Repository navigation
Marvel's Wolverine: draw with GS_FAST_LAUNCH=1 (VGT_SHADER_STAGES_EN=0x482030) after the intros; registered shader type 8 #3880
Description
What
Marvel's Wolverine (PPSA03671) now gets through its intro videos (with #757, #3588, #3633, #3846, #3867, #3831 and #3336 on main a11c8f7f7). Its next stop is a draw whose VGT_SHADER_STAGES_EN sets GS_FAST_LAUNCH:
AGC graphics: VGT_SHADER_STAGES_EN=0x482030: unsupported geometry routing, fast launch or wave-ID state
DecodeShaderStages (State.cpp) accepts 0x0047ec30 on the geometry path. Decoded with Mesa's gfx10.3 register description (src/amd/registers/gfx103.json), 0x482030 is:
| field | bits | value |
|---|---|---|
| ES_EN | 3-4 | 2 |
| GS_EN | 5 | 1 |
| PRIMGEN_EN | 13 | 1 |
| GS_FAST_LAUNCH | 19-20 | 1 (not in the accepted mask) |
| GS_W32_EN | 22 | 1 |
On gfx10.3, Mesa's RADV programs GS_FAST_LAUNCH = 1 for mesh shaders, so this looks like Wolverine's mesh-style geometry. The same run also logs, at registration, many Shader 0x11d3... has unsupported registered shader type 8; it is not prepared at registration (106 distinct in one 5-minute window). These may be the same programs; I have not tied a type-8 registration to this draw yet.
Questions
- Is anyone working on fast-launch / mesh-style NGG draws (or registered shader type 8)? fix(agc): route registered NGG workgroups through mesh #2261 routes registered type-2 NGG programs through the existing mesh path. It does not cover this state. I found no open PR or issue for
GS_FAST_LAUNCH. - If nobody is, would a first step limited to decoding this state and the draw packet that goes with it, plus a GPU test, be welcome? Or is there a planned design for mesh-style draws to follow?
Setup
Windows 11, RTX 3070, GCC 15.2.0 UCRT, Release. Stack listed above, plus local placeholders for two imports not on main and a local cap on #757's image-table size (see my comment on #757). Logs are available on request.
AI-assisted: yes (Claude Code)
Activity
oneandonlydean commentedon Oct 11, 2026
Nobody is working on GS_FAST_LAUNCH as far as I know, and no PR is open for it. PPSA02530 stops on the same bit (
0x82030, #3043). A first PR that decodes the state and its draw packet with a GPU test is welcome, as long as everything it does not cover keeps throwing.mrfelek commentedon Oct 13, 2026
Data point from the same title, on
main665ad7d plus #4962 (Linux, RTX 3080 Ti, NVIDIA 615.71.09): once the title's modules load, 131 shaders fail to prepare after registration, for example0x11d0111000and0x11d0110500:Those are vertex programs, which fits the GS_FAST_LAUNCH state this issue describes. I have not tried #4702 on this title. After the failures a game thread segfaults on a call through an empty slot of a function-pointer table in the title's own code; I could not tell whether the failed shaders cause it. The details are in the bring-up thread: #5516
AI-assisted; the runs are from my machine.
oneandonlydean commentedon Oct 14, 2026
Status: #4702 (decode GS_FAST_LAUNCH geometry with one point per subgroup) is on main as a first step. The 131 vertex programs mrfelek reported failing on
s_barrierare not covered by it, so this stays open.