Skip to content

Commit a114020

Browse files
SamTebbs33SamTebbs33
committed
Rebase
1 parent 87008e3 commit a114020

File tree

4 files changed

+64
-92
lines changed

4 files changed

+64
-92
lines changed

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

Lines changed: 12 additions & 46 deletions
Original file line numberDiff line numberDiff line change
@@ -522,40 +522,38 @@ define void @multiple_exit_conditions(ptr %src, ptr noalias %dst) #1 {
522522
; DEFAULT-LABEL: define void @multiple_exit_conditions(
523523
; DEFAULT-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) #[[ATTR2:[0-9]+]] {
524524
; DEFAULT-NEXT: [[ENTRY:.*:]]
525-
; DEFAULT-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
526-
; DEFAULT-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP0]], 1
527-
; DEFAULT-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 257, [[TMP6]]
528-
; DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]
529-
; DEFAULT: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:
530525
; DEFAULT-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
531526
; DEFAULT-NEXT: [[TMP3:%.*]] = shl nuw i64 [[TMP2]], 4
532527
; DEFAULT-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 257, [[TMP3]]
533-
; DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK1]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
528+
; DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK1]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
534529
; DEFAULT: [[VECTOR_PH]]:
535530
; DEFAULT-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
536531
; DEFAULT-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 16
537532
; DEFAULT-NEXT: [[N_MOD_VF:%.*]] = urem i64 257, [[TMP5]]
538533
; DEFAULT-NEXT: [[N_VEC:%.*]] = sub i64 257, [[N_MOD_VF]]
534+
; DEFAULT-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[N_VEC]], 8
535+
; DEFAULT-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX]]
536+
; DEFAULT-NEXT: [[TMP6:%.*]] = mul i64 [[N_VEC]], 2
539537
; DEFAULT-NEXT: br label %[[VECTOR_BODY:.*]]
540538
; DEFAULT: [[VECTOR_BODY]]:
541539
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
542-
; DEFAULT-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
543-
; DEFAULT-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX]]
540+
; DEFAULT-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX]], 8
541+
; DEFAULT-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX1]]
544542
; DEFAULT-NEXT: [[TMP1:%.*]] = load i16, ptr [[SRC]], align 2
545543
; DEFAULT-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i16> poison, i16 [[TMP1]], i64 0
546544
; DEFAULT-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i16> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i16> poison, <vscale x 4 x i32> zeroinitializer
547545
; DEFAULT-NEXT: [[TMP8:%.*]] = or <vscale x 4 x i16> [[BROADCAST_SPLAT]], splat (i16 1)
548546
; DEFAULT-NEXT: [[TMP9:%.*]] = uitofp <vscale x 4 x i16> [[TMP8]] to <vscale x 4 x double>
549547
; DEFAULT-NEXT: [[TMP10:%.*]] = call i64 @llvm.vscale.i64()
550548
; DEFAULT-NEXT: [[TMP11:%.*]] = shl nuw i64 [[TMP10]], 2
551-
; DEFAULT-NEXT: [[TMP12:%.*]] = getelementptr double, ptr [[NEXT_GEP]], i64 [[TMP11]]
549+
; DEFAULT-NEXT: [[TMP12:%.*]] = getelementptr double, ptr [[NEXT_GEP1]], i64 [[TMP11]]
552550
; DEFAULT-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
553551
; DEFAULT-NEXT: [[TMP14:%.*]] = shl nuw i64 [[TMP13]], 3
554-
; DEFAULT-NEXT: [[TMP15:%.*]] = getelementptr double, ptr [[NEXT_GEP]], i64 [[TMP14]]
552+
; DEFAULT-NEXT: [[TMP15:%.*]] = getelementptr double, ptr [[NEXT_GEP1]], i64 [[TMP14]]
555553
; DEFAULT-NEXT: [[TMP16:%.*]] = call i64 @llvm.vscale.i64()
556554
; DEFAULT-NEXT: [[TMP17:%.*]] = mul nuw i64 [[TMP16]], 12
557-
; DEFAULT-NEXT: [[TMP18:%.*]] = getelementptr double, ptr [[NEXT_GEP]], i64 [[TMP17]]
558-
; DEFAULT-NEXT: store <vscale x 4 x double> [[TMP9]], ptr [[NEXT_GEP]], align 8
555+
; DEFAULT-NEXT: [[TMP18:%.*]] = getelementptr double, ptr [[NEXT_GEP1]], i64 [[TMP17]]
556+
; DEFAULT-NEXT: store <vscale x 4 x double> [[TMP9]], ptr [[NEXT_GEP1]], align 8
559557
; DEFAULT-NEXT: store <vscale x 4 x double> [[TMP9]], ptr [[TMP12]], align 8
560558
; DEFAULT-NEXT: store <vscale x 4 x double> [[TMP9]], ptr [[TMP15]], align 8
561559
; DEFAULT-NEXT: store <vscale x 4 x double> [[TMP9]], ptr [[TMP18]], align 8
@@ -564,40 +562,8 @@ define void @multiple_exit_conditions(ptr %src, ptr noalias %dst) #1 {
564562
; DEFAULT-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
565563
; DEFAULT: [[MIDDLE_BLOCK]]:
566564
; DEFAULT-NEXT: [[CMP_N:%.*]] = icmp eq i64 257, [[N_VEC]]
567-
; DEFAULT-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]
568-
; DEFAULT: [[VEC_EPILOG_ITER_CHECK]]:
569-
; DEFAULT-NEXT: [[TMP20:%.*]] = mul i64 [[N_VEC]], 8
570-
; DEFAULT-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP20]]
571-
; DEFAULT-NEXT: [[IND_END11:%.*]] = mul i64 [[N_VEC]], 2
572-
; DEFAULT-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], [[TMP6]]
573-
; DEFAULT-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF24:![0-9]+]]
574-
; DEFAULT: [[VEC_EPILOG_PH]]:
575-
; DEFAULT-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
576-
; DEFAULT-NEXT: [[TMP21:%.*]] = call i64 @llvm.vscale.i64()
577-
; DEFAULT-NEXT: [[TMP22:%.*]] = mul nuw i64 [[TMP21]], 2
578-
; DEFAULT-NEXT: [[N_MOD_VF2:%.*]] = urem i64 257, [[TMP22]]
579-
; DEFAULT-NEXT: [[N_VEC3:%.*]] = sub i64 257, [[N_MOD_VF2]]
580-
; DEFAULT-NEXT: [[TMP23:%.*]] = mul i64 [[N_VEC3]], 8
581-
; DEFAULT-NEXT: [[TMP24:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP23]]
582-
; DEFAULT-NEXT: [[TMP25:%.*]] = mul i64 [[N_VEC3]], 2
583-
; DEFAULT-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
584-
; DEFAULT: [[VEC_EPILOG_VECTOR_BODY]]:
585-
; DEFAULT-NEXT: [[INDEX4:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT8:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
586-
; DEFAULT-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX4]], 8
587-
; DEFAULT-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX1]]
588-
; DEFAULT-NEXT: [[TMP26:%.*]] = load i16, ptr [[SRC]], align 2
589-
; DEFAULT-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <vscale x 2 x i16> poison, i16 [[TMP26]], i64 0
590-
; DEFAULT-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <vscale x 2 x i16> [[BROADCAST_SPLATINSERT6]], <vscale x 2 x i16> poison, <vscale x 2 x i32> zeroinitializer
591-
; DEFAULT-NEXT: [[TMP27:%.*]] = or <vscale x 2 x i16> [[BROADCAST_SPLAT7]], splat (i16 1)
592-
; DEFAULT-NEXT: [[TMP28:%.*]] = uitofp <vscale x 2 x i16> [[TMP27]] to <vscale x 2 x double>
593-
; DEFAULT-NEXT: store <vscale x 2 x double> [[TMP28]], ptr [[NEXT_GEP5]], align 8
594-
; DEFAULT-NEXT: [[INDEX_NEXT8]] = add nuw i64 [[INDEX4]], [[TMP22]]
595-
; DEFAULT-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT8]], [[N_VEC3]]
596-
; DEFAULT-NEXT: br i1 [[TMP29]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
597-
; DEFAULT: [[VEC_EPILOG_MIDDLE_BLOCK]]:
598-
; DEFAULT-NEXT: [[CMP_N9:%.*]] = icmp eq i64 257, [[N_VEC3]]
599-
; DEFAULT-NEXT: br i1 [[CMP_N9]], [[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]]
600-
; DEFAULT: [[VEC_EPILOG_SCALAR_PH]]:
565+
; DEFAULT-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
566+
; DEFAULT: [[SCALAR_PH]]:
601567
;
602568
; PRED-LABEL: define void @multiple_exit_conditions(
603569
; PRED-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) #[[ATTR2:[0-9]+]] {

‎llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll‎

Lines changed: 20 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -174,16 +174,16 @@ define i64 @not_dotp_i8_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %b
174174
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE14:%.*]], [[VECTOR_BODY]] ]
175175
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
176176
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
177-
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 16
178-
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 32
179-
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 48
177+
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 16
178+
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 32
179+
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 48
180180
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
181181
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
182182
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
183183
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
184-
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 16
185-
; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 32
186-
; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 48
184+
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 16
185+
; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 32
186+
; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 48
187187
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
188188
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
189189
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP18]], align 1
@@ -311,16 +311,16 @@ define i64 @not_dotp_i16_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %
311311
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]]
312312
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX2:%.*]] = mul i64 [[INDEX]], 2
313313
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX2]]
314-
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 8
315-
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 16
316-
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 24
314+
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 8
315+
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 16
316+
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 24
317317
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
318318
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i16>, ptr [[TMP0]], align 2
319319
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <8 x i16>, ptr [[TMP10]], align 2
320320
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load <8 x i16>, ptr [[TMP11]], align 2
321-
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 8
322-
; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 16
323-
; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 24
321+
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 8
322+
; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 16
323+
; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 24
324324
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <8 x i16>, ptr [[NEXT_GEP3]], align 2
325325
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2
326326
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD11:%.*]] = load <8 x i16>, ptr [[TMP18]], align 2
@@ -537,7 +537,7 @@ define i32 @not_dotp_different_types(ptr %a, ptr %b) #0 {
537537
; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = add i64 [[INDEX]], 30
538538
; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = add i64 [[INDEX]], 31
539539
; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP0]]
540-
; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = getelementptr i8, ptr [[TMP32]], i32 16
540+
; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = getelementptr i8, ptr [[TMP32]], i64 16
541541
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP32]], align 1
542542
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP34]], align 1
543543
; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
@@ -801,13 +801,13 @@ define i32 @not_dotp_not_loop_carried(ptr %a, ptr %b) #0 {
801801
; CHECK-INTERLEAVED: vector.body:
802802
; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
803803
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
804-
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP10]], i32 16
804+
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP10]], i64 16
805805
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
806806
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
807807
; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
808808
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
809809
; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
810-
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP17]], i32 16
810+
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP17]], i64 16
811811
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP17]], align 1
812812
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
813813
; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
@@ -1330,10 +1330,10 @@ define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) #0 {
13301330
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
13311331
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]]
13321332
; CHECK-INTERLEAVE1-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
1333-
; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
13341333
; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]]
13351334
; CHECK-INTERLEAVE1-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
13361335
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD1]] to <vscale x 16 x i32>
1336+
; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
13371337
; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = mul nsw <vscale x 16 x i32> [[TMP10]], [[TMP13]]
13381338
; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = select <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i32> [[TMP15]], <vscale x 16 x i32> zeroinitializer
13391339
; CHECK-INTERLEAVE1-NEXT: [[TMP19]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP12]])
@@ -1368,10 +1368,10 @@ define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) #0 {
13681368
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
13691369
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]]
13701370
; CHECK-INTERLEAVED-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
1371-
; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
13721371
; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]]
13731372
; CHECK-INTERLEAVED-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
13741373
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD1]] to <vscale x 16 x i32>
1374+
; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
13751375
; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = mul nsw <vscale x 16 x i32> [[TMP10]], [[TMP13]]
13761376
; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = select <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i32> [[TMP15]], <vscale x 16 x i32> zeroinitializer
13771377
; CHECK-INTERLEAVED-NEXT: [[TMP19]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP12]])
@@ -1406,10 +1406,10 @@ define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) #0 {
14061406
; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
14071407
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]]
14081408
; CHECK-MAXBW-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
1409-
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
14101409
; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]]
14111410
; CHECK-MAXBW-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i8> poison)
14121411
; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD1]] to <vscale x 16 x i32>
1412+
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 16 x i8> [[WIDE_MASKED_LOAD]] to <vscale x 16 x i32>
14131413
; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 16 x i32> [[TMP16]], [[TMP13]]
14141414
; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = select <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i32> [[TMP17]], <vscale x 16 x i32> zeroinitializer
14151415
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP18]])
@@ -1953,7 +1953,7 @@ define i64 @not_dotp_ext_outside_plan(ptr %a, i16 %b, i64 %n) #0 {
19531953
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ]
19541954
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
19551955
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]]
1956-
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i32 8
1956+
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i64 8
19571957
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2
19581958
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2
19591959
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
@@ -2082,7 +2082,7 @@ define i64 @not_dotp_ext_outside_plan2(ptr %a, i16 %b, i64 %n) #0 {
20822082
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ]
20832083
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
20842084
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]]
2085-
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i32 8
2085+
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i64 8
20862086
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2
20872087
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2
20882088
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>

0 commit comments

Comments
 (0)