Skip to content

[mlir][tosa] Fix conv op build functions #126321

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Feb 10, 2025
Merged

[mlir][tosa] Fix conv op build functions #126321

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 17 additions & 20 deletions mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -510,7 +510,13 @@ static void buildConvOpWithQuantInfo(OpBuilder &builder, OperationState &result,
result.addAttribute("stride", stride);
result.addAttribute("dilation", dilation);
result.addAttribute("acc_type", accType);
result.addTypes(outputType);
Type finalOutputType = outputType;
auto quantAttr = buildConvOpQuantizationAttr(builder, input, weight);
if (quantAttr) {
finalOutputType =
buildConvOpResultTypeInfo(builder, outputType, input, weight);
}
result.addTypes(finalOutputType);
}

/// Handles tosa.transpose_conv2d which has outpad and output shape
Expand All @@ -519,25 +525,19 @@ static void buildTransConvOpWithQuantInfo(
OpBuilder &builder, OperationState &result, Type outputType, Value input,
Value weight, Value bias, DenseI64ArrayAttr outpad,
DenseI64ArrayAttr stride, DenseI64ArrayAttr outputShape, TypeAttr accType) {
result.addOperands({input, weight, bias});
auto zps = createZPsAsConst(builder, input, weight);
result.addOperands({input, weight, bias, zps.first, zps.second});
result.addAttribute("out_pad", outpad);
result.addAttribute("stride", stride);
result.addAttribute("out_shape", outputShape);
result.addAttribute("acc_type", accType);
auto quantAttr = ::buildConvOpQuantizationAttr(builder, input, weight);

Type finalOutputType = outputType;
auto quantAttr = buildConvOpQuantizationAttr(builder, input, weight);
if (quantAttr) {
result.addAttribute("input_zp",
builder.getI32IntegerAttr(
static_cast<int32_t>(quantAttr.getInputZp())));
result.addAttribute("weight_zp",
builder.getI32IntegerAttr(
static_cast<int32_t>(quantAttr.getWeightZp())));
result.addTypes(
buildConvOpResultTypeInfo(builder, outputType, input, weight));
} else {
result.addTypes(outputType);
finalOutputType =
buildConvOpResultTypeInfo(builder, outputType, input, weight);
}
result.addTypes(finalOutputType);
}

/// The tosa.fully_connected op has its own builder as it does not have
Expand Down Expand Up @@ -2492,18 +2492,15 @@ LogicalResult mlir::tosa::getZeroPoint(ElementsAttr zpAttr, int64_t &zp) {
return failure();
}

// Create a rank-0 const tensor for zero point of the source tensor.
// Create a rank-1 const tensor for zero point of the source tensor.
std::optional<Value> mlir::tosa::createZeroPointTensor(OpBuilder &builder,
Location loc,
Type srcElemType,
int64_t zp) {
if (auto quantType =
llvm::dyn_cast<mlir::quant::UniformQuantizedType>(srcElemType))
srcElemType = quantType.getStorageType();

auto zpType = mlir::RankedTensorType::get({1}, srcElemType);
srcElemType = getElementTypeOrSelf(srcElemType);
if (auto quantType = llvm::dyn_cast<mlir::quant::QuantizedType>(srcElemType))
srcElemType = quantType.getStorageType();
auto zpType = mlir::RankedTensorType::get({1}, srcElemType);
if (llvm::isa<FloatType>(srcElemType)) {
auto zpAttr = DenseElementsAttr::get(
zpType, builder.getFloatAttr(srcElemType, static_cast<double>(zp)));
Expand Down