[N_VEC3]]
217216
; CHECK-VS2-NEXT: [[TMP30:%.*]] = call i64 @llvm.vscale.i64()
218217
; CHECK-VS2-NEXT: [[TMP31:%.*]] = mul i64 [[TMP30]], 4
218+
; CHECK-VS2-NEXT: [[TMP39:%.*]] = add i64 [[TMP0]], [[N_VEC3]]
219219
; CHECK-VS2-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <vscale x 4 x i8> poison, i8 [[CONV]], i64 0
220220
; CHECK-VS2-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <vscale x 4 x i8> [[BROADCAST_SPLATINSERT7]], <vscale x 4 x i8> poison, <vscale x 4 x i32> zeroinitializer
221221
; CHECK-VS2-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
@@ -235,7 +235,7 @@ define void @low_vf_ic_is_better(ptr nocapture noundef %p, i32 %tc, i16 noundef
235235
; CHECK-VS2-NEXT: [[CMP_N10:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC3]]
236236
; CHECK-VS2-NEXT: br i1 [[CMP_N10]], label %[[WHILE_END_LOOPEXIT]], label %[[VEC_EPILOG_SCALAR_PH]]
237237
; CHECK-VS2: [[VEC_EPILOG_SCALAR_PH]]:
238-
; CHECK-VS2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[TMP0]], %[[ITER_CHECK]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
238+
; CHECK-VS2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP39]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[TMP0]], %[[ITER_CHECK]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
239239
; CHECK-VS2-NEXT: br label %[[WHILE_BODY:.*]]
240240
; CHECK-VS2: [[WHILE_BODY]]:
241241
; CHECK-VS2-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[WHILE_BODY]] ]
@@ -279,7 +279,7 @@ while.end:
279279

280280
define void @trip_count_too_small(ptr nocapture noundef %p, i32 noundef %tc, i16 noundef %val) {
281281
; CHECK-LABEL: define void @trip_count_too_small(
282-
; CHECK-SAME: ptr nocapture noundef [[P:%.*]], i32 noundef [[TC:%.*]], i16 noundef [[VAL:%.*]]) #[[ATTR0]] {
282+
; CHECK-SAME: ptr nocapture noundef [[P:%.*]], i32 noundef [[TC:%.*]], i16 noundef [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
283283
; CHECK-NEXT: [[ENTRY:.*:]]
284284
; CHECK-NEXT: [[CMP7:%.*]] = icmp ult i32 [[TC]], 3
285285
; CHECK-NEXT: br i1 [[CMP7]], label %[[WHILE_PREHEADER:.*]], label %[[WHILE_END:.*]]
@@ -428,9 +428,9 @@ define void @overflow_indvar_known_false(ptr nocapture noundef %p, i32 noundef %
428428
; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP1]], [[TMP4]]
429429
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP3]]
430430
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
431-
; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[TMP0]], [[N_VEC]]
432431
; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
433432
; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 16
433+
; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[TMP0]], [[N_VEC]]
434434
; CHECK-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i64(i64 0, i64 [[TMP1]])
435435
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i8> poison, i8 [[CONV]], i64 0
436436
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i8> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i8> poison, <vscale x 16 x i32> zeroinitializer

‎llvm/test/Transforms/LoopVectorize/AArch64/mul-simplification.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,8 +19,8 @@ define i64 @mul_select_operand_known_1_via_scev() {
1919
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vector.reduce.mul.v2i64(<2 x i64> [[VEC_PHI]])
2020
; CHECK-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]]
2121
; CHECK: [[SCALAR_PH]]:
22-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 2, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
2322
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP0]], %[[MIDDLE_BLOCK]] ], [ 12, %[[ENTRY]] ]
23+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 2, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
2424
; CHECK-NEXT: br label %[[LOOP:.*]]
2525
; CHECK: [[LOOP]]:
2626
; CHECK-NEXT: [[RED:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[RED_NEXT:%.*]], %[[LOOP]] ]

‎llvm/test/Transforms/LoopVectorize/AArch64/reduction-recurrence-costs-sve.ll

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -137,9 +137,9 @@ define i32 @chained_recurrences(i32 %x, i64 %y, ptr %src.1, i32 %z, ptr %src.2)
137137
; VSCALEFORTUNING2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
138138
; VSCALEFORTUNING2-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
139139
; VSCALEFORTUNING2: scalar.ph:
140-
; VSCALEFORTUNING2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
141-
; VSCALEFORTUNING2-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ [[TMP24]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
140+
; VSCALEFORTUNING2-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ [[TMP24]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
142141
; VSCALEFORTUNING2-NEXT: [[SCALAR_RECUR_INIT11:%.*]] = phi i32 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
142+
; VSCALEFORTUNING2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
143143
; VSCALEFORTUNING2-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP50]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
144144
; VSCALEFORTUNING2-NEXT: br label [[LOOP:%.*]]
145145
; VSCALEFORTUNING2: loop:
@@ -260,9 +260,9 @@ define i32 @chained_recurrences(i32 %x, i64 %y, ptr %src.1, i32 %z, ptr %src.2)
260260
; PRED-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <vscale x 4 x i32> [[TMP29]], i32 [[TMP47]]
261261
; PRED-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
262262
; PRED: scalar.ph:
263-
; PRED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1:%.*]] ]
264-
; PRED-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ [[TMP28]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1]] ]
263+
; PRED-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ [[TMP28]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1:%.*]] ]
265264
; PRED-NEXT: [[SCALAR_RECUR_INIT8:%.*]] = phi i32 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1]] ]
265+
; PRED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1]] ]
266266
; PRED-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP44]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1]] ]
267267
; PRED-NEXT: br label [[LOOP1:%.*]]
268268
; PRED: loop:

‎llvm/test/Transforms/LoopVectorize/AArch64/scalable-avoid-scalarization.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -23,9 +23,9 @@ define void @test_no_scalarization(ptr %a, ptr noalias %b, i32 %idx, i32 %n) #0
2323
; CHECK-NEXT: [[TMP5:%.*]] = mul i32 [[TMP4]], 2
2424
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP1]], [[TMP5]]
2525
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP1]], [[N_MOD_VF]]
26-
; CHECK-NEXT: [[IND_END:%.*]] = add i32 [[IDX]], [[N_VEC]]
2726
; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vscale.i32()
2827
; CHECK-NEXT: [[TMP7:%.*]] = mul i32 [[TMP6]], 2
28+
; CHECK-NEXT: [[IND_END:%.*]] = add i32 [[IDX]], [[N_VEC]]
2929
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[IDX]], i64 0
3030
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 2 x i32> [[DOTSPLATINSERT]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
3131
; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 2 x i32> @llvm.stepvector.nxv2i32()

0 commit comments

Comments
 (0)