[RISCV][TTI] Recognize CONCAT_VECTORS if a shufflevector mask is multiple insert subvector. #111459
Conversation
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
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-transforms Author: Han-Kuan Chen (HanKuanChen) Changesreference: #110457 Full diff: https://github.com/llvm/llvm-project/pull/111459.diff 5 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index a61461681f79ed..1e5321a9ace41b 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -343,6 +343,49 @@ RISCVTTIImpl::getConstantPoolLoadCost(Type *Ty, TTI::TargetCostKind CostKind) {
/*AddressSpace=*/0, CostKind);
}
+InstructionCost
+RISCVTTIImpl::isMultipleInsertSubvector(VectorType *Tp, ArrayRef<int> Mask,
+ TTI::TargetCostKind CostKind) {
+ if (!isa<FixedVectorType>(Tp))
+ return InstructionCost::getInvalid();
+ std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(Tp);
+ if (LT.second.getScalarSizeInBits() == 1)
+ return InstructionCost::getInvalid();
+ // Try to guess SubTp.
+ for (unsigned SubVecSize = 1, E = Mask.size(); SubVecSize < E;
+ SubVecSize <<= 1) {
+ if (E % SubVecSize != 0)
+ continue;
+ SmallVector<int> RepeatedPattern(createSequentialMask(0, SubVecSize, 0));
+ bool Skip = false;
+ for (unsigned I = 0; I != E; I += SubVecSize)
+ if (!Mask.slice(I, SubVecSize).equals(RepeatedPattern)) {
+ Skip = true;
+ break;
+ }
+ if (Skip)
+ continue;
+ InstructionCost Cost = 0;
+ unsigned NumSlides = Log2_32(E / SubVecSize);
+ // The cost of extraction from a subvector is 0 if the index is 0.
+ for (unsigned I = 0; I != NumSlides; ++I) {
+ unsigned InsertIndex = SubVecSize * (1 << I);
+ FixedVectorType *SubTp = FixedVectorType::get(
+ cast<FixedVectorType>(Tp)->getElementType(), InsertIndex);
+ FixedVectorType *DesTp =
+ FixedVectorType::getDoubleElementsVectorType(SubTp);
+ std::pair<InstructionCost, MVT> DesLT = getTypeLegalizationCost(DesTp);
+ // Add the cost of whole vector register move because the destination
+ // vector register group for vslideup cannot overlap the source.
+ Cost += DesLT.first * TLI->getLMULCost(DesLT.second);
+ Cost += getShuffleCost(TTI::SK_InsertSubvector, DesTp, {}, CostKind,
+ InsertIndex, SubTp);
+ }
+ return Cost;
+ }
+ return InstructionCost::getInvalid();
+}
+
static VectorType *getVRGatherIndexType(MVT DataVT, const RISCVSubtarget &ST,
LLVMContext &C) {
assert((DataVT.getScalarSizeInBits() != 8 ||
@@ -394,6 +437,10 @@ InstructionCost RISCVTTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
LT.second, CostKind);
}
}
+ if (InstructionCost Cost =
+ isMultipleInsertSubvector(Tp, Mask, CostKind);
+ Cost.isValid())
+ return Cost;
}
// vrgather + cost of generating the mask constant.
// We model this for an unknown mask with a single vrgather.
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
index 65bbd905508557..9d6317baabc658 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -55,6 +55,12 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
/// type.
InstructionCost getConstantPoolLoadCost(Type *Ty,
TTI::TargetCostKind CostKind);
+
+ /// Return the cost if a shufflevector can be consist of multiple vslideup.
+ /// Otherwise, return InstructionCost::getInvalid().
+ InstructionCost isMultipleInsertSubvector(VectorType *Tp, ArrayRef<int> Mask,
+ TTI::TargetCostKind CostKind);
+
public:
explicit RISCVTTIImpl(const RISCVTargetMachine *TM, const Function &F)
: BaseT(TM, F.getDataLayout()), ST(TM->getSubtargetImpl(F)),
diff --git a/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll b/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll
new file mode 100644
index 00000000000000..47a2af92aee950
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll
@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: 'test'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %0 = shufflevector <8 x float> poison, <8 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %1 = shufflevector <4 x i16> poison, <4 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %2 = shufflevector <4 x float> poison, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %3 = shufflevector <2 x i1> poison, <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+entry:
+ %0 = shufflevector <8 x float> poison, <8 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %1 = shufflevector <4 x i16> poison, <4 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <4 x float> poison, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ %3 = shufflevector <2 x i1> poison, <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ ret void
+}
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
index bb806be15c71ca..23a9a654c96f9e 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
@@ -8,7 +8,7 @@
; YAML-NEXT: Function: test
; YAML-NEXT: Args:
; YAML-NEXT: - String: 'Stores SLP vectorized with cost '
-; YAML-NEXT: - Cost: '2'
+; YAML-NEXT: - Cost: '0'
; YAML-NEXT: - String: ' and with tree size '
; YAML-NEXT: - TreeSize: '7'
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
index 995cd7cfbc880b..a0cb52a853b7e6 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
@@ -8,7 +8,7 @@
; YAML: Function: test1
; YAML: Args:
; YAML: - String: 'Stores SLP vectorized with cost '
-; YAML: - Cost: '6'
+; YAML: - Cost: '4'
; YAML: - String: ' and with tree size '
; YAML: - TreeSize: '5'
@@ -47,7 +47,7 @@ declare <4 x float> @llvm.fmuladd.v4f32(<4 x float>, <4 x float>, <4 x float>)
; YAML: Function: test2
; YAML: Args:
; YAML: - String: 'Stores SLP vectorized with cost '
-; YAML: - Cost: '16'
+; YAML: - Cost: '12'
; YAML: - String: ' and with tree size '
; YAML: - TreeSize: '5'
|
|
@llvm/pr-subscribers-llvm-analysis Author: Han-Kuan Chen (HanKuanChen) Changesreference: #110457 Full diff: https://github.com/llvm/llvm-project/pull/111459.diff 5 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index a61461681f79ed..1e5321a9ace41b 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -343,6 +343,49 @@ RISCVTTIImpl::getConstantPoolLoadCost(Type *Ty, TTI::TargetCostKind CostKind) {
/*AddressSpace=*/0, CostKind);
}
+InstructionCost
+RISCVTTIImpl::isMultipleInsertSubvector(VectorType *Tp, ArrayRef<int> Mask,
+ TTI::TargetCostKind CostKind) {
+ if (!isa<FixedVectorType>(Tp))
+ return InstructionCost::getInvalid();
+ std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(Tp);
+ if (LT.second.getScalarSizeInBits() == 1)
+ return InstructionCost::getInvalid();
+ // Try to guess SubTp.
+ for (unsigned SubVecSize = 1, E = Mask.size(); SubVecSize < E;
+ SubVecSize <<= 1) {
+ if (E % SubVecSize != 0)
+ continue;
+ SmallVector<int> RepeatedPattern(createSequentialMask(0, SubVecSize, 0));
+ bool Skip = false;
+ for (unsigned I = 0; I != E; I += SubVecSize)
+ if (!Mask.slice(I, SubVecSize).equals(RepeatedPattern)) {
+ Skip = true;
+ break;
+ }
+ if (Skip)
+ continue;
+ InstructionCost Cost = 0;
+ unsigned NumSlides = Log2_32(E / SubVecSize);
+ // The cost of extraction from a subvector is 0 if the index is 0.
+ for (unsigned I = 0; I != NumSlides; ++I) {
+ unsigned InsertIndex = SubVecSize * (1 << I);
+ FixedVectorType *SubTp = FixedVectorType::get(
+ cast<FixedVectorType>(Tp)->getElementType(), InsertIndex);
+ FixedVectorType *DesTp =
+ FixedVectorType::getDoubleElementsVectorType(SubTp);
+ std::pair<InstructionCost, MVT> DesLT = getTypeLegalizationCost(DesTp);
+ // Add the cost of whole vector register move because the destination
+ // vector register group for vslideup cannot overlap the source.
+ Cost += DesLT.first * TLI->getLMULCost(DesLT.second);
+ Cost += getShuffleCost(TTI::SK_InsertSubvector, DesTp, {}, CostKind,
+ InsertIndex, SubTp);
+ }
+ return Cost;
+ }
+ return InstructionCost::getInvalid();
+}
+
static VectorType *getVRGatherIndexType(MVT DataVT, const RISCVSubtarget &ST,
LLVMContext &C) {
assert((DataVT.getScalarSizeInBits() != 8 ||
@@ -394,6 +437,10 @@ InstructionCost RISCVTTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
LT.second, CostKind);
}
}
+ if (InstructionCost Cost =
+ isMultipleInsertSubvector(Tp, Mask, CostKind);
+ Cost.isValid())
+ return Cost;
}
// vrgather + cost of generating the mask constant.
// We model this for an unknown mask with a single vrgather.
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
index 65bbd905508557..9d6317baabc658 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -55,6 +55,12 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
/// type.
InstructionCost getConstantPoolLoadCost(Type *Ty,
TTI::TargetCostKind CostKind);
+
+ /// Return the cost if a shufflevector can be consist of multiple vslideup.
+ /// Otherwise, return InstructionCost::getInvalid().
+ InstructionCost isMultipleInsertSubvector(VectorType *Tp, ArrayRef<int> Mask,
+ TTI::TargetCostKind CostKind);
+
public:
explicit RISCVTTIImpl(const RISCVTargetMachine *TM, const Function &F)
: BaseT(TM, F.getDataLayout()), ST(TM->getSubtargetImpl(F)),
diff --git a/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll b/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll
new file mode 100644
index 00000000000000..47a2af92aee950
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/fixed-vector-insert-subvector.ll
@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: 'test'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %0 = shufflevector <8 x float> poison, <8 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %1 = shufflevector <4 x i16> poison, <4 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %2 = shufflevector <4 x float> poison, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %3 = shufflevector <2 x i1> poison, <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+entry:
+ %0 = shufflevector <8 x float> poison, <8 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %1 = shufflevector <4 x i16> poison, <4 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <4 x float> poison, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ %3 = shufflevector <2 x i1> poison, <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ ret void
+}
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
index bb806be15c71ca..23a9a654c96f9e 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
@@ -8,7 +8,7 @@
; YAML-NEXT: Function: test
; YAML-NEXT: Args:
; YAML-NEXT: - String: 'Stores SLP vectorized with cost '
-; YAML-NEXT: - Cost: '2'
+; YAML-NEXT: - Cost: '0'
; YAML-NEXT: - String: ' and with tree size '
; YAML-NEXT: - TreeSize: '7'
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
index 995cd7cfbc880b..a0cb52a853b7e6 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/revec-getGatherCost.ll
@@ -8,7 +8,7 @@
; YAML: Function: test1
; YAML: Args:
; YAML: - String: 'Stores SLP vectorized with cost '
-; YAML: - Cost: '6'
+; YAML: - Cost: '4'
; YAML: - String: ' and with tree size '
; YAML: - TreeSize: '5'
@@ -47,7 +47,7 @@ declare <4 x float> @llvm.fmuladd.v4f32(<4 x float>, <4 x float>, <4 x float>)
; YAML: Function: test2
; YAML: Args:
; YAML: - String: 'Stores SLP vectorized with cost '
-; YAML: - Cost: '16'
+; YAML: - Cost: '12'
; YAML: - String: ' and with tree size '
; YAML: - TreeSize: '5'
|
topperc
reviewed
Oct 8, 2024
lukel97
reviewed
Oct 8, 2024
topperc
reviewed
Oct 8, 2024
lukel97
reviewed
Oct 8, 2024
topperc
reviewed
Oct 8, 2024
|
ping |
lukel97
reviewed
Oct 15, 2024
| @@ -343,6 +343,52 @@ RISCVTTIImpl::getConstantPoolLoadCost(Type *Ty, TTI::TargetCostKind CostKind) { | |||
| /*AddressSpace=*/0, CostKind); | |||
| } | |||
|
|
|||
| InstructionCost | |||
| RISCVTTIImpl::isMultipleInsertSubvector(VectorType *Tp, ArrayRef<int> Mask, | |||
| TTI::TargetCostKind CostKind) { | |||
There was a problem hiding this comment.
Can we split this function into something similar to ShuffleVectorInst::isReplicationMask and the other mask functions, e.g. static bool isRepeatedConcatMask(ArrayRef<int> Mask, int &SubvectorSize).
Then call that check in getShuffleCost and compute the cost there if it's true. That way it's a bit more similar to how we check other mask patterns.
There was a problem hiding this comment.
Are you saying that we should add this into ShuffleVectorInst and let RISCVTTIImpl::getShuffleCost call it?
There was a problem hiding this comment.
Not move it into getShuffleCost, but just convert it into a static function that only checks the mask
There was a problem hiding this comment.
please check 39c1368
multiple insert subvector.
HanKuanChen
force-pushed
the
riscv-tti-isMultipleInsertSubvector
branch
from
e29eb38 to
39c1368
Compare
lukel97
approved these changes
Oct 18, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
reference: #110457