Function LogConvAutotuneResultsImpl

tensorflow/core/kernels/gpu_utils.cc:134–173

template<typename TResult, typename TLog>
void LogConvAutotuneResultsImpl(se::dnn::ConvolutionKind kind,
    se::dnn::DataType element_type, se::DeviceMemoryBase input_buffer,
    se::DeviceMemoryBase filter_buffer, se::DeviceMemoryBase output_buffer,
    const se::dnn::BatchDescriptor& input_desc,
    const se::dnn::FilterDescriptor& filter_desc,
    const se::dnn::BatchDescriptor& output_desc,
    const se::dnn::ConvolutionDescriptor& conv_desc,
    se::StreamExecutor* stream_exec,
    absl::Span<const TResult> results) {
  TLog log;
  {
    ConvolutionProto instr;
    instr.set_kind(kind);
    *instr.mutable_input() = input_desc.ToProto(element_type);
    *instr.mutable_filter() = filter_desc.ToProto(element_type);
    *instr.mutable_output() = output_desc.ToProto(element_type);
    *instr.mutable_conv_desc() = conv_desc.ToProto();
    instr.set_conv_scale(1);
    instr.set_side_value_scale(0);
    instr.set_input_address(reinterpret_cast<uint64>(input_buffer.opaque()));
    instr.set_filter_address(reinterpret_cast<uint64>(filter_buffer.opaque()));
    instr.set_output_address(reinterpret_cast<uint64>(output_buffer.opaque()));
    log.mutable_instr()->PackFrom(std::move(instr));
  }
  *log.mutable_cudnn_version() = GetCudnnVersion(stream_exec);
  *log.mutable_compute_capability() = GetComputeCapability(stream_exec);
  log.set_device_pci_bus_id(stream_exec->GetDeviceDescription().pci_bus_id());
  {
    string blas_version;
    if (auto* blas = stream_exec->AsBlas()) {
      if (blas->GetVersion(&blas_version).ok()) {
        log.set_blas_version(blas_version);
      }
    }
  }
  for (const auto& result : results) {
    *log.add_results() = result;
  }
  Logger::GetSingleton()->LogProto(log);
}

template<typename TResult, typename TLog>
void LogFusedConvForwardAutotuneResultsImpl(