let builders = [Tosa_ConvOpQuantInfoBuilder];
let verifier = [{ return verifyConvOp(*this); }];
-
- let hasCanonicalizer = 1;
}
//===----------------------------------------------------------------------===//
let builders = [Tosa_ConvOpQuantInfoBuilder];
let verifier = [{ return verifyConvOp(*this); }];
-
- let hasCanonicalizer = 1;
}
//===----------------------------------------------------------------------===//
std::unique_ptr<Pass> createTosaDecomposeTransposeConvPass();
std::unique_ptr<Pass> createTosaInferShapesPass();
std::unique_ptr<Pass> createTosaMakeBroadcastablePass();
+std::unique_ptr<Pass> createTosaOptimizationPass();
std::unique_ptr<Pass> createTosaTestQuantUtilAPIPass();
#define GEN_PASS_REGISTRATION
-//===-- Passes.td - TOSA optimization pass declarations ----*- tablegen -*-===//
+//===-- Passes.td - TOSA pass declarations ----*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
//
//===----------------------------------------------------------------------===//
//
-// This file declares the optimization passes for the TOSA Dialect in MLIR.
+// This file declares the passes for the TOSA Dialect in MLIR.
//
//===----------------------------------------------------------------------===//
let constructor = "createTosaMakeBroadcastablePass()";
}
+def TosaOptimization : FunctionPass<"tosa-optimization"> {
+ let summary = "TOSA operation optimizations";
+ let description = [{
+ "Pass to perform optimizations on TOSA operations"
+ }];
+
+ let constructor = "createTosaOptimizationPass()";
+}
+
#endif // MLIR_DIALECT_TOSA_TRANSFORMS_PASSES
results.insert<MaterializePadValue>(context);
}
-struct Conv2DIsFullyConnected : public OpRewritePattern<tosa::Conv2DOp> {
- using OpRewritePattern::OpRewritePattern;
-
- LogicalResult matchAndRewrite(tosa::Conv2DOp op,
- PatternRewriter &rewriter) const override {
- Value input = op.input();
- Value weight = op.weight();
- ShapedType inputType = input.getType().cast<ShapedType>();
- ShapedType weightType = weight.getType().cast<ShapedType>();
- ShapedType resultType = op.getType().cast<ShapedType>();
-
- if (!inputType.hasStaticShape() || !weightType.hasRank()) {
- return failure();
- }
-
- for (Attribute pad : op.pad().getValue()) {
- if (!pad.cast<IntegerAttr>().getValue().isZero()) {
- return failure();
- }
- }
-
- // Stride must be 1 for this optimization.
- for (Attribute stride : op.stride().getValue()) {
- if (!stride.cast<IntegerAttr>().getValue().isOne()) {
- return failure();
- }
- }
-
- // Only works for a 1x1 kernel.
- ArrayRef<int64_t> weightShape = weightType.getShape();
- if (weightShape[1] != 1 || weightShape[2] != 1) {
- return failure();
- }
-
- // Reshape input to [N,IH,IW,IC] -> [N * IH * IW, IC].
- ArrayRef<int64_t> inputShape = inputType.getShape();
- llvm::SmallVector<int64_t, 2> revisedInputShape{
- inputShape[0] * inputShape[1] * inputShape[2], inputShape[3]};
- auto revisedInputShapeType =
- RankedTensorType::get(revisedInputShape, inputType.getElementType());
- auto reshapedInput = rewriter
- .create<tosa::ReshapeOp>(
- op.getLoc(), revisedInputShapeType, input,
- rewriter.getI64ArrayAttr(revisedInputShape))
- .getResult();
-
- // Reshape kernel to [OC,KH,KW,IC] -> [OC, IC].
- llvm::SmallVector<int64_t, 2> revisedWeightShape{weightShape[0],
- weightShape[3]};
- auto revisedWeightShapeType =
- RankedTensorType::get(revisedWeightShape, weightType.getElementType());
- auto reshapedWeight = rewriter
- .create<tosa::ReshapeOp>(
- op.getLoc(), revisedWeightShapeType, weight,
- rewriter.getI64ArrayAttr(revisedWeightShape))
- .getResult();
-
- // Perform a fully connected network over the reshaped input and weight.
- llvm::SmallVector<int64_t, 2> fullyConnectedShape{
- inputShape[0] * inputShape[1] * inputShape[2], weightShape[0]};
- auto fullyConnectedShapeType =
- RankedTensorType::get(fullyConnectedShape, resultType.getElementType());
-
- Value fullyConnectedValue;
- if (op.quantization_info()) {
- fullyConnectedValue =
- rewriter
- .create<tosa::FullyConnectedOp>(
- op.getLoc(), fullyConnectedShapeType, reshapedInput,
- reshapedWeight, op.bias(), op.quantization_info().getValue())
- .getResult();
- } else {
- fullyConnectedValue = rewriter
- .create<tosa::FullyConnectedOp>(
- op.getLoc(), fullyConnectedShapeType,
- reshapedInput, reshapedWeight, op.bias())
- .getResult();
- }
-
- // Reshape output to [N, IH, IW, OC].
- llvm::SmallVector<int64_t, 4> outputShape{inputShape[0], inputShape[1],
- inputShape[2], weightShape[0]};
- rewriter.replaceOpWithNewOp<tosa::ReshapeOp>(
- op, resultType, fullyConnectedValue,
- rewriter.getI64ArrayAttr(outputShape));
- return success();
- }
-};
-
-void Conv2DOp::getCanonicalizationPatterns(OwningRewritePatternList &results,
- MLIRContext *context) {
- results.insert<Conv2DIsFullyConnected>(context);
-}
-
-struct DepthwiseConv2DMulOptimization
- : public OpRewritePattern<tosa::DepthwiseConv2DOp> {
- using OpRewritePattern::OpRewritePattern;
-
- LogicalResult matchAndRewrite(tosa::DepthwiseConv2DOp op,
- PatternRewriter &rewriter) const override {
- Value input = op.input();
- Value weight = op.weight();
- ShapedType inputType = input.getType().cast<ShapedType>();
- ShapedType weightType = weight.getType().cast<ShapedType>();
- ShapedType resultType = op.output().getType().cast<ShapedType>();
- Type inputEType = inputType.getElementType();
-
- if (!(inputType.hasStaticShape() && weightType.hasStaticShape() &&
- resultType.hasStaticShape())) {
- return failure();
- }
-
- // Quantization information needs to still be performed.
- if (op.quantization_info() || !inputEType.isa<FloatType>()) {
- return failure();
- }
-
- // Stride must be 1 for this optimization.
- for (Attribute stride : op.stride().getValue()) {
- if (!stride.cast<IntegerAttr>().getValue().isOne()) {
- return failure();
- }
- }
-
- // Only works for a 1x1 kernel.
- ArrayRef<int64_t> weightShape = weightType.getShape();
- if (weightShape[0] != 1 || weightShape[1] != 1) {
- return failure();
- }
-
- // Reshape input to [N, H, W, C] -> [N, H, W, C, 1].
- ArrayRef<int64_t> inputShape = inputType.getShape();
- llvm::SmallVector<int64_t, 2> revisedInputShape{
- inputShape[0], inputShape[1], inputShape[2], inputShape[3], 1};
- auto revisedInputShapeType = RankedTensorType::get(
- revisedInputShape,
- input.getType().dyn_cast<RankedTensorType>().getElementType());
- auto reshapedInput = rewriter
- .create<tosa::ReshapeOp>(
- op.getLoc(), revisedInputShapeType, input,
- rewriter.getI64ArrayAttr(revisedInputShape))
- .getResult();
-
- // Reshape kernel to [KH, KW, C, M] -> [1, 1, 1, C, M].
- llvm::SmallVector<int64_t, 2> revisedWeightShape{1, 1, 1, weightShape[2],
- weightShape[3]};
- auto revisedWeightShapeType = RankedTensorType::get(
- revisedWeightShape,
- weight.getType().dyn_cast<RankedTensorType>().getElementType());
- auto reshapedWeight = rewriter
- .create<tosa::ReshapeOp>(
- op.getLoc(), revisedWeightShapeType, weight,
- rewriter.getI64ArrayAttr(revisedWeightShape))
- .getResult();
-
- // Perform an elementwise mul over the reshaped input and weight.
- llvm::SmallVector<int64_t, 2> mulShape{inputShape[0], inputShape[1],
- inputShape[2], inputShape[3],
- weightShape[3]};
- auto mulShapeType = RankedTensorType::get(
- mulShape,
- weight.getType().dyn_cast<RankedTensorType>().getElementType());
- Value mulValue =
- rewriter
- .create<tosa::MulOp>(op.getLoc(), mulShapeType, reshapedInput,
- reshapedWeight, /*shift=*/0)
- .getResult();
-
- // Reshape output to [N, H, W, C * M].
- auto outputShape = op.output().getType().cast<ShapedType>().getShape();
- auto outputShapeType = RankedTensorType::get(
- outputShape,
- input.getType().dyn_cast<RankedTensorType>().getElementType());
- auto outputValue =
- rewriter.create<tosa::ReshapeOp>(op.getLoc(), outputShapeType, mulValue,
- rewriter.getI64ArrayAttr(outputShape));
-
- // Add in the bias.
- rewriter
- .replaceOpWithNewOp<tosa::AddOp>(op, outputShapeType, outputValue,
- op.bias())
- .getResult();
- return success();
- }
-};
-
-void DepthwiseConv2DOp::getCanonicalizationPatterns(
- OwningRewritePatternList &results, MLIRContext *context) {
- results.insert<DepthwiseConv2DMulOptimization>(context);
-}
-
struct MaxPool2dIsNoOp : public OpRewritePattern<tosa::MaxPool2dOp> {
using OpRewritePattern::OpRewritePattern;
// TOSA Operator Verifiers.
//===----------------------------------------------------------------------===//
-template <typename T> static LogicalResult verifyConvOp(T op) {
+template <typename T>
+static LogicalResult verifyConvOp(T op) {
// All TOSA conv ops have an input() and weight().
auto inputType = op.input().getType().template dyn_cast<RankedTensorType>();
auto weightType = op.weight().getType().template dyn_cast<RankedTensorType>();
TosaDecomposeTransposeConv.cpp
TosaInferShapes.cpp
TosaMakeBroadcastable.cpp
+ TosaOptimization.cpp
ADDITIONAL_HEADER_DIRS
${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/Tosa/Transforms
--- /dev/null
+//===- TosaOptimization.cpp ------------------------------------------===//\r
+//\r
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.\r
+// See https://llvm.org/LICENSE.txt for license information.\r
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception\r
+//\r
+//===----------------------------------------------------------------------===//\r
+//\r
+// Pass to perform optimizations on TOSA operations\r
+//\r
+//===----------------------------------------------------------------------===//\r
+\r
+#include "mlir/Analysis/DataFlowAnalysis.h"\r
+#include "mlir/Dialect/StandardOps/IR/Ops.h"\r
+#include "mlir/Dialect/Tensor/IR/Tensor.h"\r
+#include "mlir/Dialect/Tosa/IR/TosaOps.h"\r
+#include "mlir/Dialect/Tosa/Transforms/PassDetail.h"\r
+#include "mlir/Dialect/Tosa/Transforms/Passes.h"\r
+#include "mlir/Dialect/Tosa/Utils/ShapeUtils.h"\r
+#include "mlir/IR/BlockAndValueMapping.h"\r
+#include "mlir/IR/Builders.h"\r
+#include "mlir/IR/BuiltinOps.h"\r
+#include "mlir/IR/Matchers.h"\r
+#include "mlir/Pass/Pass.h"\r
+#include "mlir/Transforms/DialectConversion.h"\r
+#include "mlir/Transforms/GreedyPatternRewriteDriver.h"\r
+#include "llvm/Support/FormatVariadic.h"\r
+\r
+using namespace mlir;\r
+using namespace mlir::tosa;\r
+\r
+#define PASS_NAME "tosa-optimization"\r
+#define DEBUG_TYPE PASS_NAME\r
+\r
+namespace {\r
+\r
+struct Conv2DIsFullyConnected : public OpRewritePattern<tosa::Conv2DOp> {\r
+ explicit Conv2DIsFullyConnected(MLIRContext *context)\r
+ : OpRewritePattern(context) {}\r
+\r
+ LogicalResult matchAndRewrite(tosa::Conv2DOp op,\r
+ PatternRewriter &rewriter) const {\r
+ Value input = op.input();\r
+ Value weight = op.weight();\r
+ ShapedType inputType = input.getType().cast<ShapedType>();\r
+ ShapedType weightType = weight.getType().cast<ShapedType>();\r
+ ShapedType resultType = op.getType().cast<ShapedType>();\r
+\r
+ if (!inputType.hasStaticShape() || !weightType.hasRank()) {\r
+ return failure();\r
+ }\r
+\r
+ // Stride must be 1 for this optimization.\r
+ for (Attribute stride : op.stride().getValue()) {\r
+ if (!stride.cast<IntegerAttr>().getValue().isOne()) {\r
+ return failure();\r
+ }\r
+ }\r
+\r
+ // Only works for a 1x1 kernel.\r
+ ArrayRef<int64_t> weightShape = weightType.getShape();\r
+ if (weightShape[1] != 1 || weightShape[2] != 1) {\r
+ return failure();\r
+ }\r
+\r
+ // Reshape input to [N,IH,IW,IC] -> [N * IH * IW, IC].\r
+ ArrayRef<int64_t> inputShape = inputType.getShape();\r
+ llvm::SmallVector<int64_t, 2> revisedInputShape{\r
+ inputShape[0] * inputShape[1] * inputShape[2], inputShape[3]};\r
+ auto revisedInputShapeType = RankedTensorType::get(\r
+ revisedInputShape,\r
+ input.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ auto reshapedInput = rewriter\r
+ .create<tosa::ReshapeOp>(\r
+ op.getLoc(), revisedInputShapeType, input,\r
+ rewriter.getI64ArrayAttr(revisedInputShape))\r
+ .getResult();\r
+\r
+ // Reshape kernel to [OC,KH,KW,IC] -> [OC, IC].\r
+ llvm::SmallVector<int64_t, 2> revisedWeightShape{weightShape[0],\r
+ weightShape[3]};\r
+ auto revisedWeightShapeType = RankedTensorType::get(\r
+ revisedWeightShape,\r
+ weight.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ auto reshapedWeight = rewriter\r
+ .create<tosa::ReshapeOp>(\r
+ op.getLoc(), revisedWeightShapeType, weight,\r
+ rewriter.getI64ArrayAttr(revisedWeightShape))\r
+ .getResult();\r
+\r
+ // Perform a fully connected network over the reshaped input and weight.\r
+ llvm::SmallVector<int64_t, 2> fullyConnectedShape{\r
+ inputShape[0] * inputShape[1] * inputShape[2], weightShape[0]};\r
+ auto fullyConnectedShapeType = RankedTensorType::get(\r
+ fullyConnectedShape,\r
+ resultType.dyn_cast<ShapedType>().getElementType());\r
+\r
+ Value fullyConnectedValue;\r
+ if (op.quantization_info()) {\r
+ fullyConnectedValue =\r
+ rewriter\r
+ .create<tosa::FullyConnectedOp>(\r
+ op.getLoc(), fullyConnectedShapeType, reshapedInput,\r
+ reshapedWeight, op.bias(), op.quantization_info().getValue())\r
+ .getResult();\r
+ } else {\r
+ fullyConnectedValue = rewriter\r
+ .create<tosa::FullyConnectedOp>(\r
+ op.getLoc(), fullyConnectedShapeType,\r
+ reshapedInput, reshapedWeight, op.bias())\r
+ .getResult();\r
+ }\r
+\r
+ // Reshape output to [N, IH, IW, OC].\r
+ llvm::SmallVector<int64_t, 4> outputShape{inputShape[0], inputShape[1],\r
+ inputShape[2], weightShape[0]};\r
+ rewriter.replaceOpWithNewOp<tosa::ReshapeOp>(\r
+ op, resultType, fullyConnectedValue,\r
+ rewriter.getI64ArrayAttr(outputShape));\r
+ return success();\r
+ }\r
+};\r
+\r
+struct DepthwiseConv2DIsMul : public OpRewritePattern<tosa::DepthwiseConv2DOp> {\r
+ explicit DepthwiseConv2DIsMul(MLIRContext *context)\r
+ : OpRewritePattern(context) {}\r
+\r
+ LogicalResult matchAndRewrite(tosa::DepthwiseConv2DOp op,\r
+ PatternRewriter &rewriter) const {\r
+ Value input = op.input();\r
+ Value weight = op.weight();\r
+ ShapedType inputType = input.getType().cast<ShapedType>();\r
+ ShapedType weightType = weight.getType().cast<ShapedType>();\r
+ ShapedType resultType = op.output().getType().cast<ShapedType>();\r
+ Type inputEType = inputType.getElementType();\r
+\r
+ if (!(inputType.hasStaticShape() && weightType.hasStaticShape() &&\r
+ resultType.hasStaticShape())) {\r
+ return failure();\r
+ }\r
+\r
+ // Quantization information needs to still be performed.\r
+ if (op.quantization_info() || !inputEType.isa<FloatType>()) {\r
+ return failure();\r
+ }\r
+\r
+ // Stride must be 1 for this optimization.\r
+ for (Attribute stride : op.stride().getValue()) {\r
+ if (!stride.cast<IntegerAttr>().getValue().isOne()) {\r
+ return failure();\r
+ }\r
+ }\r
+\r
+ // Only works for a 1x1 kernel.\r
+ ArrayRef<int64_t> weightShape = weightType.getShape();\r
+ if (weightShape[0] != 1 || weightShape[1] != 1) {\r
+ return failure();\r
+ }\r
+\r
+ // Reshape input to [N, H, W, C] -> [N, H, W, C, 1].\r
+ ArrayRef<int64_t> inputShape = inputType.getShape();\r
+ llvm::SmallVector<int64_t, 2> revisedInputShape{\r
+ inputShape[0], inputShape[1], inputShape[2], inputShape[3], 1};\r
+ auto revisedInputShapeType = RankedTensorType::get(\r
+ revisedInputShape,\r
+ input.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ auto reshapedInput = rewriter\r
+ .create<tosa::ReshapeOp>(\r
+ op.getLoc(), revisedInputShapeType, input,\r
+ rewriter.getI64ArrayAttr(revisedInputShape))\r
+ .getResult();\r
+\r
+ // Reshape kernel to [KH, KW, C, M] -> [1, 1, 1, C, M].\r
+ llvm::SmallVector<int64_t, 2> revisedWeightShape{1, 1, 1, weightShape[2],\r
+ weightShape[3]};\r
+ auto revisedWeightShapeType = RankedTensorType::get(\r
+ revisedWeightShape,\r
+ weight.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ auto reshapedWeight = rewriter\r
+ .create<tosa::ReshapeOp>(\r
+ op.getLoc(), revisedWeightShapeType, weight,\r
+ rewriter.getI64ArrayAttr(revisedWeightShape))\r
+ .getResult();\r
+\r
+ // Perform an elementwise mul over the reshaped input and weight.\r
+ llvm::SmallVector<int64_t, 2> mulShape{inputShape[0], inputShape[1],\r
+ inputShape[2], inputShape[3],\r
+ weightShape[3]};\r
+ auto mulShapeType = RankedTensorType::get(\r
+ mulShape,\r
+ weight.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ Value mulValue =\r
+ rewriter\r
+ .create<tosa::MulOp>(op.getLoc(), mulShapeType, reshapedInput,\r
+ reshapedWeight, /*shift=*/0)\r
+ .getResult();\r
+\r
+ // Reshape output to [N, H, W, C * M].\r
+ auto outputShape = op.output().getType().cast<ShapedType>().getShape();\r
+ auto outputShapeType = RankedTensorType::get(\r
+ outputShape,\r
+ input.getType().dyn_cast<RankedTensorType>().getElementType());\r
+ auto outputValue =\r
+ rewriter.create<tosa::ReshapeOp>(op.getLoc(), outputShapeType, mulValue,\r
+ rewriter.getI64ArrayAttr(outputShape));\r
+\r
+ // Add in the bias.\r
+ rewriter\r
+ .replaceOpWithNewOp<tosa::AddOp>(op, outputShapeType, outputValue,\r
+ op.bias())\r
+ .getResult();\r
+ return success();\r
+ }\r
+};\r
+\r
+class TosaOptimization : public PassWrapper<TosaOptimization, FunctionPass> {\r
+public:\r
+ explicit TosaOptimization() {}\r
+ void runOnFunction() override;\r
+\r
+ StringRef getArgument() const final { return PASS_NAME; }\r
+ StringRef getDescription() const final {\r
+ return "Applies TOSA Operation Optimizations";\r
+ }\r
+};\r
+\r
+void TosaOptimization::runOnFunction() {\r
+ OwningRewritePatternList patterns(&getContext());\r
+\r
+ patterns.insert<Conv2DIsFullyConnected>(&getContext());\r
+ patterns.insert<DepthwiseConv2DIsMul>(&getContext());\r
+\r
+ auto func = getFunction();\r
+ if (applyPatternsAndFoldGreedily(func, std::move(patterns)).failed()) {\r
+ signalPassFailure();\r
+ }\r
+}\r
+\r
+} // namespace\r
+\r
+std::unique_ptr<Pass> mlir::tosa::createTosaOptimizationPass() {\r
+ return std::make_unique<TosaOptimization>();\r
+}\r
// -----
-// CHECK-LABEL: @conv2d_as_fully_connected
-func @conv2d_as_fully_connected(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<3x1x1x2xf32>, %arg2: tensor<3xf32>) -> tensor<4x10x10x3xf32> {
- // CHECK-NOT: "tosa.conv2d"
- // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [400, 2]}
- // CHECK-SAME: -> tensor<400x2xf32>
- // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [3, 2]}
- // CHECK-SAME: -> tensor<3x2xf32>
- // CHECK: %[[VAR2:.*]] = "tosa.fully_connected"(%[[VAR0]], %[[VAR1]], %arg2)
- // CHECK-SAME: -> tensor<400x3xf32>
- // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 3]}
- // CHECK-SAME: -> tensor<4x10x10x3xf32>
- // CHECK: return %[[VAR3]]
- %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1]} : (tensor<4x10x10x2xf32>, tensor<3x1x1x2xf32>, tensor<3xf32>) -> tensor<4x10x10x3xf32>
- return %0 : tensor<4x10x10x3xf32>
-}
-
-// -----
-
-// CHECK-LABEL: @conv2d_as_fully_connected_quant
-func @conv2d_as_fully_connected_quant(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<3x1x1x2xi8>, %arg2: tensor<3xi32>) -> tensor<4x10x10x3xi32> {
- // CHECK-NOT: "tosa.conv2d"
- // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [400, 2]}
- // CHECK-SAME: -> tensor<400x2xi8>
- // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [3, 2]}
- // CHECK-SAME: -> tensor<3x2xi8>
- // CHECK: %[[VAR2:.*]] = "tosa.fully_connected"(%[[VAR0]], %[[VAR1]], %arg2)
- // CHECK-SAME: quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}
- // CHECK-SAME: -> tensor<400x3xi32>
- // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 3]}
- // CHECK-SAME: -> tensor<4x10x10x3xi32>
- // CHECK: return %[[VAR3]]
- %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}} : (tensor<4x10x10x2xi8>, tensor<3x1x1x2xi8>, tensor<3xi32>) -> tensor<4x10x10x3xi32>
- return %0 : tensor<4x10x10x3xi32>
-}
-
-// -----
-
-// CHECK-LABEL: @conv2d_padded
-func @conv2d_padded(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<3x1x1x2xf32>, %arg2: tensor<3xf32>) -> tensor<4x12x12x3xf32> {
- // CHECK: "tosa.conv2d"
- %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [1, 1, 1, 1], stride = [1, 1], dilation = [1, 1]} : (tensor<4x10x10x2xf32>, tensor<3x1x1x2xf32>, tensor<3xf32>) -> tensor<4x12x12x3xf32>
- return %0 : tensor<4x12x12x3xf32>
-}
-
-// -----
-
// CHECK-LABEL: @conv2d_stride_2
func @conv2d_stride_2(%arg0: tensor<4x10x10x2xf32>) -> tensor<4x10x10x3xf32> {
// CHECK: "tosa.conv2d"
// -----
-// CHECK-LABEL: @depthwise_conv2d_as_mul
-func @depthwise_conv2d_as_mul(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<1x1x2x3xf32>, %arg2: tensor<6xf32>) -> tensor<4x10x10x6xf32> {
- // CHECK-NOT: "tosa.depthwise_conv2d"
- // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [4, 10, 10, 2, 1]}
- // CHECK-SAME: -> tensor<4x10x10x2x1xf32>
- // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [1, 1, 1, 2, 3]}
- // CHECK-SAME: -> tensor<1x1x1x2x3xf32>
- // CHECK: %[[VAR2:.*]] = "tosa.mul"(%[[VAR0]], %[[VAR1]])
- // CHECK-SAME: -> tensor<4x10x10x2x3xf32>
- // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 6]}
- // CHECK-SAME: -> tensor<4x10x10x6xf32>
- // CHECK: %[[VAR4:.*]] = "tosa.add"(%[[VAR3]], %arg2)
- // CHECK-SAME: -> tensor<4x10x10x6xf32>
- // CHECK: return %[[VAR4]]
- %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1]} : (tensor<4x10x10x2xf32>, tensor<1x1x2x3xf32>, tensor<6xf32>) -> tensor<4x10x10x6xf32>
- return %0 : tensor<4x10x10x6xf32>
-}
-
-// -----
-
-// CHECK-LABEL: @depthwise_conv2d_as_mul_q
-func @depthwise_conv2d_as_mul_q(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<1x1x2x3xi8>, %arg2: tensor<6xi32>) -> tensor<4x10x10x6xi32> {
- // CHECK: "tosa.depthwise_conv2d"
- %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 0 : i32, weight_zp = 0 : i32}} : (tensor<4x10x10x2xi8>, tensor<1x1x2x3xi8>, tensor<6xi32>) -> tensor<4x10x10x6xi32>
- return %0 : tensor<4x10x10x6xi32>
-}
-
-// -----
-
// CHECK-LABEL: @depthwise_conv2d_stride_2
func @depthwise_conv2d_stride_2(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<1x1x2x3xf32>, %arg2: tensor<6xf32>) -> tensor<4x10x10x6xf32> {
// CHECK: "tosa.depthwise_conv2d"
--- /dev/null
+// RUN: mlir-opt --split-input-file --tosa-optimization %s | FileCheck %s\r
+\r
+// -----\r
+\r
+// CHECK-LABEL: @conv2d_as_fully_connected\r
+func @conv2d_as_fully_connected(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<3x1x1x2xf32>, %arg2: tensor<3xf32>) -> tensor<4x10x10x3xf32> {\r
+ // CHECK-NOT: "tosa.conv2d"\r
+ // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [400, 2]}\r
+ // CHECK-SAME: -> tensor<400x2xf32>\r
+ // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [3, 2]}\r
+ // CHECK-SAME: -> tensor<3x2xf32>\r
+ // CHECK: %[[VAR2:.*]] = "tosa.fully_connected"(%[[VAR0]], %[[VAR1]], %arg2)\r
+ // CHECK-SAME: -> tensor<400x3xf32>\r
+ // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 3]}\r
+ // CHECK-SAME: -> tensor<4x10x10x3xf32>\r
+ // CHECK: return %[[VAR3]]\r
+ %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1]} : (tensor<4x10x10x2xf32>, tensor<3x1x1x2xf32>, tensor<3xf32>) -> tensor<4x10x10x3xf32>\r
+ return %0 : tensor<4x10x10x3xf32>\r
+}\r
+\r
+// -----\r
+\r
+// CHECK-LABEL: @conv2d_as_fully_connected_quant\r
+func @conv2d_as_fully_connected_quant(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<3x1x1x2xi8>, %arg2: tensor<3xi32>) -> tensor<4x10x10x3xi32> {\r
+ // CHECK-NOT: "tosa.conv2d"\r
+ // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [400, 2]}\r
+ // CHECK-SAME: -> tensor<400x2xi8>\r
+ // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [3, 2]}\r
+ // CHECK-SAME: -> tensor<3x2xi8>\r
+ // CHECK: %[[VAR2:.*]] = "tosa.fully_connected"(%[[VAR0]], %[[VAR1]], %arg2)\r
+ // CHECK-SAME: quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}\r
+ // CHECK-SAME: -> tensor<400x3xi32>\r
+ // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 3]}\r
+ // CHECK-SAME: -> tensor<4x10x10x3xi32>\r
+ // CHECK: return %[[VAR3]]\r
+ %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 42 : i32, weight_zp = 24 : i32}} : (tensor<4x10x10x2xi8>, tensor<3x1x1x2xi8>, tensor<3xi32>) -> tensor<4x10x10x3xi32>\r
+ return %0 : tensor<4x10x10x3xi32>\r
+}\r
+\r
+// -----\r
+\r
+// CHECK-LABEL: @depthwise_conv2d_as_mul\r
+func @depthwise_conv2d_as_mul(%arg0: tensor<4x10x10x2xf32>, %arg1: tensor<1x1x2x3xf32>, %arg2: tensor<6xf32>) -> tensor<4x10x10x6xf32> {\r
+ // CHECK-NOT: "tosa.depthwise_conv2d"\r
+ // CHECK: %[[VAR0:.*]] = "tosa.reshape"(%arg0) {new_shape = [4, 10, 10, 2, 1]}\r
+ // CHECK-SAME: -> tensor<4x10x10x2x1xf32>\r
+ // CHECK: %[[VAR1:.*]] = "tosa.reshape"(%arg1) {new_shape = [1, 1, 1, 2, 3]}\r
+ // CHECK-SAME: -> tensor<1x1x1x2x3xf32>\r
+ // CHECK: %[[VAR2:.*]] = "tosa.mul"(%[[VAR0]], %[[VAR1]])\r
+ // CHECK-SAME: -> tensor<4x10x10x2x3xf32>\r
+ // CHECK: %[[VAR3:.*]] = "tosa.reshape"(%[[VAR2]]) {new_shape = [4, 10, 10, 6]}\r
+ // CHECK-SAME: -> tensor<4x10x10x6xf32>\r
+ // CHECK: %[[VAR4:.*]] = "tosa.add"(%[[VAR3]], %arg2)\r
+ // CHECK-SAME: -> tensor<4x10x10x6xf32>\r
+ // CHECK: return %[[VAR4]]\r
+ %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1]} : (tensor<4x10x10x2xf32>, tensor<1x1x2x3xf32>, tensor<6xf32>) -> tensor<4x10x10x6xf32>\r
+ return %0 : tensor<4x10x10x6xf32>\r
+}\r
+\r
+// -----\r
+\r
+// CHECK-LABEL: @depthwise_conv2d_as_mul_q\r
+func @depthwise_conv2d_as_mul_q(%arg0: tensor<4x10x10x2xi8>, %arg1: tensor<1x1x2x3xi8>, %arg2: tensor<6xi32>) -> tensor<4x10x10x6xi32> {\r
+ // CHECK: "tosa.depthwise_conv2d"\r
+ %0 = "tosa.depthwise_conv2d"(%arg0, %arg1, %arg2) {pad = [0, 0, 0, 0], stride = [1, 1], dilation = [1, 1], quantization_info = {input_zp = 0 : i32, weight_zp = 0 : i32}} : (tensor<4x10x10x2xi8>, tensor<1x1x2x3xi8>, tensor<6xi32>) -> tensor<4x10x10x6xi32>\r
+ return %0 : tensor<4x10x10x6xi32>\r
+}\r
+\r
+// -----\r
projects / platform / upstream / llvm.git / commitdiff