Main build 41252386309 (and 41247522136): linux job failed in check_skips because
agc_shader_disk_cache_large_file skipped (runner short of free disk or memory for a
>2 GiB file) with a message the guard does not allow. Not caused by #13998.
The windows pm4_replay timeout is a one-off flake (1 of 59 red builds).
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Say that the shader numbers are counted from the code, not from playing games,
in the progress map caption, the progress page, the docs and the README badge
text. The Still needed bars already use their own grey track and fixed fills
rather than the todo colour.
Throw sites and image formats go on the left, the form names on the right as
a table-like list. Labels sit left of each bar and the count right of it, never
over it. Bars use a neutral grey track and fixed amber and green fills that do
not follow the shared palette, so a red todo colour cannot turn the tracks red.
Throw sites scale to their 148 total and formats to their 159.
The same source file listed under two build directories forms two units,
sorted by directory, and each unit keeps its records in input order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
List the forms the decoder does not decode and the decoded ones the translator
has no case for, chart the unsupported-throw sites per recompiler area, and
chart the sampled, typed-buffer and host image formats. All of it is read from
the source by the same scan; progress.json only gains fields.
The test reads the mingwHostTests gate of build.yml and checks that it lists core/CMakeLists.txt, cmake/, 3rdparty/, .gitmodules, tools/select_tests.py and tools/configure_seed.py. Against the previous gate it fails all six.
Run 41245658018 (b9bf246ea2). linux: agc_shader_disk_cache_large_file skips
when the runner has under 3 GiB free and check_skips.py did not list that
reason. windows: guest_timed_wait_cpu aborted on a runner scheduling hiccup
(latency bound exceeded, #13710 made it serial but it is still timing-based).
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
The test takes the files progress.py reads from the module itself and checks that the report's pull request filter matches each of them, and that library sources still start it. Against the previous filter it fails on RdnaOpcode.hpp, rdna_isa.txt and nid_names.py.
The scanner fixture gains VOP2 forms at 0x100, 0x13F and 0x140 so a shifted
bound changes a known answer. The command-line summary calls the old count
what it is (opcode names declared). The PR progress report compares only
library and legacy fields, so it no longer runs for recompiler-only changes.
The progress map panel and the page table still printed the old instruction
name count (100%). Both now show the decoded forms by encoding. The scanner
fixture gains VOPC and VOP1 forms in VOP3, the interpolation emit switch, the
export family, a DPP16 refusal and the opcode range edges, so a change to any
of these rules changes a known answer.
GitHub returns some open pull request bodies with an escape that is not valid
JSON, for example a backslash followed by u00XX. gh itself rejects the page,
and pr_overlap.py stops with "Invalid \uXXXX escape" on the whole open-PR
list. Add a test for a page containing such a body, and one that checks
valid escapes are still decoded to the same text.
The first test fails on upstream main with that error. The second passes.
The shaders badge counted ISA instruction names that have an RdnaOpcode
enumerator and had reached 100%. Measure what the recompiler does instead:
decoded and translated instruction forms (base, VOP3, DPP) from the opcode
tables and emit switches, and the unsupported throws, using AMD's RDNA 2
instruction forms (MIT, vendored as a derived table). Say on the page that
this is source coverage, not game compatibility.
The Pages progress workflow now also runs when the recompiler changes, and
the Pages workflows fail instead of publishing when the scan cannot read the
recompiler.
Adrià