Skip to content

NPU Code Deployment*

1 Overview*

After generating the model file using the compiler, you can deploy and test the model on the SDK. This document is based on the SDK code example dsp/vpa/sed/vpa_algorithm.c

  • The example model implements cry detection, with build configuration configs/sed/sed_demo_8008_WUKONG_V1.0.config
  • 8008c does not have NPU, not supported

2 NPU API Call Flow*

2.1 NPU Task Configuration*

int SedInitSnpuTask(SNPU_TASK *snpu_task)
{
    snpu_task->module_id = 886;
    snpu_task->ops       = (void *)ops_content;
    snpu_task->data      = (void *)cpu_content;
    snpu_task->cmd       = (void *)npu_content;
    snpu_task->tmp_mem   = (void *)tmp_content;
    return 0;
}

    SedInitSnpuTask(&s_snpu_task);

    s_snpu_task.input  = SedGetSnpuFeatsBuffer(prev_context->snpu_buffer);
    s_snpu_task.output = SedGetSnpuStateBuffer(curr_context->snpu_buffer);

2.2 NPU Execution*

SnpuRunTask(&s_snpu_task, (void *)curr_context);

2.3 NPU Processing Callback*

After NPU execution is completed, the callback function VpaSedModelCallback will be triggered, where you can get the processing results:

int VpaSedModelCallback(unsigned int module_id, unsigned int state, void *private_data)
{
    VSP_CONTEXT *context = (VSP_CONTEXT *)private_data;

    // Get output results
    unsigned short *proOut = SedGetSnpuOutBuffer(context->snpu_buffer);
    unsigned int proOut_int = VspFloat16toFloat32(proOut[0]);
    g_prob_out = *((float*)&proOut_int);

    return 0;
}

3 Notes*

3.1 Data Management*

// Invalidate data, reload from memory
xthal_dcache_region_invalidate(buffer, size);

// Write back cache to memory
xthal_dcache_region_writeback(buffer, size);

3.2 Data Conversion*

NPU input and output only support float16 format

VspFloat32ToFloat16()  // float32 to float16 conversion
VspFloat16toFloat32()  // float16 to float32 conversion