Skip to content

Commit 2688d03

Browse files
fhahnfhahn
committed
!fixup address review comments, thanks
1 parent 11870a2 commit 2688d03

File tree

8 files changed

+29
-53
lines changed

8 files changed

+29
-53
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -7263,9 +7263,9 @@ static void fixReductionScalarResumeWhenVectorizingEpilog(
72637263
((CmpOp == StartV && isGuaranteedNotToBeUndefOrPoison(CmpOp))));
72647264
assert(IsExpectedPattern && "Unexpected reduction resume pattern");
72657265
MainResumeValue = OrigResumeV;
7266-
} else {
7267-
if (auto *VPI = dyn_cast<VPInstruction>(EpiRedHeaderPhi->getStartValue()))
7268-
MainResumeValue = VPI->getOperand(0)->getUnderlyingValue();
7266+
} else if (auto *VPI =
7267+
dyn_cast<VPInstruction>(EpiRedHeaderPhi->getStartValue())) {
7268+
MainResumeValue = VPI->getOperand(0)->getUnderlyingValue();
72697269
}
72707270

72717271
PHINode *MainResumePhi = cast<PHINode>(MainResumeValue);

‎llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -835,9 +835,9 @@ bool VPlanTransforms::legalizeUnclassifiedPhis(VPlan &Plan) {
835835
// The incoming value must be a min/max instrinsic.
836836
// TODO: Also handle the select variant.
837837
Intrinsic::ID ID = Intrinsic::not_intrinsic;
838-
if (auto *WideInt = dyn_cast<VPWidenIntrinsicRecipe>(MinMaxOp))
838+
if (auto *WideInt = dyn_cast<VPWidenIntrinsicRecipe>(MinMaxOp)) {
839839
ID = WideInt->getVectorIntrinsicID();
840-
else {
840+
} else {
841841
auto *RepR = dyn_cast<VPReplicateRecipe>(MinMaxOp);
842842
if (!RepR || !isa<IntrinsicInst>(RepR->getUnderlyingInstr()))
843843
return false;

‎llvm/test/Transforms/LoopVectorize/AArch64/select-index.ll‎

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -62,9 +62,8 @@ define i64 @test_vectorize_select_umin_last_idx(ptr %src, i64 %n) {
6262
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
6363
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
6464
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
65-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
6665
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 2
67-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8
66+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[GEP]], align 8
6867
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8
6968
; CHECK-NEXT: [[TMP3:%.*]] = icmp uge <2 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
7069
; CHECK-NEXT: [[TMP4:%.*]] = icmp uge <2 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -194,9 +193,8 @@ define i64 @test_vectorize_select_smin_last_idx(ptr %src, i64 %n) {
194193
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
195194
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
196195
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
197-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
198196
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 2
199-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8
197+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[GEP]], align 8
200198
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8
201199
; CHECK-NEXT: [[TMP3:%.*]] = icmp sge <2 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
202200
; CHECK-NEXT: [[TMP4:%.*]] = icmp sge <2 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -326,9 +324,8 @@ define i64 @test_vectorize_select_umax_last_idx(ptr %src, i64 %n) {
326324
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
327325
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
328326
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
329-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
330327
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 2
331-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8
328+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[GEP]], align 8
332329
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8
333330
; CHECK-NEXT: [[TMP3:%.*]] = icmp ule <2 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
334331
; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <2 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -458,9 +455,8 @@ define i64 @test_vectorize_select_smax_last_idx(ptr %src, i64 %n) {
458455
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
459456
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
460457
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
461-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
462458
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 2
463-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8
459+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[GEP]], align 8
464460
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8
465461
; CHECK-NEXT: [[TMP3:%.*]] = icmp sle <2 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
466462
; CHECK-NEXT: [[TMP4:%.*]] = icmp sle <2 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]

‎llvm/test/Transforms/LoopVectorize/select-index-interleaving.ll‎

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -62,9 +62,8 @@ define i64 @test_vectorize_select_umin_last_idx(ptr %src, i64 %n) {
6262
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
6363
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
6464
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
65-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
6665
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 4
67-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
66+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
6867
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 4
6968
; CHECK-NEXT: [[TMP3:%.*]] = icmp uge <4 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
7069
; CHECK-NEXT: [[TMP4:%.*]] = icmp uge <4 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -194,9 +193,8 @@ define i64 @test_vectorize_select_smin_last_idx(ptr %src, i64 %n) {
194193
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
195194
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
196195
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
197-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
198196
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 4
199-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
197+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
200198
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 4
201199
; CHECK-NEXT: [[TMP3:%.*]] = icmp sge <4 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
202200
; CHECK-NEXT: [[TMP4:%.*]] = icmp sge <4 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -326,9 +324,8 @@ define i64 @test_vectorize_select_umax_last_idx(ptr %src, i64 %n) {
326324
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
327325
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
328326
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
329-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
330327
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 4
331-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
328+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
332329
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 4
333330
; CHECK-NEXT: [[TMP3:%.*]] = icmp ule <4 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
334331
; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <4 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]
@@ -458,9 +455,8 @@ define i64 @test_vectorize_select_smax_last_idx(ptr %src, i64 %n) {
458455
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
459456
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
460457
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
461-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
462458
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[GEP]], i32 4
463-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
459+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
464460
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 4
465461
; CHECK-NEXT: [[TMP3:%.*]] = icmp sle <4 x i64> [[VEC_PHI2]], [[WIDE_LOAD]]
466462
; CHECK-NEXT: [[TMP4:%.*]] = icmp sle <4 x i64> [[VEC_PHI3]], [[WIDE_LOAD4]]

‎llvm/test/Transforms/LoopVectorize/select-smax-last-index.ll‎

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -17,8 +17,7 @@ define i64 @test_vectorize_select_smax_idx(ptr %src, i64 %n) {
1717
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
1818
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
1919
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
20-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
21-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
20+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
2221
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
2322
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
2423
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -95,8 +94,7 @@ define i64 @test_vectorize_select_smax_idx_cond_flipped(ptr %src, i64 %n) {
9594
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
9695
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
9796
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
98-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
99-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
97+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
10098
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <4 x i64> [[WIDE_LOAD]], [[VEC_PHI1]]
10199
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
102100
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -303,8 +301,7 @@ define i64 @test_vectorize_select_smax_idx_min_ops_switched(ptr %src, i64 %n) {
303301
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
304302
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
305303
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
306-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
307-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
304+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
308305
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
309306
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smax.v4i64(<4 x i64> [[WIDE_LOAD]], <4 x i64> [[VEC_PHI1]])
310307
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -425,8 +422,7 @@ define i64 @test_cmp_and_smax_use_different_values(ptr %src, i64 %x, i64 %n) {
425422
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
426423
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
427424
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
428-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
429-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
425+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
430426
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <4 x i64> [[VEC_PHI1]], [[BROADCAST_SPLAT]]
431427
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
432428
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]

‎llvm/test/Transforms/LoopVectorize/select-smin-last-index.ll‎

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -19,8 +19,7 @@ define i64 @test_vectorize_select_smin_idx(ptr %src, i64 %n) {
1919
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
2020
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
2121
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
22-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
23-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
22+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
2423
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
2524
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smin.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
2625
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -97,8 +96,7 @@ define i64 @test_vectorize_select_smin_idx_cond_flipped(ptr %src, i64 %n) {
9796
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
9897
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
9998
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
100-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
101-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
99+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
102100
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <4 x i64> [[WIDE_LOAD]], [[VEC_PHI1]]
103101
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smin.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
104102
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -305,8 +303,7 @@ define i64 @test_vectorize_select_smin_idx_min_ops_switched(ptr %src, i64 %n) {
305303
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
306304
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
307305
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
308-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
309-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
306+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
310307
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
311308
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smin.v4i64(<4 x i64> [[WIDE_LOAD]], <4 x i64> [[VEC_PHI1]])
312309
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -427,8 +424,7 @@ define i64 @test_cmp_and_smin_use_different_values(ptr %src, i64 %x, i64 %n) {
427424
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
428425
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
429426
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
430-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
431-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
427+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
432428
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <4 x i64> [[VEC_PHI1]], [[BROADCAST_SPLAT]]
433429
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.smin.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
434430
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]

‎llvm/test/Transforms/LoopVectorize/select-umax-last-index.ll‎

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -17,8 +17,7 @@ define i64 @test_vectorize_select_umax_idx(ptr %src, i64 %n) {
1717
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
1818
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
1919
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
20-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
21-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
20+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
2221
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
2322
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.umax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
2423
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -95,8 +94,7 @@ define i64 @test_vectorize_select_umax_idx_cond_flipped(ptr %src, i64 %n) {
9594
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
9695
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
9796
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV]]
98-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP]], i32 0
99-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
97+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP]], align 4
10098
; CHECK-NEXT: [[TMP2:%.*]] = icmp uge <4 x i64> [[WIDE_LOAD]], [[VEC_PHI1]]
10199
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.umax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
102100
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -303,8 +301,7 @@ define i64 @test_vectorize_select_umax_idx_min_ops_switched(ptr %src, i64 %n) {
303301
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
304302
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
305303
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
306-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
307-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
304+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
308305
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_PHI1]], [[WIDE_LOAD]]
309306
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.umax.v4i64(<4 x i64> [[WIDE_LOAD]], <4 x i64> [[VEC_PHI1]])
310307
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]
@@ -425,8 +422,7 @@ define i64 @test_cmp_and_umax_use_different_values(ptr %src, i64 %x, i64 %n) {
425422
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
426423
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
427424
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[IV1]]
428-
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[GEP1]], i32 0
429-
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP1]], align 4
425+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[GEP1]], align 4
430426
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i64> [[VEC_PHI1]], [[BROADCAST_SPLAT]]
431427
; CHECK-NEXT: [[TMP3]] = call <4 x i64> @llvm.umax.v4i64(<4 x i64> [[VEC_PHI1]], <4 x i64> [[WIDE_LOAD]])
432428
; CHECK-NEXT: [[TMP4]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]

0 commit comments

Comments
 (0)