Skip to content

QUICKSTART.md

Latest commit

 

History

History
196 lines (148 loc) · 4.61 KB

QUICKSTART.md

File metadata and controls

196 lines (148 loc) · 4.61 KB

Quick Start Guide

Prerequisites

  • Thingino firmware on Ingenic T41/T31 device
  • Cross-compilation toolchain (mipsel-linux-musl)
  • SSH access to device

Build

cd thingino-accel
./build.sh

This will create:

  • build/lib/libnna.so - Shared library
  • build/lib/libnna.a - Static library
  • build/bin/test_init - Test program

Deploy to Device

# Replace with your device IP
DEVICE_IP=192.168.x.x

# Copy library and test
scp build/lib/libnna.so root@$DEVICE_IP:/usr/lib/
scp build/bin/test_init root@$DEVICE_IP:/tmp/

Run Test

ssh root@$DEVICE_IP

# Load NNA kernel module (if not already loaded)
insmod /lib/modules/soc-nna.ko

# Run test
/tmp/test_init

Expected output:

╔══════════════════════════════════════════════════════════╗
║         thingino-accel - Initialization Test            ║
╚══════════════════════════════════════════════════════════╝

[1/6] Library version
      Version: 0.1.0-dev
  ✓ Version retrieved

[2/6] NNA initialization
NNA initialized: ORAM @ 0x12620000 (384 KB)
  ✓ NNA initialized

[3/6] Hardware information
      ORAM Physical: 0x12620000
      ORAM Virtual:  0xb6f00000
      ORAM Size:     384 KB
      NNA Version:   0x20
  ✓ Hardware info retrieved

[4/6] DDR memory allocation
      Allocated: 1048576 bytes @ 0x77f00000
  ✓ DDR memory is writable
  ✓ DDR memory freed

[5/6] ORAM allocation
      Allocated: 4096 bytes @ 0xb6f00000
      ORAM usage: 4 / 384 KB (1.0%)
  ✓ ORAM allocation successful

[6/6] Tensor operations
      Shape: [1, 224, 224, 3]
      Elements: 150528
      Bytes: 150528
  ✓ Tensor created
  ✓ Tensor data written
  ✓ Tensor destroyed

[CLEANUP] Shutting down
  ✓ NNA deinitialized

╔══════════════════════════════════════════════════════════╗
║              ✓ ALL TESTS PASSED!                        ║
╚══════════════════════════════════════════════════════════╝

Using the Library

Basic Example

#include <nna.h>
#include <nna_memory.h>
#include <nna_tensor.h>

int main() {
    // Initialize NNA
    if (nna_init() != NNA_SUCCESS) {
        return 1;
    }
    
    // Create input tensor (1x224x224x3 RGB image)
    nna_shape_t shape = nna_shape_make(1, 224, 224, 3);
    nna_tensor_t *input = nna_tensor_create(
        &shape, 
        NNA_DTYPE_UINT8, 
        NNA_FORMAT_NHWC
    );
    
    // Fill with image data
    unsigned char *data = nna_tensor_data(input);
    // ... load image into data ...
    
    // TODO: Load model and run inference
    
    // Cleanup
    nna_tensor_destroy(input);
    nna_deinit();
    
    return 0;
}

Compile Your Program

mipsel-linux-gcc -o myapp myapp.c \
    -I/path/to/thingino-accel/include \
    -L/path/to/thingino-accel/build/lib \
    -lnna -lpthread

Troubleshooting

"Failed to open /dev/soc-nna"

Make sure the kernel module is loaded:

lsmod | grep soc-nna
# If not loaded:
insmod /lib/modules/soc-nna.ko

"Out of ORAM memory"

ORAM is limited (384 KB on T41). Use DDR for large allocations:

void *large_buffer = nna_malloc(1024 * 1024);  // DDR
void *small_buffer = nna_oram_malloc(4096);    // ORAM

Build Errors

Make sure toolchain is in PATH:

export PATH="/path/to/thingino/toolchain/bin:$PATH"
export CROSS_COMPILE=mipsel-linux-

Next Steps

  1. Explore the API - See include/ directory
  2. Read DEVELOPMENT.md - Learn about internals
  3. Check REVERSE_ENGINEERING.md - Understand OEM library
  4. Contribute - Help implement model loading and inference!

API Reference

Core Functions (nna.h)

  • nna_init() - Initialize NNA
  • nna_deinit() - Cleanup
  • nna_get_hw_info() - Get hardware info
  • nna_is_ready() - Check if initialized

Memory (nna_memory.h)

  • nna_malloc() - Allocate DDR memory
  • nna_free() - Free memory
  • nna_oram_malloc() - Allocate ORAM
  • nna_oram_free() - Free ORAM

Tensors (nna_tensor.h)

  • nna_tensor_create() - Create tensor
  • nna_tensor_destroy() - Destroy tensor
  • nna_tensor_data() - Get data pointer
  • nna_shape_make() - Create shape

Support

  • GitHub Issues: (TBD)
  • Documentation: See *.md files in repo
  • Examples: See examples/ directory