Skip to content

Commit be373bf

Browse files
fhahnfhahn
committed
!fixup simplify derived IV
1 parent 4fd2d0e commit be373bf

File tree

9 files changed

+65
-38
lines changed

9 files changed

+65
-38
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -9720,12 +9720,13 @@ void VPDerivedIVRecipe::execute(VPTransformState &State) {
97209720
State.Builder.setFastMathFlags(FPBinOp->getFastMathFlags());
97219721

97229722
Value *Step = State.get(getStepValue(), VPLane(0));
9723-
Value *CanonicalIV = State.get(getOperand(1), VPLane(0));
9723+
Value *Index = State.get(getOperand(1), VPLane(0));
97249724
Value *DerivedIV = emitTransformedIndex(
9725-
State.Builder, CanonicalIV, getStartValue()->getLiveInIRValue(), Step,
9725+
State.Builder, Index, getStartValue()->getLiveInIRValue(), Step,
97269726
Kind, cast_if_present<BinaryOperator>(FPBinOp));
97279727
DerivedIV->setName(Name);
9728-
9728+
// Index may only bet set to constant 0 in prepareToExecute.
9729+
assert((DerivedIV != Index || cast<ConstantInt>(Index)->isNullValue()) && "IV didn't need transforming?");
97299730
State.set(this, DerivedIV, VPLane(0));
97309731
}
97319732

‎llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h

Lines changed: 15 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -78,6 +78,8 @@ template <unsigned BitWidth = 0> struct specific_intval {
7878
if (!VPV->isLiveIn())
7979
return false;
8080
Value *V = VPV->getLiveInIRValue();
81+
if (!V)
82+
return false;
8183
const auto *CI = dyn_cast<ConstantInt>(V);
8284
if (!CI && V->getType()->isVectorTy())
8385
if (const auto *C = dyn_cast<Constant>(V))
@@ -136,7 +138,8 @@ struct MatchRecipeAndOpcode<Opcode, RecipeTy> {
136138
// Check for recipes that do not have opcodes.
137139
if constexpr (std::is_same<RecipeTy, VPScalarIVStepsRecipe>::value ||
138140
std::is_same<RecipeTy, VPCanonicalIVPHIRecipe>::value ||
139-
std::is_same<RecipeTy, VPWidenSelectRecipe>::value)
141+
std::is_same<RecipeTy, VPWidenSelectRecipe>::value ||
142+
std::is_same<RecipeTy, VPDerivedIVRecipe>::value)
140143
return DefR;
141144
else
142145
return DefR && DefR->getOpcode() == Opcode;
@@ -382,6 +385,17 @@ inline VPScalarIVSteps_match<Op0_t, Op1_t> m_ScalarIVSteps(const Op0_t &Op0,
382385
const Op1_t &Op1) {
383386
return VPScalarIVSteps_match<Op0_t, Op1_t>(Op0, Op1);
384387
}
388+
389+
template <typename Op0_t, typename Op1_t, typename Op2_t>
390+
using VPDerivedIV_match =
391+
Recipe_match<std::tuple<Op0_t, Op1_t, Op2_t>, 0, false, VPDerivedIVRecipe>;
392+
393+
template <typename Op0_t, typename Op1_t, typename Op2_t>
394+
inline VPDerivedIV_match<Op0_t, Op1_t, Op2_t> m_DerivedIV(const Op0_t &Op0,
395+
const Op1_t &Op1, const Op2_t &Op2) {
396+
return VPDerivedIV_match<Op0_t, Op1_t, Op2_t>({Op0, Op1, Op2});
397+
}
398+
385399
} // namespace VPlanPatternMatch
386400
} // namespace llvm
387401

‎llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1060,6 +1060,9 @@ static void simplifyRecipe(VPRecipeBase &R, VPTypeAnalysis &TypeInfo) {
10601060

10611061
if (match(&R, m_c_Mul(m_VPValue(A), m_SpecificInt(1))))
10621062
return R.getVPSingleValue()->replaceAllUsesWith(A);
1063+
1064+
if ((match(&R, m_DerivedIV(m_SpecificInt(0), m_VPValue(A), m_SpecificInt(1))) || match(&R, m_DerivedIV(m_SpecificInt(0), m_SpecificInt(0), m_VPValue())) ) && TypeInfo.inferScalarType(R.getOperand(1)) == TypeInfo.inferScalarType(R.getVPSingleValue()))
1065+
return R.getVPSingleValue()->replaceAllUsesWith(R.getOperand(1));
10631066
}
10641067

10651068
/// Move loop-invariant recipes out of the vector loop region in \p Plan.

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

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -233,8 +233,8 @@ define void @test_widen_induction_variable_start(ptr %A, i64 %N, i64 %start) {
233233
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 2
234234
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
235235
; CHECK: vec.epilog.ph:
236-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
237236
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
237+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
238238
; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP0]], 2
239239
; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF2]]
240240
; CHECK-NEXT: [[IND_END4:%.*]] = add i64 [[START]], [[N_VEC3]]
@@ -409,8 +409,8 @@ define void @test_widen_extended_induction(ptr %dst) {
409409
; CHECK: vec.epilog.iter.check:
410410
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
411411
; CHECK: vec.epilog.ph:
412-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ 16, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
413412
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i32 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
413+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ 16, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
414414
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x i8> poison, i8 [[BC_RESUME_VAL]], i64 0
415415
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x i8> [[DOTSPLATINSERT]], <2 x i8> poison, <2 x i32> zeroinitializer
416416
; CHECK-NEXT: [[INDUCTION:%.*]] = add <2 x i8> [[DOTSPLAT]], <i8 0, i8 1>

‎llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -60,8 +60,8 @@ define void @test_pr59459(i64 %iv.start, ptr %arr) {
6060
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4
6161
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
6262
; CHECK: vec.epilog.ph:
63-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[IV_START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
6463
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
64+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[IV_START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
6565
; CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 [[TMP3]], 4
6666
; CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF3]]
6767
; CHECK-NEXT: [[IND_END5:%.*]] = add i64 [[IV_START]], [[N_VEC4]]
@@ -171,11 +171,11 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
171171
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i16, ptr [[K:%.*]], i64 [[TMP3]]
172172
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 0
173173
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 16
174-
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 32
174+
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 32
175175
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 48
176176
; CHECK-NEXT: store <16 x i16> [[TMP4]], ptr [[TMP9]], align 2
177177
; CHECK-NEXT: store <16 x i16> [[TMP5]], ptr [[TMP10]], align 2
178-
; CHECK-NEXT: store <16 x i16> [[TMP6]], ptr [[TMP11]], align 2
178+
; CHECK-NEXT: store <16 x i16> [[TMP6]], ptr [[TMP21]], align 2
179179
; CHECK-NEXT: store <16 x i16> [[TMP7]], ptr [[TMP12]], align 2
180180
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
181181
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <16 x i16> [[STEP_ADD_3]], [[TMP1]]
@@ -191,8 +191,8 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
191191
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
192192
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
193193
; CHECK: vec.epilog.ph:
194-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
195194
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
195+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
196196
; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[L]], 8
197197
; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[L]], [[N_MOD_VF4]]
198198
; CHECK-NEXT: [[DOTCAST7:%.*]] = trunc i64 [[N_VEC5]] to i16

‎llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll

Lines changed: 7 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -639,6 +639,8 @@ define void @test_gather_not_profitable_pr48429(i32 %d, ptr readonly noalias %pt
639639
; AVX512: vector.ph:
640640
; AVX512-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 16
641641
; AVX512-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
642+
; AVX512-NEXT: [[TMP36:%.*]] = mul i64 [[N_VEC]], 4
643+
; AVX512-NEXT: [[TMP37:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP36]]
642644
; AVX512-NEXT: [[TMP13:%.*]] = mul i64 [[N_VEC]], 64
643645
; AVX512-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP13]]
644646
; AVX512-NEXT: br label [[VECTOR_BODY:%.*]]
@@ -673,8 +675,8 @@ define void @test_gather_not_profitable_pr48429(i32 %d, ptr readonly noalias %pt
673675
; AVX512-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
674676
; AVX512-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
675677
; AVX512: vec.epilog.ph:
676-
; AVX512-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
677678
; AVX512-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
679+
; AVX512-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
678680
; AVX512-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[TMP3]], 8
679681
; AVX512-NEXT: [[N_VEC10:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF9]]
680682
; AVX512-NEXT: [[TMP24:%.*]] = mul i64 [[N_VEC10]], 4
@@ -691,12 +693,12 @@ define void @test_gather_not_profitable_pr48429(i32 %d, ptr readonly noalias %pt
691693
; AVX512-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP27]]
692694
; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, ptr [[TMP28]], i64 [[IDXPROM]]
693695
; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds float, ptr [[TMP29]], i32 0
694-
; AVX512-NEXT: [[WIDE_LOAD23:%.*]] = load <8 x float>, ptr [[TMP30]], align 4, !alias.scope [[META17:![0-9]+]]
695-
; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD23]], <8 x ptr> [[TMP26]], i32 4, <8 x i1> splat (i1 true)), !alias.scope [[META20:![0-9]+]], !noalias [[META22:![0-9]+]]
696+
; AVX512-NEXT: [[WIDE_LOAD17:%.*]] = load <8 x float>, ptr [[TMP30]], align 4, !alias.scope [[META17:![0-9]+]]
697+
; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD17]], <8 x ptr> [[TMP26]], i32 4, <8 x i1> splat (i1 true)), !alias.scope [[META20:![0-9]+]], !noalias [[META22:![0-9]+]]
696698
; AVX512-NEXT: [[TMP31:%.*]] = getelementptr float, ptr [[TMP28]], i32 0
697-
; AVX512-NEXT: [[WIDE_LOAD24:%.*]] = load <8 x float>, ptr [[TMP31]], align 4, !alias.scope [[META24:![0-9]+]]
699+
; AVX512-NEXT: [[WIDE_LOAD18:%.*]] = load <8 x float>, ptr [[TMP31]], align 4, !alias.scope [[META24:![0-9]+]]
698700
; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, <8 x ptr> [[TMP26]], i64 1
699-
; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD24]], <8 x ptr> [[TMP32]], i32 4, <8 x i1> splat (i1 true)), !alias.scope [[META20]], !noalias [[META22]]
701+
; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD18]], <8 x ptr> [[TMP32]], i32 4, <8 x i1> splat (i1 true)), !alias.scope [[META20]], !noalias [[META22]]
700702
; AVX512-NEXT: [[INDEX_NEXT24]] = add nuw i64 [[INDEX18]], 8
701703
; AVX512-NEXT: [[PTR_IND20]] = getelementptr i8, ptr [[POINTER_PHI19]], i64 512
702704
; AVX512-NEXT: [[TMP33:%.*]] = icmp eq i64 [[INDEX_NEXT24]], [[N_VEC10]]

‎llvm/test/Transforms/LoopVectorize/X86/scatter_crash.ll

Lines changed: 18 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -72,9 +72,9 @@ define void @_Z3fn1v() #0 {
7272
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
7373
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
7474
; CHECK: vec.epilog.ph:
75+
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
7576
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 8, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
7677
; CHECK-NEXT: [[BC_RESUME_VAL5:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
77-
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
7878
; CHECK-NEXT: [[N_MOD_VF6:%.*]] = urem i64 [[TMP6]], 8
7979
; CHECK-NEXT: [[N_VEC7:%.*]] = sub i64 [[TMP6]], [[N_MOD_VF6]]
8080
; CHECK-NEXT: [[TMP17:%.*]] = mul i64 [[N_VEC7]], 2
@@ -117,10 +117,10 @@ define void @_Z3fn1v() #0 {
117117
; CHECK-NEXT: [[TMP27:%.*]] = lshr i64 [[TMP26]], 1
118118
; CHECK-NEXT: [[TMP28:%.*]] = add nuw i64 [[TMP27]], 1
119119
; CHECK-NEXT: [[MIN_ITERS_CHECK25:%.*]] = icmp ult i64 [[TMP28]], 8
120-
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK25]], label [[VEC_EPILOG_SCALAR_PH46:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK29:%.*]]
120+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK25]], label [[VEC_EPILOG_SCALAR_PH46:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK25:%.*]]
121121
; CHECK: vector.main.loop.iter.check25:
122122
; CHECK-NEXT: [[MIN_ITERS_CHECK28:%.*]] = icmp ult i64 [[TMP28]], 16
123-
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK28]], label [[VEC_EPILOG_PH47:%.*]], label [[VECTOR_PH30:%.*]]
123+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK28]], label [[VEC_EPILOG_PH42:%.*]], label [[VECTOR_PH30:%.*]]
124124
; CHECK: vector.ph26:
125125
; CHECK-NEXT: [[N_MOD_VF31:%.*]] = urem i64 [[TMP28]], 16
126126
; CHECK-NEXT: [[N_VEC32:%.*]] = sub i64 [[TMP28]], [[N_MOD_VF31]]
@@ -156,16 +156,18 @@ define void @_Z3fn1v() #0 {
156156
; CHECK-NEXT: br i1 [[TMP41]], label [[MIDDLE_BLOCK20:%.*]], label [[VECTOR_BODY29]], !llvm.loop [[LOOP4:![0-9]+]]
157157
; CHECK: middle.block20:
158158
; CHECK-NEXT: [[CMP_N40:%.*]] = icmp eq i64 [[TMP28]], [[N_VEC32]]
159-
; CHECK-NEXT: br i1 [[CMP_N40]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK48:%.*]]
159+
; CHECK-NEXT: br i1 [[CMP_N40]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK43:%.*]]
160+
; CHECK: vec.epilog.iter.check43:
160161
; CHECK-NEXT: [[IND_END58:%.*]] = mul i64 [[N_VEC32]], 2
161162
; CHECK-NEXT: [[TMP42:%.*]] = mul i64 [[N_VEC32]], 2
162163
; CHECK-NEXT: [[IND_END55:%.*]] = add i64 8, [[TMP42]]
163164
; CHECK-NEXT: [[N_VEC_REMAINING49:%.*]] = sub i64 [[TMP28]], [[N_VEC32]]
164165
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK50:%.*]] = icmp ult i64 [[N_VEC_REMAINING49]], 8
165-
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK50]], label [[VEC_EPILOG_SCALAR_PH46]], label [[VEC_EPILOG_PH47]]
166-
; CHECK-NEXT: [[BC_RESUME_VAL42:%.*]] = phi i64 [ [[IND_END41]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 8, [[VECTOR_MAIN_LOOP_ITER_CHECK29]] ]
167-
; CHECK-NEXT: [[BC_RESUME_VAL44:%.*]] = phi i64 [ [[IND_END43]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK29]] ]
168-
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL51:%.*]] = phi i64 [ [[N_VEC32]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK29]] ]
166+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK50]], label [[VEC_EPILOG_SCALAR_PH46]], label [[VEC_EPILOG_PH42]]
167+
; CHECK: vec.epilog.ph42:
168+
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL51:%.*]] = phi i64 [ [[N_VEC32]], [[VEC_EPILOG_ITER_CHECK43]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
169+
; CHECK-NEXT: [[BC_RESUME_VAL42:%.*]] = phi i64 [ [[IND_END41]], [[VEC_EPILOG_ITER_CHECK43]] ], [ 8, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
170+
; CHECK-NEXT: [[BC_RESUME_VAL44:%.*]] = phi i64 [ [[IND_END43]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
169171
; CHECK-NEXT: [[N_MOD_VF52:%.*]] = urem i64 [[TMP28]], 8
170172
; CHECK-NEXT: [[N_VEC53:%.*]] = sub i64 [[TMP28]], [[N_MOD_VF52]]
171173
; CHECK-NEXT: [[TMP43:%.*]] = mul i64 [[N_VEC53]], 2
@@ -181,9 +183,9 @@ define void @_Z3fn1v() #0 {
181183
; CHECK-NEXT: [[BROADCAST_SPLAT73:%.*]] = shufflevector <8 x i1> [[BROADCAST_SPLATINSERT72]], <8 x i1> poison, <8 x i32> zeroinitializer
182184
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY50:%.*]]
183185
; CHECK: vec.epilog.vector.body50:
184-
; CHECK-NEXT: [[INDEX61:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL51]], [[VEC_EPILOG_PH47]] ], [ [[INDEX_NEXT74:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
185-
; CHECK-NEXT: [[VEC_IND65:%.*]] = phi <8 x i64> [ [[INDUCTION64]], [[VEC_EPILOG_PH47]] ], [ [[VEC_IND_NEXT66:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
186-
; CHECK-NEXT: [[VEC_IND70:%.*]] = phi <8 x i64> [ [[INDUCTION69]], [[VEC_EPILOG_PH47]] ], [ [[VEC_IND_NEXT71:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
186+
; CHECK-NEXT: [[INDEX61:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL51]], [[VEC_EPILOG_PH42]] ], [ [[INDEX_NEXT74:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
187+
; CHECK-NEXT: [[VEC_IND65:%.*]] = phi <8 x i64> [ [[INDUCTION64]], [[VEC_EPILOG_PH42]] ], [ [[VEC_IND_NEXT66:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
188+
; CHECK-NEXT: [[VEC_IND70:%.*]] = phi <8 x i64> [ [[INDUCTION69]], [[VEC_EPILOG_PH42]] ], [ [[VEC_IND_NEXT71:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
187189
; CHECK-NEXT: [[TMP44:%.*]] = sub nsw <8 x i64> splat (i64 8), [[VEC_IND65]]
188190
; CHECK-NEXT: [[TMP45:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, <8 x i64> [[VEC_IND65]]
189191
; CHECK-NEXT: [[TMP46:%.*]] = add nsw <8 x i64> [[TMP44]], [[VEC_IND70]]
@@ -203,16 +205,16 @@ define void @_Z3fn1v() #0 {
203205
; CHECK-NEXT: [[VEC_IND_NEXT66]] = add <8 x i64> [[VEC_IND65]], splat (i64 16)
204206
; CHECK-NEXT: [[VEC_IND_NEXT71]] = add <8 x i64> [[VEC_IND70]], splat (i64 16)
205207
; CHECK-NEXT: [[TMP55:%.*]] = icmp eq i64 [[INDEX_NEXT74]], [[N_VEC53]]
206-
; CHECK-NEXT: br i1 [[TMP55]], label [[VEC_EPILOG_MIDDLE_BLOCK45:%.*]], label [[VEC_EPILOG_VECTOR_BODY58]], !llvm.loop [[LOOP5:![0-9]+]]
207-
; CHECK: vec.epilog.middle.block46:
208+
; CHECK-NEXT: br i1 [[TMP55]], label [[VEC_EPILOG_MIDDLE_BLOCK45:%.*]], label [[VEC_EPILOG_VECTOR_BODY50]], !llvm.loop [[LOOP5:![0-9]+]]
209+
; CHECK: vec.epilog.middle.block40:
208210
; CHECK-NEXT: [[CMP_N75:%.*]] = icmp eq i64 [[TMP28]], [[N_VEC53]]
209-
; CHECK-NEXT: br i1 [[CMP_N75]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH46]]
210-
; CHECK: vec.epilog.scalar.ph47:
211+
; CHECK-NEXT: br i1 [[CMP_N75]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH41]]
212+
; CHECK: vec.epilog.scalar.ph41:
211213
; CHECK-NEXT: [[BC_RESUME_VAL56:%.*]] = phi i64 [ [[IND_END54]], [[VEC_EPILOG_MIDDLE_BLOCK45]] ], [ [[IND_END55]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 8, [[ITER_CHECK27]] ]
212214
; CHECK-NEXT: [[BC_RESUME_VAL59:%.*]] = phi i64 [ [[IND_END57]], [[VEC_EPILOG_MIDDLE_BLOCK45]] ], [ [[IND_END58]], [[VEC_EPILOG_ITER_CHECK48]] ], [ 0, [[ITER_CHECK27]] ]
213215
; CHECK-NEXT: br label [[FOR_BODY_US:%.*]]
214216
; CHECK: for.body.us:
215-
; CHECK-NEXT: [[INDVARS_IV78:%.*]] = phi i64 [ [[INDVARS_IV_NEXT79:%.*]], [[FOR_COND_CLEANUP4_US_LCSSA_US_US:%.*]] ], [ [[BC_RESUME_VAL56]], [[VEC_EPILOG_SCALAR_PH46]] ]
217+
; CHECK-NEXT: [[INDVARS_IV78:%.*]] = phi i64 [ [[INDVARS_IV_NEXT79:%.*]], [[FOR_COND_CLEANUP4_US_LCSSA_US_US:%.*]] ], [ [[BC_RESUME_VAL56]], [[VEC_EPILOG_SCALAR_PH41]] ]
216218
; CHECK-NEXT: [[INDVARS_IV70:%.*]] = phi i64 [ [[INDVARS_IV_NEXT71:%.*]], [[FOR_COND_CLEANUP4_US_LCSSA_US_US]] ], [ [[BC_RESUME_VAL59]], [[VEC_EPILOG_SCALAR_PH46]] ]
217219
; CHECK-NEXT: [[TMP56:%.*]] = sub nsw i64 8, [[INDVARS_IV78]]
218220
; CHECK-NEXT: [[ADD_PTR_US:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, i64 [[INDVARS_IV78]]

‎llvm/test/Transforms/LoopVectorize/vplan-iv-transforms.ll

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -35,10 +35,11 @@ define void @iv_no_binary_op_in_descriptor(i1 %c, ptr %dst) {
3535
; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
3636
; CHECK-EMPTY:
3737
; CHECK-NEXT: scalar.ph:
38+
; CHECK-NEXT: EMIT vp<[[RESUME:%.+]]> = resume-phi vp<[[VEC_TC]]>, ir<0>
3839
; CHECK-NEXT: Successor(s): ir-bb<loop.header>
3940
; CHECK-EMPTY:
4041
; CHECK-NEXT: ir-bb<loop.header>:
41-
; CHECK-NEXT: IR %iv = phi i64 [ 0, %entry ], [ %iv.next.p, %loop.latch ]
42+
; CHECK-NEXT: IR %iv = phi i64 [ 0, %entry ], [ %iv.next.p, %loop.latch ] (extra operand: vp<[[RESUME]]> from scalar.ph)
4243
; CHECK: IR %iv.next = add i64 %iv, 1
4344
; CHECK-NEXT: No successors
4445
; CHECK-EMPTY:

0 commit comments

Comments
 (0)