Merge pull request #660 from tcal-x/hps-ci

CI test and fix for #658 (HPS software build).

.github/workflows/hps.yml

@@ -0,0 +1,22 @@
+name: hps_ci
+on:
+  pull_request:
+    paths-ignore:
+      - 'docs/**'
+      - 'README.md'
+      - 'LICENSE'
+      - 'CONTRIBUTING.md'
+      - 'CODE_OF_CONDUCT.md'
+      - 'AUTHORS'
+jobs:
+  build-hps:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - uses: actions/setup-python@v2
+        with:
+          python-version: '3.7'
+      - run: bash scripts/setup -ci
+      - run: which pip3 && which python3 && which pip
+      - run: make env
+      - run: pwd && source env/conda/bin/activate cfu-common && source environment && cd proj/hps_accel && pip3 list && make PLATFORM=hps software

common/_hps/hps/ld/linker.ld

@@ -24,7 +24,7 @@ SECTIONS
 		/*
 		 * Align models and large chunks of data into a constant position
 		 */
-		. = _ftext + 0x84000;
+		. = _ftext + 0x86000;
 		_frodata = .;
 		src/models/*/*(SORT(.rodata.*))
 		src/conv2d_??.o(SORT(.rodata.*))  /* specific to hps_accel */

proj/hps_accel/src/tensorflow/lite/kernels/internal/reference/integer_ops/conv.h

@@ -236,12 +236,13 @@ inline void ConvPerChannel(
 
 // Fixed-point per-channel-quantization convolution reference kernel.
 // 16-bit data and 8-bit filter
+template <typename AccumScalar>
 inline void ConvPerChannel(
     const ConvParams& params, const int32_t* output_multiplier,
     const int32_t* output_shift, const RuntimeShape& input_shape,
     const int16_t* input_data, const RuntimeShape& filter_shape,
     const int8_t* filter_data, const RuntimeShape& bias_shape,
-    const std::int64_t* bias_data, const RuntimeShape& output_shape,
+    const AccumScalar* bias_data, const RuntimeShape& output_shape,
     int16_t* output_data) {
   // Get parameters.
   const int stride_width = params.stride_width;