diff options
Diffstat (limited to 'wiki/Graphics-Execution.md')
| -rw-r--r-- | wiki/Graphics-Execution.md | 177 |
1 files changed, 99 insertions, 78 deletions
diff --git a/wiki/Graphics-Execution.md b/wiki/Graphics-Execution.md index 6cadc98..217cb13 100644 --- a/wiki/Graphics-Execution.md +++ b/wiki/Graphics-Execution.md @@ -12,7 +12,7 @@ > [`../docs/04-implementation/04-real-render-walls.md`](../docs/04-implementation/04-real-render-walls.md) > (the measured walls); the docs tree wins on disagreement. -Status: 2026-11-16, tree `master @ a06d4b13`. Tags: `[live]` observed end to +Status: 2026-11-17, tree `master @ 1db3831e`. Tags: `[live]` observed end to end; `[wip]` partially built; `[wall]` current blocker; `[target]` defined success state; `[planned]` designed, not built. @@ -161,46 +161,48 @@ fidelity: `MME-PRELUDE`), and the chain `MME-CALL-HOLD -> MME-CARRY-SAVE -> MME-CARRY-FIRE -> MME-PACK-REARM -> MME-SA-REARM` fires; the prelude latch survives the per-fire channel recycle (`MME-LATCH-KEEP`) and every - admitted-window chunk is forwarded in order. **The decisive finding is - guest-side**: across 36 boots (c990-c1019) the client never submits a - non-null `CALL(34)` draw macro — every `0x3910/0x3914` hit is an interior - method write with `data=0x0`, no CALL header exists in the admitted - windows (`DRAW-SCAN-PRE` found=0), and the same cycles logged guest - EGL/dri2 failures and OOM kills, so the guest GL path was not producing a - real draw submission. A GZDoom video (four files, 22 s flight + 18 s - orbit, 640x480/1280x960, 30 fps, byte-verified) was delivered through the - guest's **software GL** path (Mesa llvmpipe) as a guaranteed-quality - deliverable — not a vGPU rendering claim. Stores on the vGPU path still - miss (`wrote=0`, `0/1025` pages, `px0=0`); the window/screen present is - still the frontier (§6), and the pack-completion and X-start lotteries - remain the throughput limiters. - **The real client draw uses `CALL_MME_MACRO(0)`.** The real-NVIDIA-GL - capture (c1021/c1022) shows the 12068-byte draw window at push - `0x402f9d8` carries no `0x3910/0x3914` write. Its draw is an IMMD (op 4) - write to method `0x3800` (`hdr=8d480e00`, `imm=0xd48` = 3400), followed - by CALL slots 44/57/58 (`0x3960`, `0x39c8`, `0x39d0`). That window also - carries no colour-target bind of its own and no persisted bind was live, - so the RT-gated admission refused it and it retired through the - compute-QMD lane (`stream_3d=false`). Recognition now covers the real - marker: `0x3800` (op 4; `imm>=0x100` as the admission floor) plus - `0x3960/0x39c8/0x39d0`, with `0x3910/0x3914` kept; the call-0 predicate - scans the admitted window (`start_off..at_off`) because a pre-window run - header can swallow the window head (word 316's op=0 count=1059 covers - words 400..1459 and hides the CALL at word 655 from a from-zero walk). - Admission is on the real marker even without a bind, the chunk carrying - the CALL is forced first (`draw_slice=1`, order 0), and every chunk is - replayed verbatim. The CALL is not interpreted in software: the emitted - window is walked by the physical GR front-end (the `0x114` prelude - uploads precede the window in the same launch), and - `MME-EXEC-CALL slot=0 ... op=4 imm=...` records the call reaching the - emitted stream. c1025 receipts: `GL-FLUSH ... admitted=1`, + admitted-window chunk is forwarded in order. **The late-11-16 wave crossed the guest submission gap.** An offscreen + SDL EGL/GBM shim (`GZ_EGL`) gives the guest a real NVIDIA GL context + without Xorg/DDX (`GL_RENDERER: NVIDIA GeForce RTX 2080 SUPER/PCIe/SSE2`, + `GL_VERSION: 4.6.0 NVIDIA 570.144` — no llvmpipe). The real GZDoom draw + window then surfaced: 12068 B at push `0x402f9d8` (3017 words), carrying + no `0x3910/0x3914` write. Its draw is an IMMD (op 4) CALL to method + `0x3800` (`imm=0xd48` = 3400) plus CALL slots 44/57/58 (`0x3960`, + `0x39c8`, `0x39d0`), 7x `0x1b00` inline uploads, and state words. The + window carries no colour-target bind of its own, so RT-gated admission + had refused it (it retired through the compute-QMD lane); recognition now + covers the real marker (`0x3800` op=4 with `imm>=0x100`, plus the three + secondaries; `0x3910/0x3914` kept), admission happens on the real marker + even without a bind, the CALL-bearing chunk is forced first, and every + chunk is replayed verbatim. c1025 receipts: `GL-FLUSH ... admitted=1`, `DRAW-WRITE ... m=0x3800 op=4 data=0xd48`, `3D mark admitted (CALL(0) - 0x3800 draw marker)`, `3D stream chunked`, `3D chunk launch ... draw=1`, - `MME-CALL-ARGS slot=0 data=0xd48`, `MME-EXEC-CALL slot=0 ... imm=0xd48`. - The captured SA[0] is still unresolved (`addr=0x0`), the exact - 12068-byte window's post-fix admission was not re-captured (the run - wedged on an earlier draw window first), and stores / `DRAW_IAB - mac=0x22` remain the open frontier. + 0x3800 draw marker)`, `3D stream chunked` (4 chunks), `3D chunk launch + ... draw=1`, `MME-CALL-ARGS slot=0 data=0xd48`, `MME-EXEC-CALL slot=0`. + **CALL(0) no longer wedges**: eng89t resolves the CALL(0) immediate as an + IR address through the client SA[0] (then golden) and elides an op=4 + CALL(0) whose body is not resident for this launch to a NOP + (`MME-CALL-UNMAPPED ... reason=not-resident`) so the ladder continues. + On c1026/c1027 the real draw's CALL(0) entered its captured macro + (`macro-status=00000004`, `mpc` 0->4) without trapping, and c1027 + produced the first DS with real stage data (`DS-FAST2 + ds_after=83044206 stage=015e0900,015e0a00,015c0400`). The trap code + `0x80001528` is decoded as the sticky `NV_PGRAPH_TRAPPED_ADDR` latch + (bit31 INVALID, subch 0, method `0x1528` = `SET_SHADER_EXCEPTIONS`), not + a cause/PC; the surviving wedge is the synthetic golden macro body + (`badf1201 pc=513 mac=0xf1`) on a non-CALL(0) launch. **Remaining gap + (precise):** the client's per-draw MME macro body was never captured — + the 59580-byte prelude carries only the 12 setup macros (573 code words, + pc 0..0x98), the draw push has no `0x114/0x11c`, and `0xd48` is a macro + length/argument, not an IR entry. The engine has the real draw method + stream but not the body to execute. Next: capture the per-draw `0x114` + upload (every beef push plus the X/2D channel) or implement/verify the + MME64 body, install it before the draw launch, then check `DS-FAST2 + ds_after=8[01]`, `GUEST-WIT va=0x15d`, `wrote>0`, `px0!=0`. No vGPU + surface store or pixel is proven yet (`wrote=0`, no `px_nonblack`). The + software-GL GZDoom video (four files, 22 s flight + 18 s orbit, + 640x480/1280x960, 30 fps, byte-verified) remains the guaranteed-quality + deliverable and is not a vGPU rendering claim; the pack-completion and + X-start lotteries still gate run rate. ## 4. The channel and object picture * **Every submission lives on a channel**: a GPFIFO ring plus engine objects, @@ -263,9 +265,12 @@ swaps, and the GR backend no longer times out (the pair-close/pair-safe replay executes the `DRAW_IAB` (`mac=0x22`) with a synthetic viewport, and the 11-16 wave proved the whole host chain deterministic (admission of the real pushes, clean DS, golden pack, CALL hold, carry fire, SA re-arm). The - remaining question is now guest-side — a real non-null `CALL(34)` draw - submission — after which the vertex/store and the window/screen present - path are next. + late-11-16 wave then crossed the guest submission gap: a real NVIDIA EGL + context without Xorg/DDX, the real GZDoom op=4 `CALL(0)` `0x3800` draw + admitted and replayed, `CALL(0)` entering its captured macro without + trapping, and the first DS with real stage data. The remaining gate is + the client's per-draw MME macro body (never captured), then the + vertex/store and the window/screen present. During bring-up the module also carries a **diagnostics-only mirror** for apps that spin on a shared progress word; it exists so a run can proceed while the @@ -274,34 +279,43 @@ requires it switched off, along with any external poke. ## 6. Where the frontier is (ranked walls) -1. **Guest draw submission and the direct-GL store** `[wall]`. The - **host-side chain is proven end to end** (2026-11-16): the real frame - pushes are admitted (persisted colour-target bind from an earlier trusted - push; draw markers accepted inside counted method runs; the 14.8-thousand- - word window admitted and replayed as 40 chunks with a data-led start snap - and a 128-slot candidate ring), the beef draw lane latches a **clean DS** - (`DS-FAST2 ... ds_after=80000000`, c963) at the live-derived GS attribute - map, capture-less boots get a **golden 2799-word MME pack** from the - golden library, and `MME-CALL-HOLD -> MME-CARRY-SAVE -> MME-CARRY-FIRE -> - MME-PACK-REARM -> MME-SA-REARM` fires with the prelude latch kept across - the per-fire channel recycle; chunks are forwarded in order. The decisive - finding of c990-c1019 (36 boots) is **guest-side**: the client never - submits a non-null `CALL(34)` draw macro — every `0x3910/0x3914` hit is an - interior method write with `data=0x0`, no CALL header exists in the - admitted windows (`DRAW-SCAN-PRE` found=0), and those cycles logged guest - EGL/dri2 failures and OOM kills, so the guest GL path was not producing a - real draw submission in that boot class. The next work is guest-first - (repair/verify the guest EGL/dri2 path and memory, only then capture one - frame's submission); `MME-CALL-ARGS slot=34 data=`, `slot34=1`, - `mac=0x22`, `VA=0xd74` and colour pages at `0x15d...` are the acceptance - receipts. Closed on the way: the QMD-carrier theory (refuted: no QMD bit - in the entry format, no QMD magic in 256 MiB BAR1), RT routing (the clear - targets RT0 while the window binds RT1; RT0 bind verified 16x with correct - target/geometry, still 61/61 `wrote=0`), and the `CODESRC` fault-park class - (suspect program page relocated to the engine-mapped window-0 shadow, - `17c7ed7c`; validated boot: 31 relocations, 51 GR launches, 79 GP_GET - advances, 0 `FAULT_PDE`, 0 `WEDGE-REARM`, `backend_ok=35`). Stores on the - vGPU path still miss (`wrote=0`, `0/1025`, `px0=0`). +1. **The per-draw MME macro body and the store** `[wall]`. The + **host-side chain is proven end to end** and the **guest submission gap + was crossed late on 2026-11-16**: an offscreen SDL EGL/GBM shim (`GZ_EGL`) + gives the guest a real NVIDIA GL context without Xorg/DDX + (`GL_RENDERER: NVIDIA GeForce RTX 2080 SUPER/PCIe/SSE2`, + `GL_VERSION: 4.6.0 NVIDIA 570.144`), and the real GZDoom draw window + surfaced — 12068 B at push `0x402f9d8`, an op=4 IMMD CALL to method + `0x3800` (`imm=0xd48=3400`) plus CALL slots 44/57/58 + (`0x3960/0x39c8/0x39d0`), with no colour-target bind of its own. + Recognition now covers that real marker, admission happens on the marker + even without a bind, the CALL-bearing chunk is forced first, and every + chunk replays verbatim (c1025: `GL-FLUSH admitted=1`, `DRAW-WRITE + m=0x3800 op=4 data=0xd48`, `3D chunk launch ... draw=1`, + `MME-CALL-ARGS slot=0 data=0xd48`, `MME-EXEC-CALL slot=0`). **CALL(0) no + longer wedges**: eng89t resolves the immediate as an IR address through + the client SA[0] (then golden) and elides a not-resident op=4 CALL(0) to + a NOP (`MME-CALL-UNMAPPED ... reason=not-resident`); c1026/c1027 entered + the captured macro (`macro-status=00000004`, `mpc` 0->4) without + trapping, and c1027 produced the first DS with real stage data + (`DS-FAST2 ds_after=83044206 stage=015e0900,015e0a00,015c0400`). The trap + code `0x80001528` is a sticky `NV_PGRAPH_TRAPPED_ADDR` latch (bit31 + INVALID, subch 0, method `0x1528` = `SET_SHADER_EXCEPTIONS`), not a + cause/PC; the surviving wedge is the synthetic golden macro body + (`badf1201 pc=513 mac=0xf1`) on a non-CALL(0) launch. **The remaining + gap is the client's per-draw MME macro body**, which was never captured: + the 59580-byte prelude carries only the 12 setup macros (573 code words, + pc 0..0x98), the draw push has no `0x114/0x11c`, and `0xd48` is a macro + length/argument, not an IR entry. Next: capture the per-draw `0x114` + upload (every beef push plus the X/2D channel) or implement/verify the + MME64 body, install it before the draw launch, then check `DS-FAST2 + ds_after=8[01]`, `GUEST-WIT va=0x15d`, `wrote>0`, `px0!=0`. Stores on + the vGPU path still miss (`wrote=0`, `0/1025`, `px0=0`). Closed on the + way: the QMD-carrier theory (refuted), RT routing (the clear targets RT0 + while the window binds RT1; RT0 bind verified 16x, still 61/61 + `wrote=0`), and the `CODESRC` fault-park class (`17c7ed7c`; validated + boot: 31 relocations, 51 GR launches, 79 GP_GET advances, 0 `FAULT_PDE`, + 0 `WEDGE-REARM`, `backend_ok=35`). 2. **Window/screen present** `[wall]`. The pbuffer and off-screen frame paths are proven (`pixel=ff0000ff`, 500/500 non-black frames, captured videos), but a windowed present does not land: glxgears' window is black, the DDX @@ -372,7 +386,10 @@ requires it switched off, along with any external poke. | Real frame pushes admitted: persisted colour-target bind from an earlier trusted push (`RT-PERSIST`), draw markers accepted inside counted method runs, 14.8-thousand-word window admitted and replayed as 40 chunks (`3D stream chunked ... -> 40 chunk(s)`), data-led start snap (`WIN-START-SNAP`), 128-slot candidate ring | `[live]` | | Clean DS on the beef draw lane after the full-window chunking (`DS-FAST2 ... ds_after=80000000`, c963) at the live-derived GS attribute map (maps=6), no SPH18; draw anchor after the first `0x3810` and raster enables emitted | `[live]` | | Golden 2799-word MME pack on capture-less boots (`MME-GOLDEN overlay`, `MME-PRELUDE pack ir_words=2799`); `MME-CALL-HOLD -> MME-CARRY-SAVE -> MME-CARRY-FIRE -> MME-PACK-REARM -> MME-SA-REARM` fires; prelude latch kept across the per-fire channel recycle (`MME-LATCH-KEEP`); admitted-window chunks forwarded in order | `[live]` | -| Guest draw submission: across 36 boots (c990-c1019) the client never submits a non-null `CALL(34)` draw macro — every `0x3910/0x3914` hit is an interior method write with `data=0x0`; no CALL header in the admitted windows (`DRAW-SCAN-PRE` found=0); guest EGL/dri2 failures and OOM kills in the same cycles; stores `wrote=0` / `0/1025` | `[wall]` | +| Guest GL submission crossed (late 11-16): offscreen SDL EGL/GBM shim (`GZ_EGL`) gives GZDoom a real NVIDIA GL context without Xorg/DDX (`GL_RENDERER: NVIDIA GeForce RTX 2080 SUPER/PCIe/SSE2`, `GL_VERSION: 4.6.0 NVIDIA 570.144`; no llvmpipe) | `[live]` | +| Real GZDoom draw decoded and admitted: 12068 B window at push `0x402f9d8`, op=4 IMMD CALL to method `0x3800` (`imm=0xd48`) plus slots 44/57/58 (`0x3960/0x39c8/0x39d0`), no `0x3910/0x3914`; marker recognition, bind-free admission, draw-chunk-first, verbatim chunk replay (`3D chunk launch ... draw=1`) | `[live]` | +| `CALL(0)` executes without trapping: immediate resolved as an IR address through client SA[0] then golden (`MME-CALL-RESOLVE`), not-resident calls elided (`MME-CALL-UNMAPPED ... reason=not-resident`); c1026/c1027 entered the captured macro (`macro-status=00000004`, mpc 0->4), first DS with real stage data (`DS-FAST2 ds_after=83044206 stage=015e0900,015e0a00,015c0400`); trap `0x80001528` decoded as a sticky TRAPPED_ADDR latch, not a cause/PC | `[live]` | +| Per-draw MME macro body: never captured (the 59580-B prelude carries only the 12 setup macros, 573 code words pc 0..0x98; the draw push has no `0x114/0x11c`; `0xd48` is a length/argument, not an IR entry); the surviving wedge is the synthetic golden body (`badf1201 pc=513 mac=0xf1`) on a non-CALL(0) launch; stores `wrote=0`, no `px_nonblack` | `[wall]` | | GZDoom video via the guest **software GL** path (Mesa llvmpipe): four byte-verified files (22 s flight + 18 s orbit, 640x480/1280x960, 30 fps) — a guaranteed-quality deliverable, **not** a vGPU rendering claim | `[live]` (software path) | | Pack-completion and shallow-boot lotteries gate the deterministic release (throughput limiters) | `[wip]` | | Window/screen present produces non-black window pixels | `[wall]` | @@ -395,12 +412,16 @@ requires it switched off, along with any external poke. GZDoom. * Display/KMS bring-up (NvKms, `/dev/dri`, X, a real GL context) is not a rendered-frame claim. GZDoom reaching `V_Init: allocate screen` proves the - client's GL init and the direct-GL GR lane, not the game loop or a - presented frame; the host-side chain is proven end to end (admission, - clean DS, golden pack, CALL hold, carry fire, SA re-arm), but the decisive - 11-16 finding is that the client did not submit a non-null `CALL(34)` draw - macro in the recorded boots, so the stores are still missing and the - window stays black. + client's GL init, not the game loop or a presented frame. The late-11-16 + wave went further — a real NVIDIA GL context without Xorg/DDX, the real + op=4 CALL(0) draw admitted and replayed, `CALL(0)` entering its captured + macro without trapping, and a clean DS with real stage data — but the + client's per-draw MME macro body was never captured, the surviving wedge + is the synthetic golden body, and no surface store or pixel exists yet + (`wrote=0`, no `px_nonblack`), so the window stays black. +* The real-NVIDIA EGL context is a guest-GL-path result, not a vGPU pixel + result: it proves the guest's GL stack now submits through the emulated + device, not that the draw's stores reached a render target. * The engine-zero-write (NOWRITE) campaign is not a rendering result: it and the draw census together proved the submission is macro-driven (no method-grammar draw header exists), that RT routing is not the cause, and |
