fix big-endian build for completeness sake

ci: build aarch64 wheel (for raspberry pi os 64-bit)

.github/workflows/pip.yml

@@ -4,8 +4,6 @@ on:
   workflow_dispatch:
   pull_request:
   push:
-    branches:
-      - main
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}

.github/workflows/wheels.yml

@@ -4,8 +4,6 @@ on:
   workflow_dispatch:
   pull_request:
   push:
-    branches:
-     - main
   release:
     types:
       - published
@@ -33,19 +31,37 @@ jobs:
 
 
   build_wheels:
-    name: Wheels on ${{ matrix.os }}
+    name: Wheels on ${{ matrix.os }}${{ matrix.extra }}
     runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
       matrix:
         os: [ubuntu-latest, windows-latest, macos-latest]
-
+        arch_linux:  ["auto"]
+        extra: [""]
+        include:
+          - os: ubuntu-latest
+            arch_linux: "aarch64"
+            extra: "- aarch64"
+          - os: ubuntu-latest
+            arch_linux: "ppc64le"
+            extra: "- ppc64le"
+          - os: ubuntu-latest
+            arch_linux: "s390x"
+            extra: "- s390x"
     steps:
     - uses: actions/checkout@v3
 
+    - name: Set up QEMU
+      if: runner.os == 'Linux'
+      uses: docker/setup-qemu-action@v2
+      with:
+        platforms: all
+
     - uses: pypa/cibuildwheel@v2.14.1
       env:
-        CIBW_ARCHS_MACOS: auto universal2
+        CIBW_ARCHS_MACOS: "x86_64 universal2 arm64"
+        CIBW_ARCHS_LINUX: ${{ matrix.arch_linux }}
         CIBW_PRERELEASE_PYTHONS: true
 
     - name: Verify clean directory

src/main.cpp

@@ -1,4 +1,5 @@
 #include <pybind11/pybind11.h>
+#include <algorithm>
 
 extern "C" {
 #include "au/au_header.c"
@@ -18,6 +19,14 @@ extern "C" {
 #define STRINGIFY(x) #x
 #define MACRO_STRINGIFY(x) STRINGIFY(x)
 
+#if defined(__linux__)
+#include <endian.h>
+#else
+// all others assumed to be little-endian
+#define htole16(x) (x)
+#define le16toh(x) (x)
+#endif
+
 namespace py = pybind11;
 
 #define MAX_SAMPLE_RATE 32000
@@ -38,6 +47,9 @@ py::bytes encode(py::bytes bytes_in, size_t input_frame_size=320, size_t output_
         size_t e = std::min(std::size(data_in), i + 2*input_frame_size);
         std::fill(std::copy(reinterpret_cast<Word16*>(&data_in[i]), reinterpret_cast<Word16*>(&data_in[e]), input), std::end(input), 0);
 
+        for(size_t i=0; i < MAX_FRAMESIZE; i++)
+            input[i] = le16toh(input[i]);
+
         auto mag_shift = samples_to_rmlt_coefs(input, history, mlt_coefs, input_frame_size);
 
         /* Encode the mlt coefs */
@@ -47,6 +59,9 @@ py::bytes encode(py::bytes bytes_in, size_t input_frame_size=320, size_t output_
                 mag_shift,
                 out_words);
 
+        for(size_t i=0; i < MAX_BITS_PER_FRAME / 16; i++)
+            out_words[i] = htole16(out_words[i]);
+
         result.append(reinterpret_cast<char*>(&out_words[0]), output_frame_size);
     }