• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===- UniformSupport.cpp - Support utilities for uniform quant -----------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "mlir/Dialect/Quant/UniformSupport.h"
10 #include "mlir/IR/BuiltinTypes.h"
11 #include <numeric>
12 
13 using namespace mlir;
14 using namespace mlir::quant;
15 
isQuantizablePrimitiveType(Type inputType)16 static bool isQuantizablePrimitiveType(Type inputType) {
17   return inputType.isa<FloatType>();
18 }
19 
20 const ExpressedToQuantizedConverter
forInputType(Type inputType)21 ExpressedToQuantizedConverter::forInputType(Type inputType) {
22   if (inputType.isa<TensorType, VectorType>()) {
23     Type elementType = inputType.cast<ShapedType>().getElementType();
24     if (!isQuantizablePrimitiveType(elementType))
25       return ExpressedToQuantizedConverter{inputType, nullptr};
26     return ExpressedToQuantizedConverter{inputType, elementType};
27   }
28   // Supported primitive type (which just is the expressed type).
29   if (isQuantizablePrimitiveType(inputType))
30     return ExpressedToQuantizedConverter{inputType, inputType};
31   // Unsupported.
32   return ExpressedToQuantizedConverter{inputType, nullptr};
33 }
34 
convert(QuantizedType elementalType) const35 Type ExpressedToQuantizedConverter::convert(QuantizedType elementalType) const {
36   assert(expressedType && "convert() on unsupported conversion");
37   if (auto tensorType = inputType.dyn_cast<RankedTensorType>())
38     return RankedTensorType::get(tensorType.getShape(), elementalType);
39   if (auto tensorType = inputType.dyn_cast<UnrankedTensorType>())
40     return UnrankedTensorType::get(elementalType);
41   if (auto vectorType = inputType.dyn_cast<VectorType>())
42     return VectorType::get(vectorType.getShape(), elementalType);
43 
44   // If the expressed types match, just use the new elemental type.
45   if (elementalType.getExpressedType() == expressedType)
46     return elementalType;
47   // Unsupported.
48   return nullptr;
49 }
50 
51 ElementsAttr
convert(Attribute realValue)52 UniformQuantizedPerAxisValueConverter::convert(Attribute realValue) {
53   if (auto attr = realValue.dyn_cast<DenseFPElementsAttr>()) {
54     return convert(attr);
55   }
56   // TODO: handles sparse elements attribute
57   return nullptr;
58 }
59 
60 DenseElementsAttr
convert(DenseFPElementsAttr attr)61 UniformQuantizedPerAxisValueConverter::convert(DenseFPElementsAttr attr) {
62   // Creates the converter for each chunk. Normally the size of the
63   // quantization dim is 3, so we can cache all the converters.
64   ShapedType type = attr.getType();
65   size_t dimSize = type.getDimSize(quantizationDim);
66   if (dimSize != scales.size()) {
67     return {};
68   }
69   SmallVector<UniformQuantizedValueConverter, 4> converters;
70   converters.reserve(dimSize);
71   for (int i = 0, e = dimSize; i != e; ++i) {
72     converters.push_back(getPerChunkConverter(i));
73   }
74 
75   // Scan the elements of the dense elements attributes and quantize them by
76   // using the right quantization parameters.
77   int64_t flattenIndex = 0;
78   auto shape = type.getShape();
79   int64_t chunkSize =
80       std::accumulate(std::next(shape.begin(), quantizationDim + 1),
81                       shape.end(), 1, std::multiplies<int64_t>());
82   Type newElementType = IntegerType::get(storageBitWidth, attr.getContext());
83   return attr.mapValues(newElementType, [&](const APFloat &old) {
84     int chunkIndex = (flattenIndex++) / chunkSize;
85     return converters[chunkIndex % dimSize].quantizeFloatToInt(old);
86   });
87 }
88