nnicolosi/Raptor
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

replace deprecated "rewriter.create()" pattern

Browse Source 
refactor PIM to Pim everywhere except for the accelerator name
This commit is contained in:
NiccoloN
2026-03-20 13:30:53 +01:00
parent 916a09414c
commit bb6dcd38a3
32 changed files with 222 additions and 212 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
src/PIM/Conversion/ONNXToSpatial/ONNXToSpatialCommon.cpp
View File

@@ -47,7 +47,7 @@ SmallVector<Value> sliceTensor(
if (i == numSlices - 1 && lastSliceSize != 0)
sizes[axis] = rewriter.getIndexAttr(lastSliceSize);
Value slice = rewriter.create<tensor::ExtractSliceOp>(loc, tensorToSlice, offsets, sizes, strides);
Value slice = tensor::ExtractSliceOp::create(rewriter, loc, tensorToSlice, offsets, sizes, strides);
slices.push_back(slice);
}
@@ -100,11 +100,11 @@ broadcastToVector(Value scalarToBroadcast, int64_t length, ConversionPatternRewr
int64_t shape[2] = {1, length};
Type type = oldType.cloneWith(ArrayRef(shape), elementType);
auto zero = rewriter.create<arith::ConstantIndexOp>(loc, 0).getResult();
auto zero = arith::ConstantIndexOp::create(rewriter, loc, 0).getResult();
SmallVector<Value> index(oldType.getRank(), zero);
auto elementValue = rewriter.create<tensor::ExtractOp>(loc, scalarToBroadcast, index).getResult();
auto elementValue = tensor::ExtractOp::create(rewriter, loc, scalarToBroadcast, index).getResult();
return rewriter.create<tensor::SplatOp>(loc, type, elementValue);
return tensor::SplatOp::create(rewriter, loc, type, elementValue);
}
Value sumTensors(ArrayRef<Value> tensors, ConversionPatternRewriter& rewriter) {
@@ -122,7 +122,7 @@ Value sumTensors(ArrayRef<Value> tensors, ConversionPatternRewriter& rewriter) {
Value a = (*currTensors)[i];
Value b = (*currTensors)[i + 1];
rewriter.setInsertionPointAfterValue(b);
auto addedValue = rewriter.create<spatial::SpatVAddOp>(a.getLoc(), a.getType(), a, b);
auto addedValue = spatial::SpatVAddOp::create(rewriter, a.getLoc(), a.getType(), a, b);
nextTensors->push_back(addedValue);
}
if (currTensors->size() % 2 == 1)
@@ -137,10 +137,10 @@ Value sumTensors(ArrayRef<Value> tensors, ConversionPatternRewriter& rewriter) {
Value createMapOperation(PatternRewriter& rewriter, MapOperations mapOp, const Value& input) {
switch (mapOp) {
case MapOperations::None: assert(false && "Invalid map operation during map operation creation.");
case MapOperations::ONNXSoftmaxOp: return rewriter.create<ONNXSoftmaxOp>(input.getLoc(), input.getType(), input);
case MapOperations::ONNXReluOp: return rewriter.create<ONNXReluOp>(input.getLoc(), input.getType(), input);
case MapOperations::ONNXLeakyReluOp: return rewriter.create<ONNXLeakyReluOp>(input.getLoc(), input.getType(), input);
case MapOperations::ONNXExpOp: return rewriter.create<ONNXExpOp>(input.getLoc(), input.getType(), input);
case MapOperations::ONNXSoftmaxOp: return ONNXSoftmaxOp::create(rewriter, input.getLoc(), input.getType(), input);
case MapOperations::ONNXReluOp: return ONNXReluOp::create(rewriter, input.getLoc(), input.getType(), input);
case MapOperations::ONNXLeakyReluOp: return ONNXLeakyReluOp::create(rewriter, input.getLoc(), input.getType(), input);
case MapOperations::ONNXExpOp: return ONNXExpOp::create(rewriter, input.getLoc(), input.getType(), input);
}
}
@@ -201,7 +201,7 @@ void tileImageTensorByChannel(Value imageTensor,
offsets[2] = rewriter.getIndexAttr(x);
offsets[3] = rewriter.getIndexAttr(y);
tiles[i][x][y] = rewriter.create<tensor::ExtractSliceOp>(loc, imageTensor, offsets, sizes, strides);
tiles[i][x][y] = tensor::ExtractSliceOp::create(rewriter, loc, imageTensor, offsets, sizes, strides);
}
}
}
@@ -225,7 +225,7 @@ Value createImgConcatOp(SmallVector<SmallVector<SmallVector<Value>>>& outputTile
for (size_t outTile = 0; outTile < outputTileCount; outTile++)
tilesToConcat.push_back(outputTiles[outTile][outX][outY]);
return rewriter.create<spatial::SpatImgConcatOp>(loc, outputType, tilesToConcat);
return spatial::SpatImgConcatOp::create(rewriter, loc, outputType, tilesToConcat);
}
LogicalResult
@@ -271,7 +271,7 @@ Value createExtractSliceImg(Value valToSlice,
offsets[2] = rewriter.getIndexAttr(x);
offsets[3] = rewriter.getIndexAttr(y);
return rewriter.create<tensor::ExtractSliceOp>(valToSlice.getLoc(), valToSlice, offsets, sizes, strides);
return tensor::ExtractSliceOp::create(rewriter, valToSlice.getLoc(), valToSlice, offsets, sizes, strides);
}
Value indexImgValue(Value v,
@@ -384,7 +384,7 @@ void resolveInputTensorTilesBlockArg(Value wholeInputTensor,
offsets[2] = rewriter.getIndexAttr(x);
offsets[3] = rewriter.getIndexAttr(y);
inputTiles[t][x][y] = rewriter.create<tensor::ExtractSliceOp>(loc, wholeInputTensor, offsets, sizes, strides);
inputTiles[t][x][y] = tensor::ExtractSliceOp::create(rewriter, loc, wholeInputTensor, offsets, sizes, strides);
}
}
}
@@ -452,7 +452,7 @@ LogicalResult handleFlattenLikeOp(SmallVector<SmallVector<Value>>& inputTiles,
SmallVector<int64_t> newShapeVals = {curTileShape.getDimSize(0), curTileShape.getDimSize(1)};
auto shapeType = RankedTensorType::get({static_cast<int64_t>(newShapeVals.size())}, rewriter.getI64Type());
Value shapeTensor =
rewriter.create<arith::ConstantOp>(reshapeInput.getLoc(), DenseIntElementsAttr::get(shapeType, newShapeVals));
arith::ConstantOp::create(rewriter, reshapeInput.getLoc(), DenseIntElementsAttr::get(shapeType, newShapeVals));
auto reshapedType = RankedTensorType::get(newShapeVals, curTileShape.getElementType());
auto reshapedCurTile = tosa::ReshapeOp::create(rewriter, reshapeInput.getLoc(), reshapedType, curTile, shapeTensor);