agc_driver_vopc_compare failed on every Linux runner with an
AVX512-FP16 Xeon (Sapphire, Emerald and Granite Rapids) and passed on
AMD EPYC and Ice Lake runners. Ubuntu 24.04's lavapipe compiles with
LLVM 20.1.2, which on these CPUs merges a masked fcmp ult/uge on half
vectors with the same compare of swapped operands, so v_cmp_nle_f16,
v_cmp_nlt_f16 and v_cmpx_nle_f16 return the result of their mirror.
The recompiler's SPIR-V is the same on every device and correct.
GALLIUM_OVERRIDE_CPU_CAPS=avx keeps LLVM off AVX-512, so every runner
compiles the tests the same way.
MixPrecision and SdwaDestination dispatch hand-assembled RDNA through the
recompiler and compare every lane with a CPU model, but were never in ctest.
Move them to tests/execution/, which is globbed into one test per file with
the same target names, so new execution tests add a file and no CMake lines.
They exit 77 (skipped) when the host has no Vulkan device; the Linux runner
installs lavapipe and sets ANYPS5_REQUIRE_VULKAN=1 so they cannot skip there.
Co-Authored-By: Claude <noreply@anthropic.com>
WinLibs puts its own cmake.exe (4.3, OpenSSL) first in PATH and ships no
CA certificates, so file(DOWNLOAD) of the FFmpeg package failed with
curl error 60. SSL_CERT_FILE points it at the bundle of Git for Windows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Feat/provide release
Added a release workflow for v* tags: build Windows and Linux, run tests, and publish only if both platforms are successful. The release contains PRX archives, relinker executables, and documentation from docs/user. The build cache is reused.
- `prx-windows-v0.1.0.zip`
- `prx-windows-v0.1.0.tar.gz`
- `prx-linux-v0.1.0.zip`
- `prx-linux-v0.1.0.tar.gz`
- `relinker-v0.1.0.exe`
- `relinker-v0.1.0`
- `COMPATIBILITY.md`
- `INPUT_MAPPING.md`
Builds with tests enabled and runs ctest on every push to main and every
pull request: GCC on Linux, and on Windows the MinGW-w64 winlibs GCC
15.2.0 toolchain required by the README. ccache and the toolchain are
cached, and only pushes to main save the ccache, so pull requests reuse
it without evicting it.
Windows build steps run in cmd because Git Bash puts its own mingw64 DLLs
first in PATH, which breaks build tools such as nid_patcher that link
libstdc++ dynamically.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
progress.py stores the names of implemented and pending items, can
measure another source tree with --root and prints a Markdown diff of
two runs with --compare. The Progress report workflow compares every
pull request with its base, and Progress comment posts the result as a
single comment, updated on every push and removed when progress no
longer changes. Commenting runs as a separate workflow_run so pull
requests from forks never run with write permissions.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
tools/progress.py counts implemented prx functions and RDNA instructions
recognized by the shader recompiler, and renders treemaps, badges and an
index page. The Progress workflow deploys them to GitHub Pages on every
push to main, so nothing generated is committed to the repository.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Dean Galvin
boykopovar
Adrià Franch
STL1te