[TOSA] FFT2D operator #77005
Merged
[TOSA] FFT2D operator #77005
Conversation
This file contains 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-mlir @llvm/pr-subscribers-mlir-linalg Author: Dmitriy Smirnov (d-smirnov) ChangesThis PR adds lowering for TOSA Fft2d operator. @eric-k256 @GeorgeARM Full diff: https://github.com/llvm/llvm-project/pull/77005.diff 2 Files Affected:
diff --git a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
index 7c35389f1e9290..c28bd496f4a697 100644
--- a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
+++ b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
@@ -2341,6 +2341,134 @@ struct RFFT2dConverter final : public OpRewritePattern<RFFT2dOp> {
}
};
+struct FFT2dConverter final : public OpRewritePattern<FFT2dOp> {
+ using OpRewritePattern<FFT2dOp>::OpRewritePattern;
+
+ LogicalResult matchAndRewrite(FFT2dOp fft2d,
+ PatternRewriter &rewriter) const override {
+ if (!llvm::all_of(fft2d->getOperandTypes(),
+ RFFT2dConverter::isRankedTensor) ||
+ !llvm::all_of(fft2d->getResultTypes(),
+ RFFT2dConverter::isRankedTensor)) {
+ return rewriter.notifyMatchFailure(fft2d, "only supports ranked tensors");
+ }
+
+ auto loc = fft2d.getLoc();
+ auto input_real = fft2d.getInputReal();
+ auto input_imag = fft2d.getInputImag();
+ auto inverse = fft2d.getInverseAttr();
+
+ auto real_el_ty = input_real.getType()
+ .cast<ShapedType>()
+ .getElementType()
+ .cast<FloatType>();
+ auto imag_el_ty = input_imag.getType()
+ .cast<ShapedType>()
+ .getElementType()
+ .cast<FloatType>();
+
+ assert(real_el_ty == imag_el_ty);
+
+ // Compute the output type and set of dynamic sizes
+ llvm::SmallVector<Value> dynamicSizes;
+
+ // Get [N, H, W]
+ auto dims = tensor::getMixedSizes(rewriter, loc, input_real);
+
+ llvm::SmallVector<int64_t, 3> staticSizes;
+ dispatchIndexOpFoldResults(dims, dynamicSizes, staticSizes);
+
+ auto outputType = RankedTensorType::get(staticSizes, real_el_ty);
+
+ // Iterator types for the linalg.generic implementation
+ llvm::SmallVector<utils::IteratorType, 5> iteratorTypes = {
+ utils::IteratorType::parallel, utils::IteratorType::parallel,
+ utils::IteratorType::parallel, utils::IteratorType::reduction,
+ utils::IteratorType::reduction};
+
+ // Inputs/outputs to the linalg.generic implementation
+ llvm::SmallVector<Value> genericOpInputs = {input_real, input_imag};
+ llvm::SmallVector<Value> genericOpOutputs = {
+ RFFT2dConverter::createZeroTensor(rewriter, loc, outputType,
+ dynamicSizes),
+ RFFT2dConverter::createZeroTensor(rewriter, loc, outputType,
+ dynamicSizes)};
+
+ // Indexing maps for input and output tensors
+ auto indexingMaps = AffineMap::inferFromExprList(
+ llvm::ArrayRef{RFFT2dConverter::affineDimsExpr(rewriter, 0, 3, 4),
+ RFFT2dConverter::affineDimsExpr(rewriter, 0, 3, 4),
+ RFFT2dConverter::affineDimsExpr(rewriter, 0, 1, 2),
+ RFFT2dConverter::affineDimsExpr(rewriter, 0, 1, 2)});
+
+ // Width and height dimensions of the original input.
+ auto dimH = rewriter.createOrFold<tensor::DimOp>(loc, input_real, 1);
+ auto dimW = rewriter.createOrFold<tensor::DimOp>(loc, input_real, 2);
+
+ // Constants and dimension sizes
+ auto twoPiAttr = rewriter.getFloatAttr(real_el_ty, 6.283185307179586);
+ auto twoPi = rewriter.create<arith::ConstantOp>(loc, twoPiAttr);
+ auto constH =
+ RFFT2dConverter::castIndexToFloat(rewriter, loc, real_el_ty, dimH);
+ auto constW =
+ RFFT2dConverter::castIndexToFloat(rewriter, loc, real_el_ty, dimW);
+
+ auto buildBody = [&](OpBuilder &builder, Location loc, ValueRange args) {
+ Value valReal = args[0];
+ Value valImag = args[1];
+ Value sumReal = args[2];
+ Value sumImag = args[3];
+
+ // Indices for angle computation
+ auto oy = RFFT2dConverter::createLinalgIndex(builder, loc, real_el_ty, 1);
+ auto ox = RFFT2dConverter::createLinalgIndex(builder, loc, real_el_ty, 2);
+ auto iy = RFFT2dConverter::createLinalgIndex(builder, loc, real_el_ty, 3);
+ auto ix = RFFT2dConverter::createLinalgIndex(builder, loc, real_el_ty, 4);
+
+ // float_t angle = sign_val * 2 * pi() * ((iy * oy) / H + (ix * ox) / W);
+ auto iyXoy = builder.create<arith::MulFOp>(loc, iy, oy);
+ auto ixXox = builder.create<arith::MulFOp>(loc, ix, ox);
+ auto yComponent = builder.create<arith::DivFOp>(loc, iyXoy, constH);
+ auto xComponent = builder.create<arith::DivFOp>(loc, ixXox, constW);
+ auto sumXY = builder.create<arith::AddFOp>(loc, yComponent, xComponent);
+ auto angle = builder.create<arith::MulFOp>(loc, twoPi, sumXY);
+ if (inverse.getValue()) {
+ angle = builder.create<arith::MulFOp>(
+ loc, angle,
+ rewriter.create<arith::ConstantOp>(
+ loc, rewriter.getFloatAttr(real_el_ty, -1.0)));
+ }
+
+ // realComponent = val_real * cos(a) + val_imag * sin(a);
+ // imagComponent = -val_real * sin(a) + val_imag * cos(a);
+ auto cosAngle = builder.create<math::CosOp>(loc, angle);
+ auto sinAngle = builder.create<math::SinOp>(loc, angle);
+
+ auto rcos = builder.create<arith::MulFOp>(loc, valReal, cosAngle);
+ auto rsin = builder.create<arith::MulFOp>(loc, valImag, sinAngle);
+ auto realComponent = builder.create<arith::AddFOp>(loc, rcos, rsin);
+
+ auto icos = builder.create<arith::MulFOp>(loc, valImag, cosAngle);
+ auto isin = builder.create<arith::MulFOp>(loc, valReal, sinAngle);
+
+ auto imagComponent = builder.create<arith::SubFOp>(loc, icos, isin);
+
+ // outReal = sumReal + realComponent
+ // outImag = sumImag - imagComponent
+ auto outReal = builder.create<arith::AddFOp>(loc, sumReal, realComponent);
+ auto outImag = builder.create<arith::AddFOp>(loc, sumImag, imagComponent);
+
+ builder.create<linalg::YieldOp>(loc, ValueRange{outReal, outImag});
+ };
+
+ rewriter.replaceOpWithNewOp<linalg::GenericOp>(
+ fft2d, fft2d.getResultTypes(), genericOpInputs, genericOpOutputs,
+ indexingMaps, iteratorTypes, buildBody);
+
+ return success();
+ }
+};
+
} // namespace
void mlir::tosa::populateTosaToLinalgConversionPatterns(
@@ -2404,6 +2532,7 @@ void mlir::tosa::populateTosaToLinalgConversionPatterns(
RescaleConverter,
ReverseConverter,
RFFT2dConverter,
+ FFT2dConverter,
TableConverter,
TileConverter>(patterns->getContext());
// clang-format on
diff --git a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir
index 535316e5a37210..1cccddd8cd9179 100644
--- a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir
+++ b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg.mlir
@@ -1739,3 +1739,137 @@ func.func @test_dynamic_rfft2d(%arg0: tensor<?x?x?xf32>) -> (tensor<?x?x?xf32>,
%output_real, %output_imag = "tosa.rfft2d"(%arg0) {} : (tensor<?x?x?xf32>) -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>)
return %output_real, %output_imag : tensor<?x?x?xf32>, tensor<?x?x?xf32>
}
+
+// -----
+// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
+
+// CHECK: #[[$ATTR_0:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d3, d4)>
+// CHECK: #[[$ATTR_1:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2)>
+
+// CHECK-LABEL: func.func @test_static_fft2d(
+// CHECK-SAME: %[[VAL_0:.*]]: tensor<8x8x8xf32>,
+// CHECK-SAME: %[[VAL_1:.*]]: tensor<8x8x8xf32>) -> (tensor<8x8x8xf32>, tensor<8x8x8xf32>) {
+// CHECK: %[[VAL_2:.*]] = tensor.empty() : tensor<8x8x8xf32>
+// CHECK: %[[VAL_3:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_4:.*]] = linalg.fill ins(%[[VAL_3]] : f32) outs(%[[VAL_2]] : tensor<8x8x8xf32>) -> tensor<8x8x8xf32>
+// CHECK: %[[VAL_5:.*]] = tensor.empty() : tensor<8x8x8xf32>
+// CHECK: %[[VAL_6:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_7:.*]] = linalg.fill ins(%[[VAL_6]] : f32) outs(%[[VAL_5]] : tensor<8x8x8xf32>) -> tensor<8x8x8xf32>
+// CHECK: %[[VAL_8:.*]] = arith.constant 1 : index
+// CHECK: %[[VAL_9:.*]] = arith.constant 8 : index
+// CHECK: %[[VAL_10:.*]] = arith.constant 2 : index
+// CHECK: %[[VAL_11:.*]] = arith.constant 8 : index
+// CHECK: %[[VAL_12:.*]] = arith.constant 6.28318548 : f32
+// CHECK: %[[VAL_13:.*]] = arith.index_castui %[[VAL_9]] : index to i32
+// CHECK: %[[VAL_14:.*]] = arith.uitofp %[[VAL_13]] : i32 to f32
+// CHECK: %[[VAL_15:.*]] = arith.index_castui %[[VAL_11]] : index to i32
+// CHECK: %[[VAL_16:.*]] = arith.uitofp %[[VAL_15]] : i32 to f32
+// CHECK: %[[VAL_17:.*]]:2 = linalg.generic {indexing_maps = [#[[$ATTR_0]], #[[$ATTR_0]], #[[$ATTR_1]], #[[$ATTR_1]]], iterator_types = ["parallel", "parallel", "parallel", "reduction", "reduction"]} ins(%[[VAL_0]], %[[VAL_1]] : tensor<8x8x8xf32>, tensor<8x8x8xf32>) outs(%[[VAL_4]], %[[VAL_7]] : tensor<8x8x8xf32>, tensor<8x8x8xf32>) {
+// CHECK: ^bb0(%[[VAL_18:.*]]: f32, %[[VAL_19:.*]]: f32, %[[VAL_20:.*]]: f32, %[[VAL_21:.*]]: f32):
+// CHECK: %[[VAL_22:.*]] = linalg.index 1 : index
+// CHECK: %[[VAL_23:.*]] = arith.index_castui %[[VAL_22]] : index to i32
+// CHECK: %[[VAL_24:.*]] = arith.uitofp %[[VAL_23]] : i32 to f32
+// CHECK: %[[VAL_25:.*]] = linalg.index 2 : index
+// CHECK: %[[VAL_26:.*]] = arith.index_castui %[[VAL_25]] : index to i32
+// CHECK: %[[VAL_27:.*]] = arith.uitofp %[[VAL_26]] : i32 to f32
+// CHECK: %[[VAL_28:.*]] = linalg.index 3 : index
+// CHECK: %[[VAL_29:.*]] = arith.index_castui %[[VAL_28]] : index to i32
+// CHECK: %[[VAL_30:.*]] = arith.uitofp %[[VAL_29]] : i32 to f32
+// CHECK: %[[VAL_31:.*]] = linalg.index 4 : index
+// CHECK: %[[VAL_32:.*]] = arith.index_castui %[[VAL_31]] : index to i32
+// CHECK: %[[VAL_33:.*]] = arith.uitofp %[[VAL_32]] : i32 to f32
+// CHECK: %[[VAL_34:.*]] = arith.mulf %[[VAL_30]], %[[VAL_24]] : f32
+// CHECK: %[[VAL_35:.*]] = arith.mulf %[[VAL_33]], %[[VAL_27]] : f32
+// CHECK: %[[VAL_36:.*]] = arith.divf %[[VAL_34]], %[[VAL_14]] : f32
+// CHECK: %[[VAL_37:.*]] = arith.divf %[[VAL_35]], %[[VAL_16]] : f32
+// CHECK: %[[VAL_38:.*]] = arith.addf %[[VAL_36]], %[[VAL_37]] : f32
+// CHECK: %[[VAL_39:.*]] = arith.mulf %[[VAL_12]], %[[VAL_38]] : f32
+// CHECK: %[[VAL_40:.*]] = math.cos %[[VAL_39]] : f32
+// CHECK: %[[VAL_41:.*]] = math.sin %[[VAL_39]] : f32
+// CHECK: %[[VAL_42:.*]] = arith.mulf %[[VAL_18]], %[[VAL_40]] : f32
+// CHECK: %[[VAL_43:.*]] = arith.mulf %[[VAL_19]], %[[VAL_41]] : f32
+// CHECK: %[[VAL_44:.*]] = arith.addf %[[VAL_42]], %[[VAL_43]] : f32
+// CHECK: %[[VAL_45:.*]] = arith.mulf %[[VAL_19]], %[[VAL_40]] : f32
+// CHECK: %[[VAL_46:.*]] = arith.mulf %[[VAL_18]], %[[VAL_41]] : f32
+// CHECK: %[[VAL_47:.*]] = arith.subf %[[VAL_45]], %[[VAL_46]] : f32
+// CHECK: %[[VAL_48:.*]] = arith.addf %[[VAL_20]], %[[VAL_44]] : f32
+// CHECK: %[[VAL_49:.*]] = arith.addf %[[VAL_21]], %[[VAL_47]] : f32
+// CHECK: linalg.yield %[[VAL_48]], %[[VAL_49]] : f32, f32
+// CHECK: } -> (tensor<8x8x8xf32>, tensor<8x8x8xf32>)
+// CHECK: return %[[VAL_50:.*]]#0, %[[VAL_50]]#1 : tensor<8x8x8xf32>, tensor<8x8x8xf32>
+// CHECK: }
+func.func @test_static_fft2d(%arg0: tensor<8x8x8xf32>, %arg1: tensor<8x8x8xf32>) -> (tensor<8x8x8xf32>, tensor<8x8x8xf32>) {
+ %output_real, %output_imag = "tosa.fft2d"(%arg0, %arg1) {inverse=false} : (tensor<8x8x8xf32>, tensor<8x8x8xf32>) -> (tensor<8x8x8xf32>, tensor<8x8x8xf32>)
+ return %output_real, %output_imag : tensor<8x8x8xf32>, tensor<8x8x8xf32>
+}
+
+// -----
+// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
+
+// CHECK: #[[$ATTR_2:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d3, d4)>
+// CHECK: #[[$ATTR_3:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2)>
+
+// CHECK-LABEL: func.func @test_dynamic_fft2d(
+// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?x?xf32>,
+// CHECK-SAME: %[[VAL_1:.*]]: tensor<?x?x?xf32>) -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>) {
+// CHECK: %[[VAL_2:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_3:.*]] = tensor.dim %[[VAL_0]], %[[VAL_2]] : tensor<?x?x?xf32>
+// CHECK: %[[VAL_4:.*]] = arith.constant 1 : index
+// CHECK: %[[VAL_5:.*]] = tensor.dim %[[VAL_0]], %[[VAL_4]] : tensor<?x?x?xf32>
+// CHECK: %[[VAL_6:.*]] = arith.constant 2 : index
+// CHECK: %[[VAL_7:.*]] = tensor.dim %[[VAL_0]], %[[VAL_6]] : tensor<?x?x?xf32>
+// CHECK: %[[VAL_8:.*]] = tensor.empty(%[[VAL_3]], %[[VAL_5]], %[[VAL_7]]) : tensor<?x?x?xf32>
+// CHECK: %[[VAL_9:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_10:.*]] = linalg.fill ins(%[[VAL_9]] : f32) outs(%[[VAL_8]] : tensor<?x?x?xf32>) -> tensor<?x?x?xf32>
+// CHECK: %[[VAL_11:.*]] = tensor.empty(%[[VAL_3]], %[[VAL_5]], %[[VAL_7]]) : tensor<?x?x?xf32>
+// CHECK: %[[VAL_12:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_13:.*]] = linalg.fill ins(%[[VAL_12]] : f32) outs(%[[VAL_11]] : tensor<?x?x?xf32>) -> tensor<?x?x?xf32>
+// CHECK: %[[VAL_14:.*]] = arith.constant 1 : index
+// CHECK: %[[VAL_15:.*]] = tensor.dim %[[VAL_0]], %[[VAL_14]] : tensor<?x?x?xf32>
+// CHECK: %[[VAL_16:.*]] = arith.constant 2 : index
+// CHECK: %[[VAL_17:.*]] = tensor.dim %[[VAL_0]], %[[VAL_16]] : tensor<?x?x?xf32>
+// CHECK: %[[VAL_18:.*]] = arith.constant 6.28318548 : f32
+// CHECK: %[[VAL_19:.*]] = arith.index_castui %[[VAL_15]] : index to i32
+// CHECK: %[[VAL_20:.*]] = arith.uitofp %[[VAL_19]] : i32 to f32
+// CHECK: %[[VAL_21:.*]] = arith.index_castui %[[VAL_17]] : index to i32
+// CHECK: %[[VAL_22:.*]] = arith.uitofp %[[VAL_21]] : i32 to f32
+// CHECK: %[[VAL_23:.*]]:2 = linalg.generic {indexing_maps = [#[[$ATTR_2]], #[[$ATTR_2]], #[[$ATTR_3]], #[[$ATTR_3]]], iterator_types = ["parallel", "parallel", "parallel", "reduction", "reduction"]} ins(%[[VAL_0]], %[[VAL_1]] : tensor<?x?x?xf32>, tensor<?x?x?xf32>) outs(%[[VAL_10]], %[[VAL_13]] : tensor<?x?x?xf32>, tensor<?x?x?xf32>) {
+// CHECK: ^bb0(%[[VAL_24:.*]]: f32, %[[VAL_25:.*]]: f32, %[[VAL_26:.*]]: f32, %[[VAL_27:.*]]: f32):
+// CHECK: %[[VAL_28:.*]] = linalg.index 1 : index
+// CHECK: %[[VAL_29:.*]] = arith.index_castui %[[VAL_28]] : index to i32
+// CHECK: %[[VAL_30:.*]] = arith.uitofp %[[VAL_29]] : i32 to f32
+// CHECK: %[[VAL_31:.*]] = linalg.index 2 : index
+// CHECK: %[[VAL_32:.*]] = arith.index_castui %[[VAL_31]] : index to i32
+// CHECK: %[[VAL_33:.*]] = arith.uitofp %[[VAL_32]] : i32 to f32
+// CHECK: %[[VAL_34:.*]] = linalg.index 3 : index
+// CHECK: %[[VAL_35:.*]] = arith.index_castui %[[VAL_34]] : index to i32
+// CHECK: %[[VAL_36:.*]] = arith.uitofp %[[VAL_35]] : i32 to f32
+// CHECK: %[[VAL_37:.*]] = linalg.index 4 : index
+// CHECK: %[[VAL_38:.*]] = arith.index_castui %[[VAL_37]] : index to i32
+// CHECK: %[[VAL_39:.*]] = arith.uitofp %[[VAL_38]] : i32 to f32
+// CHECK: %[[VAL_40:.*]] = arith.mulf %[[VAL_36]], %[[VAL_30]] : f32
+// CHECK: %[[VAL_41:.*]] = arith.mulf %[[VAL_39]], %[[VAL_33]] : f32
+// CHECK: %[[VAL_42:.*]] = arith.divf %[[VAL_40]], %[[VAL_20]] : f32
+// CHECK: %[[VAL_43:.*]] = arith.divf %[[VAL_41]], %[[VAL_22]] : f32
+// CHECK: %[[VAL_44:.*]] = arith.addf %[[VAL_42]], %[[VAL_43]] : f32
+// CHECK: %[[VAL_45:.*]] = arith.mulf %[[VAL_18]], %[[VAL_44]] : f32
+// CHECK: %[[VAL_46:.*]] = arith.constant -1.000000e+00 : f32
+// CHECK: %[[VAL_47:.*]] = arith.mulf %[[VAL_45]], %[[VAL_46]] : f32
+// CHECK: %[[VAL_48:.*]] = math.cos %[[VAL_47]] : f32
+// CHECK: %[[VAL_49:.*]] = math.sin %[[VAL_47]] : f32
+// CHECK: %[[VAL_50:.*]] = arith.mulf %[[VAL_24]], %[[VAL_48]] : f32
+// CHECK: %[[VAL_51:.*]] = arith.mulf %[[VAL_25]], %[[VAL_49]] : f32
+// CHECK: %[[VAL_52:.*]] = arith.addf %[[VAL_50]], %[[VAL_51]] : f32
+// CHECK: %[[VAL_53:.*]] = arith.mulf %[[VAL_25]], %[[VAL_48]] : f32
+// CHECK: %[[VAL_54:.*]] = arith.mulf %[[VAL_24]], %[[VAL_49]] : f32
+// CHECK: %[[VAL_55:.*]] = arith.subf %[[VAL_53]], %[[VAL_54]] : f32
+// CHECK: %[[VAL_56:.*]] = arith.addf %[[VAL_26]], %[[VAL_52]] : f32
+// CHECK: %[[VAL_57:.*]] = arith.addf %[[VAL_27]], %[[VAL_55]] : f32
+// CHECK: linalg.yield %[[VAL_56]], %[[VAL_57]] : f32, f32
+// CHECK: } -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>)
+// CHECK: return %[[VAL_58:.*]]#0, %[[VAL_58]]#1 : tensor<?x?x?xf32>, tensor<?x?x?xf32>
+// CHECK: }
+func.func @test_dynamic_fft2d(%arg0: tensor<?x?x?xf32>, %arg1: tensor<?x?x?xf32>) -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>) {
+ %output_real, %output_imag = "tosa.fft2d"(%arg0, %arg1) {inverse = true} : (tensor<?x?x?xf32>, tensor<?x?x?xf32>) -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>)
+ return %output_real, %output_imag : tensor<?x?x?xf32>, tensor<?x?x?xf32>
+}
|
|
Tests failed due to dependency on 75098 |
kuhar
reviewed
Jan 10, 2024
|
Thanks @d-smirnov! |
GeorgeARM
approved these changes
Jan 11, 2024
|
There is a bug in mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp line 2375 : tensor::getMixedSizes returns a SmallVector so using an ArrayRef would mean the actual data will be destroyed as soon as this call is done. You should either use auto or SmallVector. Also, can you add [[maybe_unused]] to imag_el_ty as it is only used in an assert? (line 2366) Can you please fix? |
Someone sent this fix 21133f1da4 |
justinfargnoli
pushed a commit
to justinfargnoli/llvm-project
that referenced
this pull request
Jan 28, 2024
This PR adds lowering for TOSA Fft2d operator down to Linalg.
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.
This PR adds lowering for TOSA Fft2d operator.
@eric-k256 @GeorgeARM