[VPlan] Directly unroll VectorPointerRecipe#168886
Merged
Conversation
artagnon
requested review from
fhahn,
kmclaughlin-arm,
lukel97 and
paulwalker-arm
Member
|
@llvm/pr-subscribers-backend-powerpc @llvm/pr-subscribers-backend-risc-v Author: Ramkumar Ramachandra (artagnon) ChangesSimilar to how the runtime VF is modeled as an operand in VectorEndPointerRecipe, model it as an operand in VectorPointerRecipe as well. Patch is 320.47 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/168886.diff 44 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 8013c47f53db0..feb699f738f23 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7608,10 +7608,10 @@ VPWidenMemoryRecipe *VPRecipeBuilder::tryToWidenMemory(VPInstruction *VPI,
Ptr, &Plan.getVF(), getLoadStoreType(I),
/*Stride*/ -1, Flags, VPI->getDebugLoc());
} else {
- VectorPtr = new VPVectorPointerRecipe(Ptr, getLoadStoreType(I),
- GEP ? GEP->getNoWrapFlags()
- : GEPNoWrapFlags::none(),
- VPI->getDebugLoc());
+ VectorPtr = new VPVectorPointerRecipe(
+ Ptr, &Plan.getVF(), getLoadStoreType(I),
+ GEP ? GEP->getNoWrapFlags() : GEPNoWrapFlags::none(),
+ VPI->getDebugLoc());
}
Builder.insert(VectorPtr);
Ptr = VectorPtr;
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 13582f8bd2d62..7f817a3f8a218 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1974,18 +1974,20 @@ class VPVectorEndPointerRecipe : public VPRecipeWithIRFlags,
/// A recipe to compute the pointers for widened memory accesses of IndexTy.
class VPVectorPointerRecipe : public VPRecipeWithIRFlags,
- public VPUnrollPartAccessor<1> {
+ public VPUnrollPartAccessor<2> {
Type *SourceElementTy;
public:
- VPVectorPointerRecipe(VPValue *Ptr, Type *SourceElementTy,
+ VPVectorPointerRecipe(VPValue *Ptr, VPValue *VF, Type *SourceElementTy,
GEPNoWrapFlags GEPFlags, DebugLoc DL)
- : VPRecipeWithIRFlags(VPDef::VPVectorPointerSC, ArrayRef<VPValue *>(Ptr),
- GEPFlags, DL),
+ : VPRecipeWithIRFlags(VPDef::VPVectorPointerSC, {Ptr, VF}, GEPFlags, DL),
SourceElementTy(SourceElementTy) {}
VP_CLASSOF_IMPL(VPDef::VPVectorPointerSC)
+ VPValue *getVFValue() { return getOperand(1); }
+ const VPValue *getVFValue() const { return getOperand(1); }
+
void execute(VPTransformState &State) override;
Type *getSourceElementType() const { return SourceElementTy; }
@@ -2005,8 +2007,9 @@ class VPVectorPointerRecipe : public VPRecipeWithIRFlags,
}
VPVectorPointerRecipe *clone() override {
- return new VPVectorPointerRecipe(getOperand(0), SourceElementTy,
- getGEPNoWrapFlags(), getDebugLoc());
+ return new VPVectorPointerRecipe(getOperand(0), getVFValue(),
+ SourceElementTy, getGEPNoWrapFlags(),
+ getDebugLoc());
}
/// Return true if this VPVectorPointerRecipe corresponds to part 0. Note that
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 7c9302860a3b5..6f0e5a2e93206 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2651,7 +2651,12 @@ void VPVectorPointerRecipe::execute(VPTransformState &State) {
/*IsUnitStride*/ true, CurrentPart, Builder);
Value *Ptr = State.get(getOperand(0), VPLane(0));
- Value *Increment = createStepForVF(Builder, IndexTy, State.VF, CurrentPart);
+ Value *RuntimeVF = State.get(getVFValue(), VPLane(0));
+ RuntimeVF = Builder.CreateZExtOrTrunc(RuntimeVF, IndexTy);
+ Value *Increment =
+ CurrentPart == 1 ? RuntimeVF
+ : Builder.CreateNUWMul(
+ RuntimeVF, ConstantInt::get(IndexTy, CurrentPart));
Value *ResultPtr = Builder.CreateGEP(getSourceElementType(), Ptr, Increment,
"", getGEPNoWrapFlags());
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 25557f1d5d651..8f3d67d9f2313 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2710,10 +2710,11 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
VPRegionBlock *LoopRegion = Plan.getVectorLoopRegion();
VPBasicBlock *Header = LoopRegion->getEntryBasicBlock();
- assert(all_of(Plan.getVF().users(),
- IsaPred<VPVectorEndPointerRecipe, VPScalarIVStepsRecipe,
- VPWidenIntOrFpInductionRecipe>) &&
- "User of VF that we can't transform to EVL.");
+ assert(
+ all_of(Plan.getVF().users(),
+ IsaPred<VPVectorPointerRecipe, VPVectorEndPointerRecipe,
+ VPScalarIVStepsRecipe, VPWidenIntOrFpInductionRecipe>) &&
+ "User of VF that we can't transform to EVL.");
Plan.getVF().replaceUsesWithIf(&EVL, [](VPUser &U, unsigned Idx) {
return isa<VPWidenIntOrFpInductionRecipe, VPScalarIVStepsRecipe>(U);
});
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/divs-with-scalable-vfs.ll b/llvm/test/Transforms/LoopVectorize/AArch64/divs-with-scalable-vfs.ll
index 72e813b62025f..3227862c7350f 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/divs-with-scalable-vfs.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/divs-with-scalable-vfs.ll
@@ -21,7 +21,8 @@ define void @sdiv_feeding_gep(ptr %dst, i32 %x, i64 %M, i64 %conv6, i64 %N) {
; CHECK-NEXT: br i1 [[TMP7]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP9:%.*]] = mul nuw i64 [[TMP8]], 4
+; CHECK-NEXT: [[TMP11:%.*]] = mul nuw i64 [[TMP8]], 2
+; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP11]], 2
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP9]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT: [[TMP18:%.*]] = sdiv i64 [[M]], [[CONV6]]
@@ -36,9 +37,7 @@ define void @sdiv_feeding_gep(ptr %dst, i32 %x, i64 %M, i64 %conv6, i64 %N) {
; CHECK-NEXT: [[TMP30:%.*]] = add i32 [[TMP28]], [[TMP26]]
; CHECK-NEXT: [[TMP32:%.*]] = sext i32 [[TMP30]] to i64
; CHECK-NEXT: [[TMP34:%.*]] = getelementptr double, ptr [[DST]], i64 [[TMP32]]
-; CHECK-NEXT: [[TMP37:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP38:%.*]] = shl nuw i64 [[TMP37]], 1
-; CHECK-NEXT: [[TMP39:%.*]] = getelementptr double, ptr [[TMP34]], i64 [[TMP38]]
+; CHECK-NEXT: [[TMP39:%.*]] = getelementptr double, ptr [[TMP34]], i64 [[TMP11]]
; CHECK-NEXT: store <vscale x 2 x double> zeroinitializer, ptr [[TMP34]], align 8
; CHECK-NEXT: store <vscale x 2 x double> zeroinitializer, ptr [[TMP39]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll b/llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll
index 4b097ba2422e4..34bba684b3bb1 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll
@@ -30,7 +30,8 @@ define void @iv_casts(ptr %dst, ptr %src, i32 %x, i64 %N) #0 {
; DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK3]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
; DEFAULT: [[VECTOR_PH]]:
; DEFAULT-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
-; DEFAULT-NEXT: [[TMP10:%.*]] = mul nuw i64 [[TMP9]], 16
+; DEFAULT-NEXT: [[TMP13:%.*]] = mul nuw i64 [[TMP9]], 8
+; DEFAULT-NEXT: [[TMP10:%.*]] = mul i64 [[TMP13]], 2
; DEFAULT-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP10]]
; DEFAULT-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
; DEFAULT-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[X]], i64 0
@@ -40,9 +41,7 @@ define void @iv_casts(ptr %dst, ptr %src, i32 %x, i64 %N) #0 {
; DEFAULT: [[VECTOR_BODY]]:
; DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; DEFAULT-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[INDEX]]
-; DEFAULT-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
-; DEFAULT-NEXT: [[TMP14:%.*]] = shl nuw i64 [[TMP13]], 3
-; DEFAULT-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[TMP12]], i64 [[TMP14]]
+; DEFAULT-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[TMP12]], i64 [[TMP13]]
; DEFAULT-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP12]], align 1
; DEFAULT-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP15]], align 1
; DEFAULT-NEXT: [[TMP16:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i16>
@@ -56,9 +55,7 @@ define void @iv_casts(ptr %dst, ptr %src, i32 %x, i64 %N) #0 {
; DEFAULT-NEXT: [[TMP24:%.*]] = trunc <vscale x 8 x i16> [[TMP22]] to <vscale x 8 x i8>
; DEFAULT-NEXT: [[TMP25:%.*]] = trunc <vscale x 8 x i16> [[TMP23]] to <vscale x 8 x i8>
; DEFAULT-NEXT: [[TMP26:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX]]
-; DEFAULT-NEXT: [[TMP27:%.*]] = call i64 @llvm.vscale.i64()
-; DEFAULT-NEXT: [[TMP28:%.*]] = shl nuw i64 [[TMP27]], 3
-; DEFAULT-NEXT: [[TMP29:%.*]] = getelementptr i8, ptr [[TMP26]], i64 [[TMP28]]
+; DEFAULT-NEXT: [[TMP29:%.*]] = getelementptr i8, ptr [[TMP26]], i64 [[TMP13]]
; DEFAULT-NEXT: store <vscale x 8 x i8> [[TMP24]], ptr [[TMP26]], align 1
; DEFAULT-NEXT: store <vscale x 8 x i8> [[TMP25]], ptr [[TMP29]], align 1
; DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP10]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleaving-reduction.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleaving-reduction.ll
index aa94763b44a30..1f83ce55e153a 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/interleaving-reduction.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleaving-reduction.ll
@@ -144,7 +144,8 @@ define i32 @interleave_integer_reduction(ptr %src, i64 %N) {
; INTERLEAVE-4-VLA-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; INTERLEAVE-4-VLA: vector.ph:
; INTERLEAVE-4-VLA-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; INTERLEAVE-4-VLA-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16
+; INTERLEAVE-4-VLA-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP2]], 4
+; INTERLEAVE-4-VLA-NEXT: [[TMP3:%.*]] = mul i64 [[TMP5]], 4
; INTERLEAVE-4-VLA-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
; INTERLEAVE-4-VLA-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; INTERLEAVE-4-VLA-NEXT: br label [[VECTOR_BODY:%.*]]
@@ -155,14 +156,10 @@ define i32 @interleave_integer_reduction(ptr %src, i64 %N) {
; INTERLEAVE-4-VLA-NEXT: [[VEC_PHI3:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ]
; INTERLEAVE-4-VLA-NEXT: [[VEC_PHI4:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
; INTERLEAVE-4-VLA-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i64 [[INDEX]]
-; INTERLEAVE-4-VLA-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
-; INTERLEAVE-4-VLA-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 2
-; INTERLEAVE-4-VLA-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[TMP6]]
-; INTERLEAVE-4-VLA-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; INTERLEAVE-4-VLA-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 3
+; INTERLEAVE-4-VLA-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[TMP5]]
+; INTERLEAVE-4-VLA-NEXT: [[TMP9:%.*]] = mul nuw i64 [[TMP5]], 2
; INTERLEAVE-4-VLA-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[TMP9]]
-; INTERLEAVE-4-VLA-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()
-; INTERLEAVE-4-VLA-NEXT: [[TMP12:%.*]] = mul nuw i64 [[TMP11]], 12
+; INTERLEAVE-4-VLA-NEXT: [[TMP12:%.*]] = mul nuw i64 [[TMP5]], 3
; INTERLEAVE-4-VLA-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[TMP12]]
; INTERLEAVE-4-VLA-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP4]], align 1
; INTERLEAVE-4-VLA-NEXT: [[WIDE_LOAD5:%.*]] = load <vscale x 4 x i32>, ptr [[TMP7]], align 1
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
index 31453e9509ea3..d8cc8167f00af 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
@@ -37,7 +37,7 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFNONE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[N_VEC]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]
; TFNONE-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[B]], i64 [[INDVARS_IV]]
; TFNONE-NEXT: [[LOAD:%.*]] = load i64, ptr [[GEP]], align 8
-; TFNONE-NEXT: [[CALL:%.*]] = call i64 @foo(i64 [[LOAD]]) #[[ATTR3:[0-9]+]]
+; TFNONE-NEXT: [[CALL:%.*]] = call i64 @foo(i64 [[LOAD]]) #[[ATTR4:[0-9]+]]
; TFNONE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDVARS_IV]]
; TFNONE-NEXT: store i64 [[CALL]], ptr [[ARRAYIDX]], align 8
; TFNONE-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
@@ -72,7 +72,8 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-SAME: ptr noalias [[A:%.*]], ptr readnone [[B:%.*]]) #[[ATTR0:[0-9]+]] {
; TFA_INTERLEAVE-NEXT: [[ENTRY:.*]]:
; TFA_INTERLEAVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 4
+; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = mul nuw i64 [[TMP3]], 2
+; TFA_INTERLEAVE-NEXT: [[TMP4:%.*]] = mul i64 [[TMP9]], 2
; TFA_INTERLEAVE-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 1
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP6]], i64 1025)
@@ -82,17 +83,13 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ splat (i1 true), %[[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], %[[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT4:%.*]], %[[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP7:%.*]] = getelementptr i64, ptr [[B]], i64 [[INDEX]]
-; TFA_INTERLEAVE-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 1
; TFA_INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr i64, ptr [[TMP7]], i64 [[TMP9]]
; TFA_INTERLEAVE-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 8 [[TMP7]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; TFA_INTERLEAVE-NEXT: [[WIDE_MASKED_LOAD3:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 8 [[TMP10]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]], <vscale x 2 x i64> poison)
; TFA_INTERLEAVE-NEXT: [[TMP11:%.*]] = call <vscale x 2 x i64> @foo_vector(<vscale x 2 x i64> [[WIDE_MASKED_LOAD]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: [[TMP12:%.*]] = call <vscale x 2 x i64> @foo_vector(<vscale x 2 x i64> [[WIDE_MASKED_LOAD3]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
; TFA_INTERLEAVE-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]]
-; TFA_INTERLEAVE-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP15:%.*]] = shl nuw i64 [[TMP14]], 1
-; TFA_INTERLEAVE-NEXT: [[TMP16:%.*]] = getelementptr inbounds i64, ptr [[TMP13]], i64 [[TMP15]]
+; TFA_INTERLEAVE-NEXT: [[TMP16:%.*]] = getelementptr inbounds i64, ptr [[TMP13]], i64 [[TMP9]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP11]], ptr align 8 [[TMP13]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP12]], ptr align 8 [[TMP16]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]]
@@ -160,7 +157,7 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFNONE-NEXT: [[CMP:%.*]] = icmp ugt i64 [[TMP12]], 50
; TFNONE-NEXT: br i1 [[CMP]], label %[[IF_THEN:.*]], label %[[IF_END]]
; TFNONE: [[IF_THEN]]:
-; TFNONE-NEXT: [[TMP9:%.*]] = call i64 @foo(i64 [[TMP12]]) #[[ATTR3]]
+; TFNONE-NEXT: [[TMP9:%.*]] = call i64 @foo(i64 [[TMP12]]) #[[ATTR4]]
; TFNONE-NEXT: br label %[[IF_END]]
; TFNONE: [[IF_END]]:
; TFNONE-NEXT: [[TMP14:%.*]] = phi i64 [ [[TMP9]], %[[IF_THEN]] ], [ 0, %[[FOR_BODY]] ]
@@ -201,7 +198,8 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-SAME: ptr noalias [[A:%.*]], ptr readnone [[B:%.*]]) #[[ATTR0]] {
; TFA_INTERLEAVE-NEXT: [[ENTRY:.*]]:
; TFA_INTERLEAVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 4
+; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = mul nuw i64 [[TMP3]], 2
+; TFA_INTERLEAVE-NEXT: [[TMP4:%.*]] = mul i64 [[TMP9]], 2
; TFA_INTERLEAVE-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 1
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP6]], i64 1025)
@@ -211,8 +209,6 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ splat (i1 true), %[[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], %[[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], %[[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX]]
-; TFA_INTERLEAVE-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 1
; TFA_INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[TMP7]], i64 [[TMP9]]
; TFA_INTERLEAVE-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 8 [[TMP7]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]], <vscale x 2 x i64> poison)
; TFA_INTERLEAVE-NEXT: [[WIDE_MASKED_LOAD3:%.*]] = call <vscale x 2 x i64> @llvm.masked.load.nxv2i64.p0(ptr align 8 [[TMP10]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]], <vscale x 2 x i64> poison)
@@ -225,9 +221,7 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[PREDPHI:%.*]] = select <vscale x 2 x i1> [[TMP11]], <vscale x 2 x i64> [[TMP15]], <vscale x 2 x i64> zeroinitializer
; TFA_INTERLEAVE-NEXT: [[PREDPHI4:%.*]] = select <vscale x 2 x i1> [[TMP12]], <vscale x 2 x i64> [[TMP16]], <vscale x 2 x i64> zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP17:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[INDEX]]
-; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64()
-; TFA_INTERLEAVE-NEXT: [[TMP19:%.*]] = shl nuw i64 [[TMP18]], 1
-; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[TMP17]], i64 [[TMP19]]
+; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[TMP17]], i64 [[TMP9]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI]], ptr align 8 [[TMP17]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI4]], ptr align 8 [[TMP20]], <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]]
@@ -308,10 +302,10 @@ define void @test_widen_if_then_else(ptr noalias %a, ptr readnone %b) #4 {
; TFNONE-NEXT: [[CMP:%.*]] = icmp ugt i64 [[TMP13]], 50
; TFNONE-NEXT: br i1 [[CMP]], label %[[IF_THEN:.*]], label %[[IF_ELSE:.*]]
; TFNONE: [[IF_THEN]]:
-; TFNONE-NEXT: [[TMP14:%.*]] = call i64 @foo(i64 [[TMP13]]) #[[ATTR4:[0-9]+]]
+; TFNONE-NEXT: [[TMP14:%.*]] = call i64 @foo(i64 [[TMP13]]) #[[ATTR5:[0-9]+]]
; TFNONE-NEXT: ...
[truncated]
|
🐧 Linux x64 Test Results
✅ The build succeeded and all tests passed. |
fhahn
reviewed
Nov 23, 2025
artagnon
changed the title
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
e398f4f to
3e7cda1
Compare
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
3e7cda1 to
02787c9
Compare
fhahn
reviewed
Nov 25, 2025
llvm/test/Transforms/LoopVectorize/AArch64/divs-with-scalable-vfs.ll
Outdated
Show resolved
Hide resolved
artagnon
added a commit
that referenced
this pull request
Nov 26, 2025
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see #168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
02787c9 to
6a75337
Compare
tanji-dg
pushed a commit
to tanji-dg/llvm-project
that referenced
this pull request
Nov 27, 2025
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see llvm#168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
GeneraluseAI
pushed a commit
to GeneraluseAI/llvm-project
that referenced
this pull request
Nov 27, 2025
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see llvm#168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
6a75337 to
24dd807
Compare
fhahn
reviewed
Dec 2, 2025
lukel97
reviewed
Dec 2, 2025
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll
Outdated
Show resolved
Hide resolved
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
24dd807 to
c2baf7f
Compare
fhahn
reviewed
Dec 2, 2025
augusto2112
pushed a commit
to augusto2112/llvm-project
that referenced
this pull request
Dec 3, 2025
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see llvm#168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
fhahn
reviewed
Dec 3, 2025
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
5a01ad6 to
9c0eda6
Compare
kcloudy0717
pushed a commit
to kcloudy0717/llvm-project
that referenced
this pull request
Dec 4, 2025
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see llvm#168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
fhahn
reviewed
Dec 4, 2025
llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll
Outdated
Show resolved
Hide resolved
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
3 times, most recently
from
8e143cb to
401b077
Compare
Contributor
Author
|
Gentle ping. |
fhahn
reviewed
Dec 14, 2025
fhahn
approved these changes
Dec 14, 2025
Contributor
fhahn
left a comment
fhahn
left a comment
There was a problem hiding this comment.
LGTM, with comments adjusted, thanks
fhahn
reviewed
Dec 14, 2025
artagnon
force-pushed
the
vplan-vector-pointer-vf-op
branch
from
401b077 to
fbaef19
Compare
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/88/builds/19028 Here is the relevant piece of the build log for the reference |
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Dec 19, 2025
In an effort to get rid of VPUnrollPartAccessor and directly unroll recipes, start by directly unrolling VectorPointerRecipe, allowing for VPlan-based simplifications and simplification of the corresponding execute.
valadaptive
pushed a commit
to valadaptive/llvm-project
that referenced
this pull request
Dec 24, 2025
In an effort to get rid of VPUnrollPartAccessor and directly unroll recipes, start by directly unrolling VectorPointerRecipe, allowing for VPlan-based simplifications and simplification of the corresponding execute.
This was referenced Jan 13, 2026
Xubaidu
pushed a commit
to Xubaidu/llvm-project
that referenced
this pull request
Jan 26, 2026
In preparation to strip VPUnrollPartAccessor and unroll recipes directly, strip unnecessary complication in getGEPIndexTy, as the unroll part will no longer be available in follow-ups (see llvm#168886 for instance). The patch also helps by doing a mass test update up-front. Narrowing the GEP index type conditionally does not yield any benefit, and the change is non-functional in terms of emitted assembly. While at it, avoid hard-coding address-space 0, and use the pointer operand's address space to get the GEP index type.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
In an effort to get rid of VPUnrollPartAccessor and directly unroll recipes, start by directly unrolling VectorPointerRecipe, allowing for VPlan-based simplifications and simplification of the corresponding execute.