Add armv6 and armv7 to multiarch CI and fix build issues

.github/workflows/build-debian-multiarch.yml

@@ -35,26 +35,45 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-debian-multiarch
   cancel-in-progress: true
 
+# this command is called in two places, so save it in an env first
+env:
+  INSTALL_CMD: |
+    apt-get update --fix-missing
+    apt-get upgrade -y
+    apt-get install build-essential meson -y
+    apt-get install libsdl2-dev libsdl2-image-dev libsdl2-mixer-dev libsdl2-ttf-dev -y
+    apt-get install libfreetype6-dev libportmidi-dev fontconfig -y
+    apt-get install python3-dev python3-pip python3-wheel python3-sphinx -y
+    pip3 install meson-python --break-system-packages
+
 jobs:
   build-multiarch:
-    name: Debian (Bullseye - 11) [${{ matrix.arch }}]
+    name: Debian (Bookworm - 12) [${{ matrix.arch }}]
     runs-on: ubuntu-22.04
 
     strategy:
       fail-fast: false  # if a particular matrix build fails, don't skip the rest
       matrix:
         # maybe more things could be added in here in the future (if needed)
-        arch: [s390x, ppc64le]
+        include:
+          - { arch: s390x, base_image: '' }
+          - { arch: ppc64le, base_image: '' }
+          - { arch: armv6, base_image: '' }
+          # a custom base_image is specified in the armv7 case. This is done because
+          # the armv6 image is just raspbian in disguise. And the wheel built on armv7
+          # is going to be tested on armv6
+          - { arch: armv7, base_image: 'balenalib/raspberrypi3-debian:bookworm' }
 
     steps:
     - uses: actions/[email protected]
 
     - name: Build sources and run tests
-      uses: uraimo/[email protected].1
+      uses: uraimo/[email protected].2
       id: build
       with:
-        arch: ${{ matrix.arch }}
-        distro: bullseye
+        arch: ${{ matrix.base_image && 'none' || matrix.arch }}
+        distro: ${{ matrix.base_image && 'none' || 'bookworm' }}
+        base_image: ${{ matrix.base_image }}
 
         # Not required, but speeds up builds
         githubToken: ${{ github.token }}
@@ -74,19 +93,61 @@ jobs:
         # builds don't have to re-install them. The image layer is cached
         # publicly in your project's package repository, so it is vital that
         # no secrets are present in the container state or logs.
-        install: |
-          apt-get update --fix-missing
-          apt-get upgrade -y
-          apt-get install libsdl2-dev libsdl2-image-dev libsdl2-mixer-dev libsdl2-ttf-dev libfreetype6-dev libportmidi-dev libjpeg-dev fontconfig -y
-          apt-get install python3-setuptools python3-dev python3-pip python3-wheel python3-sphinx -y
+        install: ${{ env.INSTALL_CMD }}
 
-        # Build a wheel, install it for running unit tests
+        # Build a wheel, install it for running unit tests.
+        # --no-build-isolation is passed so that preinstalled meson-python can be used
+        # (done for optimization reasons)
         run: |
           echo "\nBuilding pygame wheel\n"
           python3 setup.py docs
-          pip3 wheel . --wheel-dir /artifacts -vvv
+          pip3 wheel . --no-build-isolation --wheel-dir /artifacts -vvv
+          echo "\nInstalling wheel\n"
+          pip3 install --no-index --pre --break-system-packages --find-links /artifacts pygame-ce
+          echo "\nRunning tests\n"
+          export SDL_VIDEODRIVER=dummy
+          export SDL_AUDIODRIVER=disk
+          python3 -m pygame.tests -v --exclude opengl,music,timing --time_out 300
+    
+    # Upload the generated files under github actions assets section
+    - name: Upload dist
+      uses: actions/upload-artifact@v4
+      with:
+        name: pygame-multiarch-${{ matrix.arch }}-dist
+        path: ~/artifacts/*.whl
+
+  # test wheels built on armv7 on armv6. Why?
+  # because piwheels expects the same armv7 wheel to work on both armv7 and armv6
+  test-armv7-on-armv6:
+    needs: build-multiarch
+    name: Debian (Bookworm - 12) [build - armv7, test - armv6]
+    runs-on: ubuntu-22.04
+    steps:
+    - name: Download all multiarch artifacts
+      uses: actions/download-artifact@v4
+      with:
+        name: pygame-multiarch-armv7-dist
+        path: ~/artifacts
+
+    - name: Rename arm wheel in artifacts
+      run: |
+        cd ~/artifacts
+        for f in *; do
+          mv "$f" "${f//armv7l/armv6l}"
+        done
+
+    - name: Test armv7 wheel on armv6
+      uses: uraimo/[email protected]
+      with:
+        arch: armv6
+        distro: bookworm
+        githubToken: ${{ github.token }}
+        dockerRunArgs: --volume ~/artifacts:/artifacts_new
+        shell: /bin/sh
+        install: ${{ env.INSTALL_CMD }}
+        run: |
           echo "\nInstalling wheel\n"
-          pip3 install --no-index --pre --find-links /artifacts pygame-ce
+          pip3 install --no-index --pre --break-system-packages --find-links /artifacts_new pygame-ce
           echo "\nRunning tests\n"
           export SDL_VIDEODRIVER=dummy
           export SDL_AUDIODRIVER=disk

meson.build

@@ -293,8 +293,8 @@ if host_machine.cpu_family() == 'arm'
     # first check if compiler supports the flag, and use it then. Needed only
     # on 32-bit armv7.
     flag = '-mfpu=neon'
-    if cc.has_argument(flag)
-        simd_sse2_neon_flags += flag
+    if cc.has_argument(flag) and host_machine.cpu() == 'armv7l'
+        simd_sse2_neon_flags += [flag, '-march=armv7-a']
         simd_sse2_neon = true
     endif
 elif host_machine.cpu_family() == 'aarch64'

src_c/simd_fill.h

@@ -21,11 +21,6 @@
 #endif
 #endif
 
-#if PG_ENABLE_ARM_NEON
-// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
-#include "include/sse2neon.h"
-#endif /* PG_ENABLE_ARM_NEON */
-
 #if defined(__SSE2__)
 #define PG_ENABLE_SSE_NEON 1
 #elif PG_ENABLE_ARM_NEON

src_c/simd_shared.h

@@ -33,11 +33,6 @@ pg_has_avx2();
 #endif
 #endif
 
-#if PG_ENABLE_ARM_NEON
-// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
-#include "include/sse2neon.h"
-#endif /* PG_ENABLE_ARM_NEON */
-
 /* This defines PG_ENABLE_SSE_NEON as True if either SSE or NEON is available
  * at compile time. Since we do compile time translation of SSE2->NEON, they
  * have the same code paths, so this reduces code duplication of those paths.

src_c/simd_surface_fill_sse2.c

@@ -1,5 +1,10 @@
 #include "simd_fill.h"
 
+#if PG_ENABLE_ARM_NEON
+// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
+#include "include/sse2neon.h"
+#endif /* PG_ENABLE_ARM_NEON */
+
 #define BAD_SSE2_FUNCTION_CALL                                               \
     printf(                                                                  \
         "Fatal Error: Attempted calling an SSE2 function when both compile " \