Skip to content

Commit 0877a4c

Browse files
fhahnfhahn
committed
[VPlan] Remove PtrIV::IsScalarAfterVectorization, use VPlan analysis.
1 parent 9da6627 commit 0877a4c

File tree

8 files changed

+41
-63
lines changed

8 files changed

+41
-63
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp‎

Lines changed: 5 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -7655,7 +7655,7 @@ createWidenInductionRecipes(VPInstruction *PhiR,
76557655
}
76567656

76577657
VPHeaderPHIRecipe *
7658-
VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI, VFRange &Range) {
7658+
VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI) {
76597659
auto *Phi = cast<PHINode>(VPI->getUnderlyingInstr());
76607660

76617661
// Check if this is an integer or fp induction. If so, build the recipe that
@@ -7666,14 +7666,9 @@ VPRecipeBuilder::tryToOptimizeInductionPHI(VPInstruction *VPI, VFRange &Range) {
76667666
// Check if this is pointer induction. If so, build the recipe for it.
76677667
if (auto *II = Legal->getPointerInductionDescriptor(Phi)) {
76687668
VPValue *Step = vputils::getOrCreateVPValueForSCEVExpr(Plan, II->getStep());
7669-
return new VPWidenPointerInductionRecipe(
7670-
Phi, VPI->getOperand(0), Step, &Plan.getVFxUF(), *II,
7671-
LoopVectorizationPlanner::getDecisionAndClampRange(
7672-
[&](ElementCount VF) {
7673-
return CM.isScalarAfterVectorization(Phi, VF);
7674-
},
7675-
Range),
7676-
VPI->getDebugLoc());
7669+
return new VPWidenPointerInductionRecipe(Phi, VPI->getOperand(0), Step,
7670+
&Plan.getVFxUF(), *II,
7671+
VPI->getDebugLoc());
76777672
}
76787673
return nullptr;
76797674
}
@@ -8162,7 +8157,7 @@ VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(VPSingleDefRecipe *R,
81628157
"Non-header phis should have been handled during predication");
81638158
auto *Phi = cast<PHINode>(R->getUnderlyingInstr());
81648159
assert(R->getNumOperands() == 2 && "Must have 2 operands for header phis");
8165-
if ((Recipe = tryToOptimizeInductionPHI(PhiR, Range)))
8160+
if ((Recipe = tryToOptimizeInductionPHI(PhiR)))
81668161
return Recipe;
81678162

81688163
VPHeaderPHIRecipe *PhiRecipe = nullptr;

‎llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h‎

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -100,8 +100,7 @@ class VPRecipeBuilder {
100100

101101
/// Check if an induction recipe should be constructed for \p VPI. If so build
102102
/// and return it. If not, return null.
103-
VPHeaderPHIRecipe *tryToOptimizeInductionPHI(VPInstruction *VPI,
104-
VFRange &Range);
103+
VPHeaderPHIRecipe *tryToOptimizeInductionPHI(VPInstruction *VPI);
105104

106105
/// Optimize the special case where the operand of \p VPI is a constant
107106
/// integer induction variable.

‎llvm/lib/Transforms/Vectorize/VPlan.h‎

Lines changed: 3 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -2213,19 +2213,15 @@ class VPWidenIntOrFpInductionRecipe : public VPWidenInductionRecipe {
22132213
};
22142214

22152215
class VPWidenPointerInductionRecipe : public VPWidenInductionRecipe {
2216-
bool IsScalarAfterVectorization;
2217-
22182216
public:
22192217
/// Create a new VPWidenPointerInductionRecipe for \p Phi with start value \p
22202218
/// Start and the number of elements unrolled \p NumUnrolledElems, typically
22212219
/// VF*UF.
22222220
VPWidenPointerInductionRecipe(PHINode *Phi, VPValue *Start, VPValue *Step,
22232221
VPValue *NumUnrolledElems,
2224-
const InductionDescriptor &IndDesc,
2225-
bool IsScalarAfterVectorization, DebugLoc DL)
2222+
const InductionDescriptor &IndDesc, DebugLoc DL)
22262223
: VPWidenInductionRecipe(VPDef::VPWidenPointerInductionSC, Phi, Start,
2227-
Step, IndDesc, DL),
2228-
IsScalarAfterVectorization(IsScalarAfterVectorization) {
2224+
Step, IndDesc, DL) {
22292225
addOperand(NumUnrolledElems);
22302226
}
22312227

@@ -2234,8 +2230,7 @@ class VPWidenPointerInductionRecipe : public VPWidenInductionRecipe {
22342230
VPWidenPointerInductionRecipe *clone() override {
22352231
return new VPWidenPointerInductionRecipe(
22362232
cast<PHINode>(getUnderlyingInstr()), getOperand(0), getOperand(1),
2237-
getOperand(2), getInductionDescriptor(), IsScalarAfterVectorization,
2238-
getDebugLoc());
2233+
getOperand(2), getInductionDescriptor(), getDebugLoc());
22392234
}
22402235

22412236
VP_CLASSOF_IMPL(VPDef::VPWidenPointerInductionSC)

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4310,7 +4310,7 @@ void VPCanonicalIVPHIRecipe::print(raw_ostream &O, const Twine &Indent,
43104310
#endif
43114311

43124312
bool VPWidenPointerInductionRecipe::onlyScalarsGenerated(bool IsScalable) {
4313-
return IsScalarAfterVectorization &&
4313+
return vputils::onlyScalarValuesUsed(this) &&
43144314
(!IsScalable || vputils::onlyFirstLaneUsed(this));
43154315
}
43164316

‎llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll‎

Lines changed: 16 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -11,21 +11,14 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) {
1111
; CHECK: vector.main.loop.iter.check:
1212
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
1313
; CHECK: vector.ph:
14+
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 10000
1415
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
1516
; CHECK: vector.body:
1617
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
17-
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
18-
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
19-
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
20-
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
21-
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 [[TMP0]]
22-
; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP1]]
23-
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP]], i32 0
24-
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x ptr> [[TMP4]], ptr [[NEXT_GEP1]], i32 1
25-
; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP2]]
26-
; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP3]]
27-
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP2]], i32 0
28-
; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x ptr> [[TMP6]], ptr [[NEXT_GEP3]], i32 1
18+
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[PTR_START_1]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
19+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <2 x i64> <i64 0, i64 1>
20+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = extractelement <2 x ptr> [[TMP5]], i32 0
21+
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, <2 x ptr> [[TMP5]], <2 x i64> splat (i64 2)
2922
; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <2 x ptr> [[TMP5]], zeroinitializer
3023
; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP8]], i32 0
3124
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[TMP8]], i32 1
@@ -40,32 +33,32 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) {
4033
; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[NEXT_GEP]], align 1
4134
; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP15]], align 1
4235
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
36+
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 4
4337
; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
4438
; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
4539
; CHECK: middle.block:
4640
; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
4741
; CHECK: vec.epilog.iter.check:
4842
; CHECK-NEXT: [[IND_END4:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
49-
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
43+
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]
5044
; CHECK: vec.epilog.ph:
5145
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
46+
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[TMP0]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START_1]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
5247
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
5348
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
5449
; CHECK: vec.epilog.vector.body:
5550
; CHECK-NEXT: [[INDEX6:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
56-
; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[INDEX6]], 0
57-
; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[INDEX6]], 1
58-
; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP17]]
59-
; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP18]]
60-
; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP7]], i32 0
61-
; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x ptr> [[TMP19]], ptr [[NEXT_GEP8]], i32 1
51+
; CHECK-NEXT: [[POINTER_PHI2:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], [[VEC_EPILOG_PH]] ], [ [[PTR_IND5:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
52+
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[POINTER_PHI2]], <2 x i64> <i64 0, i64 1>
53+
; CHECK-NEXT: [[NEXT_GEP7:%.*]] = extractelement <2 x ptr> [[TMP20]], i32 0
6254
; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <2 x ptr> [[TMP20]], zeroinitializer
6355
; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x i1> [[TMP21]], i32 0
6456
; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP21]], i32 1
6557
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP22]])
6658
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP23]])
6759
; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[NEXT_GEP7]], align 1
6860
; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX6]], 2
61+
; CHECK-NEXT: [[PTR_IND5]] = getelementptr i8, ptr [[POINTER_PHI2]], i64 2
6962
; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT9]], 10000
7063
; CHECK-NEXT: br i1 [[TMP25]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
7164
; CHECK: vec.epilog.middle.block:
@@ -136,7 +129,7 @@ define void @test_widen_induction(ptr %A, i64 %N) {
136129
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
137130
; CHECK: vec.epilog.iter.check:
138131
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2
139-
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
132+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]]
140133
; CHECK: vec.epilog.ph:
141134
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
142135
; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 2
@@ -221,7 +214,7 @@ define void @test_widen_induction_variable_start(ptr %A, i64 %N, i64 %start) {
221214
; CHECK: vec.epilog.iter.check:
222215
; CHECK-NEXT: [[IND_END5:%.*]] = add i64 [[START]], [[N_VEC]]
223216
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2
224-
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
217+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]]
225218
; CHECK: vec.epilog.ph:
226219
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
227220
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
@@ -304,7 +297,7 @@ define void @test_widen_induction_step_2(ptr %A, i64 %N, i32 %step) {
304297
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
305298
; CHECK: vec.epilog.iter.check:
306299
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2
307-
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
300+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]]
308301
; CHECK: vec.epilog.ph:
309302
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
310303
; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 2
@@ -416,7 +409,7 @@ define void @test_widen_truncated_induction(ptr %A) {
416409
; CHECK: middle.block:
417410
; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
418411
; CHECK: vec.epilog.iter.check:
419-
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
412+
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3]]
420413
; CHECK: vec.epilog.ph:
421414
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
422415
; CHECK-NEXT: [[TMP5:%.*]] = trunc i64 [[VEC_EPILOG_RESUME_VAL]] to i8

‎llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll‎

Lines changed: 1 addition & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -112,15 +112,11 @@ define void @pointer_induction(ptr noalias %start, i64 %N) {
112112
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
113113
; CHECK: vector.body:
114114
; CHECK-NEXT: [[INDEX2:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
115-
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
116-
; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
117-
; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[TMP12]]
118-
; CHECK-NEXT: [[TMP15:%.*]] = extractelement <vscale x 2 x ptr> [[VECTOR_GEP]], i32 0
115+
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[START]], i64 [[INDEX2]]
119116
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i8>, ptr [[TMP15]], align 1
120117
; CHECK-NEXT: [[TMP17:%.*]] = add <vscale x 2 x i8> [[WIDE_LOAD]], splat (i8 1)
121118
; CHECK-NEXT: store <vscale x 2 x i8> [[TMP17]], ptr [[TMP15]], align 1
122119
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX2]], [[TMP4]]
123-
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP4]]
124120
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
125121
; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
126122
; CHECK: middle.block:

‎llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll‎

Lines changed: 4 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -25,17 +25,14 @@ define void @test_pr55375_interleave_opaque_ptr(ptr %start, ptr %end) {
2525
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
2626
; CHECK: vector.body:
2727
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
28-
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16
29-
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0
30-
; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 16
31-
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
32-
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
33-
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP7]], i32 0
34-
; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x ptr> [[TMP9]], ptr [[TMP8]], i32 1
28+
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
29+
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <2 x i64> <i64 0, i64 16>
30+
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x ptr> [[TMP10]], i32 0
3531
; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x ptr> zeroinitializer, <2 x ptr> [[TMP10]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
3632
; CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x ptr> [[TMP12]], <4 x ptr> poison, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
3733
; CHECK-NEXT: store <4 x ptr> [[INTERLEAVED_VEC]], ptr [[TMP7]], align 8
3834
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
35+
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 32
3936
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4037
; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
4138
; CHECK: middle.block:

‎llvm/test/Transforms/LoopVectorize/pointer-induction-index-width-smaller-than-iv-width.ll‎

Lines changed: 10 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -15,16 +15,20 @@ define void @wide_ptr_induction_index_width_smaller_than_iv_width(ptr noalias %s
1515
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
1616
; CHECK: [[VECTOR_BODY]]:
1717
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
18-
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[SRC]], %[[VECTOR_PH]] ], [ [[PTR_IND:%.*]], %[[VECTOR_BODY]] ]
19-
; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i32> <i32 0, i32 8, i32 16, i32 24>
20-
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 0
21-
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 1
22-
; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 2
23-
; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x ptr> [[VECTOR_GEP]], i32 3
2418
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 0
2519
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1
2620
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 2
2721
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 3
22+
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[INDEX]] to i32
23+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i32 [[DOTCAST]], 8
24+
; CHECK-NEXT: [[TMP11:%.*]] = add i32 [[OFFSET_IDX]], 0
25+
; CHECK-NEXT: [[TMP6:%.*]] = add i32 [[OFFSET_IDX]], 8
26+
; CHECK-NEXT: [[TMP16:%.*]] = add i32 [[OFFSET_IDX]], 16
27+
; CHECK-NEXT: [[TMP17:%.*]] = add i32 [[OFFSET_IDX]], 24
28+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP11]]
29+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP6]]
30+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP16]]
31+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP17]]
2832
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP5]], align 1
2933
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[DST_0]], i64 [[TMP1]]
3034
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[DST_0]], i64 [[TMP2]]
@@ -36,7 +40,6 @@ define void @wide_ptr_induction_index_width_smaller_than_iv_width(ptr noalias %s
3640
; CHECK-NEXT: store ptr [[TMP13]], ptr [[TMP9]], align 8
3741
; CHECK-NEXT: store ptr [[TMP14]], ptr [[TMP10]], align 8
3842
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
39-
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i32 32
4043
; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
4144
; CHECK-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
4245
; CHECK: [[MIDDLE_BLOCK]]:

0 commit comments

Comments
 (0)