* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file histogram_fusion_pass.cpp
* \brief histogram fusion pass (Histogram --> HistogramV2)
*
* Pattern:
* x x, min, max
* | |
* Histogram ==> HistogramV2
* | |
* y y
*
* Key differences:
* - Histogram: min/max are attributes
* - HistogramV2: min/max are input tensors
* - min/max tensors dtype should match x's dtype
* - is_out_dtype_int32 attr depends on x dtype (false for float/f16)
*/
#include <vector>
#include <string>
#include "es_math_ops.h"
#include "platform/platform_info.h"
#include "ge/ge_utils.h"
#include "log/log.h"
#include "histogram_fusion_pass.h"
using namespace ge;
using namespace fe;
using namespace fusion;
namespace ops {
static const std::string kPassName = "HistogramFusionPass";
static const int64_t kCaptureHistogramNode = 0l;
std::vector<PatternUniqPtr> HistogramFusionPass::Patterns()
{
OP_LOGD(kPassName.c_str(), "Enter Patterns for HistogramFusionPass");
std::vector<PatternUniqPtr> patternGraphs;
auto graphBuilder = es::EsGraphBuilder(kPassName.c_str());
auto x = graphBuilder.CreateInput(0);
auto output = es::Histogram(x, 0.0, 0.0, 100);
auto graph = graphBuilder.BuildAndReset({output});
auto pattern = std::make_unique<Pattern>(std::move(*graph));
pattern->CaptureTensor({*output.GetProducer(), 0});
patternGraphs.emplace_back(std::move(pattern));
return patternGraphs;
}
bool HistogramFusionPass::MeetRequirements(const std::unique_ptr<MatchResult> &match_result)
{
OP_LOGD(kPassName.c_str(), "Enter MeetRequirements for HistogramFusionPass");
PlatformInfo platformInfo;
OptionalInfo optionalInfo;
if (PlatformInfoManager::Instance().GetPlatformInfoWithOutSocVersion(platformInfo, optionalInfo) != SUCCESS) {
OP_LOGE(kPassName.c_str(), "Get platformInfo failed.");
return false;
}
const std::string soc = platformInfo.str_info.short_soc_version;
if (soc != "Ascend950") {
OP_LOGD(kPassName.c_str(), "Platform %s is not supported, only Ascend950.", soc.c_str());
return false;
}
NodeIo histogramNodeIo;
if (match_result->GetCapturedTensor(kCaptureHistogramNode, histogramNodeIo) != SUCCESS) {
OP_LOGE(kPassName.c_str(), "Failed to GetCaptured tensor");
return false;
}
AscendString nodeType;
histogramNodeIo.node.GetType(nodeType);
std::string typeStr = nodeType.GetString();
if (typeStr != "Histogram") {
OP_LOGD(kPassName.c_str(), "Node type %s is not Histogram, skip.", typeStr.c_str());
return false;
}
TensorDesc inputDesc;
histogramNodeIo.node.GetInputDesc(0, inputDesc);
DataType inputDtype = inputDesc.GetDataType();
if (inputDtype != DT_FLOAT16 && inputDtype != DT_FLOAT && inputDtype != DT_INT64 && inputDtype != DT_INT32 &&
inputDtype != DT_INT16 && inputDtype != DT_INT8 && inputDtype != DT_UINT8) {
OP_LOGD(kPassName.c_str(), "Input dtype %d not supported, skip.", inputDtype);
return false;
}
TensorDesc outputDesc;
histogramNodeIo.node.GetOutputDesc(0, outputDesc);
DataType outputDtype = outputDesc.GetDataType();
if (outputDtype != DT_FLOAT && outputDtype != DT_INT32) {
OP_LOGD(kPassName.c_str(), "Output dtype %d is not DT_FLOAT or DT_INT32, skip fusion.", outputDtype);
return false;
}
return true;
}
std::unique_ptr<Graph> HistogramFusionPass::Replacement(const std::unique_ptr<MatchResult> &match_result)
{
OP_LOGD(kPassName.c_str(), "Enter Replacement for HistogramFusionPass");
std::vector<SubgraphInput> subgraphInputs;
match_result->ToSubgraphBoundary()->GetAllInputs(subgraphInputs);
std::vector<Shape> inputShapes;
std::vector<DataType> inputDtypes;
std::vector<Format> inputFormats;
GetInputsInfo(subgraphInputs, inputShapes, inputDtypes, inputFormats);
NodeIo histogramNodeIo;
if (match_result->GetCapturedTensor(kCaptureHistogramNode, histogramNodeIo) != SUCCESS) {
OP_LOGE(kPassName.c_str(), "Failed to GetCaptured tensor in Replacement");
return nullptr;
}
int64_t bins = 100;
histogramNodeIo.node.GetAttr("bins", bins);
OP_LOGD(kPassName.c_str(), "bins: %ld", bins);
float minVal = 0.0f;
histogramNodeIo.node.GetAttr("min", minVal);
OP_LOGD(kPassName.c_str(), "min: %f", minVal);
float maxVal = 0.0f;
histogramNodeIo.node.GetAttr("max", maxVal);
OP_LOGD(kPassName.c_str(), "max: %f", maxVal);
DataType inputDtype = inputDtypes[0];
DataType yDtype = inputDtype;
if (inputDtype == DT_FLOAT || inputDtype == DT_FLOAT16) {
yDtype = DT_FLOAT;
}
OP_LOGD(kPassName.c_str(), "input dtype: %d, y_dtype: %d", inputDtype, yDtype);
auto replaceGraphBuilder = es::EsGraphBuilder("replacement");
std::vector<int64_t> xDims;
for (size_t i = 0; i < inputShapes[0].GetDimNum(); i++) {
xDims.push_back(inputShapes[0].GetDim(i));
}
auto rX = replaceGraphBuilder.CreateInput(0, "x", inputDtypes[0], inputFormats[0], xDims);
auto rMin = replaceGraphBuilder.CreateScalar(minVal);
auto rMax = replaceGraphBuilder.CreateScalar(maxVal);
auto output = es::HistogramV2(rX, rMin, rMax, bins, yDtype);
std::vector<es::EsTensorHolder> outputs = {output};
GraphUniqPtr replaceGraph = replaceGraphBuilder.BuildAndReset(outputs);
if (InferShape(replaceGraph, subgraphInputs) != SUCCESS) {
OP_LOGE(kPassName.c_str(), "Infershape for replacement failed.");
return nullptr;
}
return replaceGraph;
}
static void GetInputsInfo(const std::vector<SubgraphInput> &subgraphInputs, std::vector<Shape> &inputShapes,
std::vector<DataType> &inputDtypes, std::vector<Format> &inputFormats)
{
for (const auto &subgraphInput : subgraphInputs) {
auto matchNode = subgraphInput.GetAllInputs().at(0);
TensorDesc tensorDesc;
matchNode.node.GetInputDesc(matchNode.index, tensorDesc);
inputShapes.emplace_back(tensorDesc.GetShape());
inputDtypes.emplace_back(tensorDesc.GetDataType());
inputFormats.emplace_back(tensorDesc.GetFormat());
}
}
static Status InferShape(const GraphUniqPtr &replaceGraph, const std::vector<SubgraphInput> &subgraphInputs)
{
OP_LOGD(kPassName.c_str(), "Begin infershape for replacements.");
std::vector<Shape> inputShapes;
for (const auto &subgraphInput : subgraphInputs) {
auto matchNode = subgraphInput.GetAllInputs().at(0);
TensorDesc tensorDesc;
matchNode.node.GetInputDesc(matchNode.index, tensorDesc);
inputShapes.emplace_back(tensorDesc.GetShape());
}
return GeUtils::InferShape(*replaceGraph, inputShapes);
}
REG_FUSION_PASS(HistogramFusionPass).Stage(CustomPassStage::kAfterInferShape);
}
