[mlir][spirv] Linearize ND vectors. #80451
Closed
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
SPIR-V only supports 1D vectors, try to linearize vector in type converter.
|
@llvm/pr-subscribers-mlir-spirv Author: Ivan Butygin (Hardcode84) ChangesSPIR-V only supports 1D vectors, try to linearize vector in type converter. Not sure is this is a right approach or we should have a dedicated vector linearization pass. Full diff: https://github.com/llvm/llvm-project/pull/80451.diff 4 Files Affected:
diff --git a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
index edf81bd7a8f39..b55eda69f99ec 100644
--- a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
+++ b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
@@ -229,8 +229,8 @@ struct ConstantCompositeOpPattern final
if (!srcType || srcType.getNumElements() == 1)
return failure();
- // arith.constant should only have vector or tenor types.
- assert((isa<VectorType, RankedTensorType>(srcType)));
+ assert((isa<VectorType, RankedTensorType>(srcType) &&
+ "arith.constant should only have vector or tensor types"));
Type dstType = getTypeConverter()->convertType(srcType);
if (!dstType)
@@ -250,8 +250,9 @@ struct ConstantCompositeOpPattern final
srcType.getElementType());
dstElementsAttr = dstElementsAttr.reshape(dstAttrType);
} else {
- // TODO: add support for large vectors.
- return failure();
+ dstAttrType =
+ VectorType::get(srcType.getNumElements(), srcType.getElementType());
+ dstElementsAttr = dstElementsAttr.reshape(dstAttrType);
}
}
diff --git a/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp b/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
index 2b79c8022b8e8..1ce7dff8ff0e4 100644
--- a/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
+++ b/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
@@ -330,6 +330,10 @@ convertVectorType(const spirv::TargetEnv &targetEnv,
if (type.getRank() <= 1 && type.getNumElements() == 1)
return convertScalarType(targetEnv, options, scalarType, storageClass);
+ // Linearize ND vectors
+ if (type.getRank() > 1)
+ type = VectorType::get(type.getNumElements(), scalarType);
+
if (!spirv::CompositeType::isValid(type)) {
LLVM_DEBUG(llvm::dbgs()
<< type << " illegal: not a valid composite type\n");
diff --git a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
index 0d92a8e676d85..551b036ba85e5 100644
--- a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
+++ b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
@@ -26,9 +26,9 @@ module attributes {
#spirv.vce<v1.0, [Int8, Int16, Int64, Float16, Float64, Shader], []>, #spirv.resource_limits<>>
} {
-func.func @unsupported_2x2elem_vector(%arg0: vector<2x2xi32>) {
+func.func @unsupported_2x2elem_vector(%arg0: vector<3x5xi32>) {
// expected-error@+1 {{failed to legalize operation 'arith.muli'}}
- %2 = arith.muli %arg0, %arg0: vector<2x2xi32>
+ %2 = arith.muli %arg0, %arg0: vector<3x5xi32>
return
}
diff --git a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
index ae47ae36ca51c..4a2ef1f0275c6 100644
--- a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
+++ b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
@@ -556,6 +556,8 @@ func.func @constant() {
%9 = arith.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi32>
// CHECK: spirv.Constant dense<{{\[}}1, 2, 3, 4, 5, 6]> : tensor<6xi32> : !spirv.array<6 x i32>
%10 = arith.constant dense<[1, 2, 3, 4, 5, 6]> : tensor<6xi32>
+ // CHECK: spirv.Constant dense<[1.000000e+00, 2.000000e+00, 3.000000e+00, 4.000000e+00]> : vector<4xf32>
+ %11 = arith.constant dense<[[1.0, 2.0], [3.0, 4.0]]> : vector<2x2xf32>
return
}
|
|
@llvm/pr-subscribers-mlir Author: Ivan Butygin (Hardcode84) ChangesSPIR-V only supports 1D vectors, try to linearize vector in type converter. Not sure is this is a right approach or we should have a dedicated vector linearization pass. Full diff: https://github.com/llvm/llvm-project/pull/80451.diff 4 Files Affected:
diff --git a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
index edf81bd7a8f39..b55eda69f99ec 100644
--- a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
+++ b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
@@ -229,8 +229,8 @@ struct ConstantCompositeOpPattern final
if (!srcType || srcType.getNumElements() == 1)
return failure();
- // arith.constant should only have vector or tenor types.
- assert((isa<VectorType, RankedTensorType>(srcType)));
+ assert((isa<VectorType, RankedTensorType>(srcType) &&
+ "arith.constant should only have vector or tensor types"));
Type dstType = getTypeConverter()->convertType(srcType);
if (!dstType)
@@ -250,8 +250,9 @@ struct ConstantCompositeOpPattern final
srcType.getElementType());
dstElementsAttr = dstElementsAttr.reshape(dstAttrType);
} else {
- // TODO: add support for large vectors.
- return failure();
+ dstAttrType =
+ VectorType::get(srcType.getNumElements(), srcType.getElementType());
+ dstElementsAttr = dstElementsAttr.reshape(dstAttrType);
}
}
diff --git a/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp b/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
index 2b79c8022b8e8..1ce7dff8ff0e4 100644
--- a/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
+++ b/mlir/lib/Dialect/SPIRV/Transforms/SPIRVConversion.cpp
@@ -330,6 +330,10 @@ convertVectorType(const spirv::TargetEnv &targetEnv,
if (type.getRank() <= 1 && type.getNumElements() == 1)
return convertScalarType(targetEnv, options, scalarType, storageClass);
+ // Linearize ND vectors
+ if (type.getRank() > 1)
+ type = VectorType::get(type.getNumElements(), scalarType);
+
if (!spirv::CompositeType::isValid(type)) {
LLVM_DEBUG(llvm::dbgs()
<< type << " illegal: not a valid composite type\n");
diff --git a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
index 0d92a8e676d85..551b036ba85e5 100644
--- a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
+++ b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv-unsupported.mlir
@@ -26,9 +26,9 @@ module attributes {
#spirv.vce<v1.0, [Int8, Int16, Int64, Float16, Float64, Shader], []>, #spirv.resource_limits<>>
} {
-func.func @unsupported_2x2elem_vector(%arg0: vector<2x2xi32>) {
+func.func @unsupported_2x2elem_vector(%arg0: vector<3x5xi32>) {
// expected-error@+1 {{failed to legalize operation 'arith.muli'}}
- %2 = arith.muli %arg0, %arg0: vector<2x2xi32>
+ %2 = arith.muli %arg0, %arg0: vector<3x5xi32>
return
}
diff --git a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
index ae47ae36ca51c..4a2ef1f0275c6 100644
--- a/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
+++ b/mlir/test/Conversion/ArithToSPIRV/arith-to-spirv.mlir
@@ -556,6 +556,8 @@ func.func @constant() {
%9 = arith.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi32>
// CHECK: spirv.Constant dense<{{\[}}1, 2, 3, 4, 5, 6]> : tensor<6xi32> : !spirv.array<6 x i32>
%10 = arith.constant dense<[1, 2, 3, 4, 5, 6]> : tensor<6xi32>
+ // CHECK: spirv.Constant dense<[1.000000e+00, 2.000000e+00, 3.000000e+00, 4.000000e+00]> : vector<4xf32>
+ %11 = arith.constant dense<[[1.0, 2.0], [3.0, 4.0]]> : vector<2x2xf32>
return
}
|
|
Alternative approach #81159 |
|
Closed in favor of #81159 |
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.
SPIR-V only supports 1D vectors, try to linearize vector in type converter.
Not sure is this is a right approach or we should have a dedicated vector linearization pass.