ysh329 / OpenCL-101

Learn OpenCL step by step.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

【竞品调研】TensorFlow Lite GPU OpenCL在卷积上的选择策略

ysh329 opened this issue · comments

// tensorflow/tensorflow/lite/delegates/gpu/cl/selectors/convolution_selector.cc
// https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/delegates/gpu/cl/selectors/convolution_selector.cc#L40
std::unique_ptr<GPUOperation> SelectConvolution(
    const Convolution2DAttributes& attr, const BHWC& dst_shape,
    const DeviceInfo& device_info, const OperationDef& op_def,
    ModelHints hints) {
  if (device_info.IsAdreno()) {
    return SelectConvolutionAdreno(attr, dst_shape, device_info, op_def, hints);
  } else if (device_info.IsPowerVR() || device_info.IsAMD() ||
             device_info.IsIntel()) {
    return SelectConvolutionPowerVR(attr, device_info, op_def);
  } else if (device_info.IsNvidia()) {
    return SelectConvolutionNVidia(attr, dst_shape, device_info, op_def);
  } else if (device_info.IsMali()) {
    return SelectConvolutionMali(attr, dst_shape, device_info, op_def);
  } else {
    return SelectConvolutionAdreno(attr, dst_shape, device_info, op_def, hints);
  }
}

std::unique_ptr<GPUOperation> SelectConvolutionForWinograd(
    const Convolution2DAttributes& attr, const BHWC& dst_shape,
    const DeviceInfo& device_info, const OperationDef& op_def,
    ModelHints hints) {
  if (device_info.IsAdreno()) {
    return SelectConvolutionWinogradAdreno(attr, dst_shape, device_info, op_def,
                                           hints);
  } else if (device_info.IsPowerVR() || device_info.IsAMD() ||
             device_info.IsNvidia() || device_info.IsIntel()) {
    ConvPowerVR conv =
        CreateConvPowerVRWino4x4To6x6(device_info, op_def, attr, &dst_shape);
    return absl::make_unique<ConvPowerVR>(std::move(conv));
  } else if (device_info.IsMali()) {
    return SelectConvolutionWinogradMali(attr, dst_shape, device_info, op_def);
  } else {
    return SelectConvolutionWinogradAdreno(attr, dst_shape, device_info, op_def,
                                           hints);
  }
}

std::unique_ptr<GPUOperation> SelectConvolutionWithDynamicWeights(
    const Convolution2DAttributes& attr, const BHWC& weights_shape,
    const BHWC& dst_shape, const DeviceInfo& device_info,
    const OperationDef& op_def, ModelHints hints,
    ConvWeightsDescription* weights_desc) {
  if (device_info.IsAdreno()) {
    return SelectConvolutionDynamicWeightsAdreno(attr, weights_shape, dst_shape,
                                                 device_info, op_def, hints,
                                                 weights_desc);
  } else if (device_info.IsMali()) {
    return SelectConvolutionDynamicWeightsMali(attr, weights_shape, dst_shape,
                                               device_info, op_def, hints,
                                               weights_desc);
  } else {
    ConvPowerVR conv = CreateConvPowerVRDynamicWeights(
        device_info, op_def, attr, weights_shape, &dst_shape);
    *weights_desc = conv.GetConvWeightsDescription();
    return absl::make_unique<ConvPowerVR>(std::move(conv));
  }
}