![\"\"](\"https:\/\/www.emu-france.com\/wp-content\/uploads\/2026\/09\/prosper.png\")
<\/a>\n <\/p>\n
Les changements sont les suivants:
\n <\/p>\n
– ci: add the platform-stub ratchet rule for HLE(*_stub) platform-arm handlers by @ymatrix1 in #4366
\n– gpu: preserve independent logical64 WAITCNT completion by @mattias800 in #4372
\n– docs: list all eleven ratchet rules in the charter and state the platform-arm rule by @ymatrix1 in #4367
\n– docs: clarify APR proposal and preserve historical Dragon Quest observations by @ymatrix1 in #4365
\n– fix(gpu): retain producing raster launch controls through capture replay by @mattias800 in #4376
\n– ci: enforce pytest policy on new Python tests by @ymatrix1 in #4314
\n– Bound native shader walkers by authentic readable source windows by @mattias800 in #4377
\n– docs: record current GTA Stage 2 ABBA checkpoint by @mattias800 in #4379
\n– docs: record Sonic deferred-wait world falsification by @mattias800 in #4380
\n– docs: record GTA warm target-count checkpoint by @mattias800 in #4381
\n– feat(gpu): connect original scalar-buffer reads to shared wave64 bank by @mattias800 in #4374
\n– test: use SDK-relative pointers in scalar-bank fixtures by @mattias800 in #4382
\n– test: adversarial pins for rdna2 closed-tail block admission by @ymatrix1 in #4386
\n– test: LLVM oracle pins for SOP2 destination widths by @ymatrix1 in #4390
\n– test: follow v71 capture tail and Wave32 eager routing so main’s Linux CI is green by @mattias800 in #4396
\n– test: pin mandatory-literal K capture for VOP2 mul-adds and SOPK setreg by @ymatrix1 in #4401
\n– feat(gpu): implement VOP3P integer dot product lowering (v_dot2, v_dot4, v_dot8) by @ymatrix1 in #4400
\n– test: pin SOPP direct-branch set against the ISA table by @ymatrix1 in #4405
\n– test: oracle pins for scalar opcode predicates by @ymatrix1 in #4408
\n– feat(gpu): implement v_min3_i32, v_max3_i32, and v_med3_i32 lowering by @ymatrix1 in #4404
\n– feat(gpu): implement subword SAD instructions (v_sad_u8, v_sad_hi_u8, v_sad_u16, v_msad_u8) by @ymatrix1 in #4388
\n– feat(gpu): implement v_perm_b32 byte permutation lowering by @ymatrix1 in #4389
\n– feat(gpu): implement image_gather4 texture gather lowering by @ymatrix1 in #4392
\n– test: ISA-table pins for memory-writer classification by @ymatrix1 in #4393
\n– feat(gpu): implement v_div_fixup_f32 division fixup lowering by @ymatrix1 in #4391
\n– feat(gpu): implement s_movrels_b32 relative SGPR move lowering by @ymatrix1 in #4394
\n– perf: emit opt-in per-flip stage timing samples by @ymatrix1 in #4406
\n– feat(gpu): implement v_mad_i64_i32 lowering by @ymatrix1 in #4387
\n– gpu: bound s_movrels_b32’s constant M0 before adding it to the base by @mattias800 in #4413
\n– test: migrate normal_magnitude_predicate to GTest by @ymatrix1 in #4414
\n– test: extract minmax3 kernels from rdna2_to_spirv main (pilot split) by @ymatrix1 in #4415
\n– gpu: recover GTA V’s #4270 bank-scene regression (2.9 -> 6.5 fps) by @mattias800 in #4421
\n– feat(hle): implement libSceAudio3d\/AudioPropagation\/Acm spatial audio stubs by @ymatrix1 in #4418
\n– recompiler: stop phantom raw-wide refusals of register-offset descriptor loads (#4422) by @mattias800 in #4425
\n– feat(hle): implement libSceNet sockets and libSceNp offline stubs by @ymatrix1 in #4417
\n– gpu: seed BGRA render targets into compute storage on the device (+ lift live compute’s storage codec) by @mattias800 in #4428
\n– gpu: read back the span before an ordered DMA only when the copy touches it (Alex Kidd holds 60 fps) by @mattias800 in #4459
\n– docs: Alex Kidd plateau finding and async pair outcome by @mattias800 in #4461
\n– test: migrate 8 host\/loader\/HLE suites to GoogleTest by @ymatrix1 in #4430
\n– test: migrate the fragment float-flag identity suites to GoogleTest by @ymatrix1 in #4441
\n– test: migrate the select\/pselect pure-sleep suite to GoogleTest by @ymatrix1 in #4446
\n– test: migrate the guest-facing HLE handler contracts to GoogleTest by @ymatrix1 in #4434
\n– test: migrate the diagnostics enabled-path and JSON suites to GoogleTest by @ymatrix1 in #4449
\n– test: extract scalar-ALU kernels from rdna2_to_spirv main by @ymatrix1 in #4431
\n– test: migrate the pthread-key destructor suite to GoogleTest by @ymatrix1 in #4443
\n– test: extract VOP3 batch kernels from rdna2_to_spirv main by @ymatrix1 in #4432
\n– test: migrate the clock and timed-wait HLE contracts to GoogleTest by @ymatrix1 in #4437
\n– test(gpu): pin f64 transcendental refusal with rcp_f32 control by @ymatrix1 in #4453
\n– test(gpu): pin div-fmas refusal with fixup control by @ymatrix1 in #4454
\n– test(gpu): pin scalar-misc refusal with movk\/mov controls by @ymatrix1 in #4451
\n– test(gpu): pin float-dot refusal with integer-dot control by @ymatrix1 in #4456
\n– test(gpu): pin x2 buffer-atomic refusal with 32-bit control by @ymatrix1 in #4460
\n– test(gpu): pin SOPP halt\/kill refusal with s_nop control by @ymatrix1 in #4445
\n– test(gpu): pin 64-bit DS refusal with ds_add_u32 control by @ymatrix1 in #4448
\n– test(gpu): pin WRXCHG2 DS refusal with write2 control by @ymatrix1 in #4450
\n– test(gpu): pin multi-level IMAGE_LOAD_MIP refusal with plain-load control by @ymatrix1 in #4438
\n– test(gpu): pin NSA IMAGE_STORE_MIP refusal with consecutive-store control by @ymatrix1 in #4444
\n– feat(hle): register remaining dialog entry points (Login\/MsgProgress\/WebBrowser) by @ymatrix1 in #4458
\n– feat(hle): register Trophy2 icon getters alongside the info queries by @ymatrix1 in #4440
\n– feat(hle): register Http option setters, epoll lifecycle and Https setup by @ymatrix1 in #4442
\n– gpu: bound MSAA\/unusual-swizzle colour targets so ordered DMAs skip the span readback (Summer Sports) by @mattias800 in #4464
\n– feat(hle): register sceSaveDataLoadIcon completing the icon round-trip by @ymatrix1 in #4436
\n– feat(hle): register UDS DestroyHandle\/SetInt32 and sceSaveDataDelete by @ymatrix1 in #4433
\n– feat(hle): register remaining UDS setters, lifecycle and stats (25 exports) by @ymatrix1 in #4455
\n– feat(hle): register libSceNpWebApi2 request and push-event surface by @ymatrix1 in #4435
\n– feat(hle): implement libSceRtc calendar arithmetic (22 exports) by @ymatrix1 in #4447
\n– hle: fix np.cpp splice that broke the build (GetStorageStat body) by @mattias800 in #4468
\n– feat(hle): register libSceNpManager remainder (14 exports) by @ymatrix1 in #4476
\n– feat(hle): register notice-skip setters, PowerTick and privacy\/Init2 by @ymatrix1 in #4478
\n– feat(hle): report voice unavailable across the port surface by @ymatrix1 in #4473
\n– feat(hle): register MouseClose and pad mic-in-use vibration quirk by @ymatrix1 in #4471
\n– feat(hle): register Net epoll\/introspection\/resolver and Rudp setup by @ymatrix1 in #4482
\n– test(gpu): pin float image-atomic refusal with add control by @ymatrix1 in #4470
\n– test(gpu): pin arithmetic x2 buffer-atomic refusal with 32-bit control by @ymatrix1 in #4474
\n– test(gpu): pin swappc\/code-end refusal with mov\/nop controls by @ymatrix1 in #4475
\n– test(gpu): pin bfe64\/mode-set refusal with mov\/nop controls by @ymatrix1 in #4477
\n– test(gpu): pin endpgm-variant refusal with endpgm control by @ymatrix1 in #4479
\n– test(gpu): pin setreg-imm32 refusal with flat-scratch control by @ymatrix1 in #4480
\n– test(gpu): pin deferred 32-bit atomic refusal with add control by @ymatrix1 in #4481
\n– test(gpu): pin image_atomic_fcmpswap and image_atomic_dec refusals (supersedes #4472) by @mattias800 in #4488
\n– memory: a fixed direct map replaces tracked direct memory on both platforms (replaces #4189) by @ymatrix1 in #4486
\n– feat(hle): implement NGS2 allocator-create\/destroy\/grain\/lock lifecycle (6 exports) by @ymatrix1 in #4487
\n– feat(hle): implement Fiber\/Ult remainder (7 exports) by @ymatrix1 in #4491
\n– feat(hle): implement AvPlayer trick-speed\/bandwidth\/printf remainder (4 exports) by @ymatrix1 in #4490
\n– feat(hle): implement Font render-state remainder (13 exports) by @ymatrix1 in #4494
\n– feat(hle): implement GameUpdate request lifecycle (5 exports) by @ymatrix1 in #4497
\n– gpu: restore the Unity UI vertex shader (Greak, Alex Kidd, Summer Sports menus and HUDs) + fix the macOS build by @mattias800 in #4495
\n– gpu: Summer Sports’ track is the right colour again (BC6H lightmaps clamp until the pipeline is linear) by @mattias800 in #4501
\n– feat(hle): register AJM direct-job shapes (14 exports) by @ymatrix1 in #4484
\n– feat(hle): implement Ime text-entry session (11 exports) by @ymatrix1 in #4503
\n– feat(hle): implement libSceCesCs conversion and libScePngDec decode by @ymatrix1 in #4452
\n– hle: sceRtcGetTick is the same on every host (fixes the Rtc tests on macOS and Windows) by @mattias800 in #4505
\n– host(win): keep the host’s allocations out of the guest’s address range \u2014 the UE4 launch hang (#4426) by @mattias800 in #4507
\n– hle: re-derive NGS2\/AvPlayer\/Font contracts from the .native modules PS5 titles link by @mattias800 in #4524
\n– feat(hle): implement NGS2 waveform parse\/calc\/reset-option (3 exports) by @ymatrix1 in #4510
\n– hle(apr): a counter completion is never delivered below one already delivered \u2014 Kena’s boot crash (#4504) by @mattias800 in #4517
\n– docs(kena): Windows -> Linux\/AMD handoff of the title-screen 3D bring-up by @mattias800 in #4500
\n– gpu: Space Adventure Cobra’s in-game 3D renders again (two recompiler defects in its per-light loop) by @mattias800 in #4522
\n– recompiler: accept UE4’s save-then-v_cmpx EXEC guard around a counted loop (Kena, #4427) by @mattias800 in #4534
\n– gpu: close two raw-wide classifier gaps found in review (wrong relative-move opcodes, implicit VCC readers) by @mattias800 in #4535
\n– gpu: a loop exit drops the scratch a mask overwrote, for every loop by @mattias800 in #4542
\n– tools\/ci: decode git and ruff I\/O as UTF-8 in check_python_quality by @ymatrix1 in #4539
\n– feat(hle): implement Ult pool\/runtime destroys (2 exports) by @ymatrix1 in #4528
\n– recompiler: admit several bottom-tested SCC do-while loops in one program (#4518) by @mattias800 in #4546
\n– tools: static HLE progress census by @ymatrix1 in #4532
\n– diagnostics: say whether Wave64 was the GPU’s doing or prosper’s by @mattias800 in #4531
\n– docs: architecture spec and decision records (ADRs 0001-0021), gated by check_arch_docs by @ymatrix1 in #4536
\n– gpu: s_movrels_b32 reads a range of registers, and the liveness proofs now know it (#4538) by @mattias800 in #4547
\n– tools\/docs: unbreak the Docs job (check_arch_docs’ \u00ab\u00a0ctest\u00a0\u00bb kind is not a ctest invocation) by @mattias800 in #4561
\n– docs: add Rayman 30th Anniversary Edition and MOUSE: P.I. For Hire with their first captures by @mattias800 in #4558
\n– gpu: two leftovers from the #4547 review (raw-x2 escape list; dispatcher SCC after a relative read) (#4559) by @mattias800 in #4562
\n– gpu: keep the program a draw lost to the owned-wave gate, and the gate’s reason (#4555) by @mattias800 in #4565
\n– gpu: AGC’s eliminate-fast-clear rectangle writes no colour (#1588) by @mattias800 in #4568
\n– gpu: four raw-wide classifier gaps, and one emitter fix they depend on, that put GTA V’s and MOUSE’s fragment programs on the owned-wave path (#4555) by @mattias800 in #4569
\n– hle\/agc: sceAgcCbBranch builds its packet; the branch runs where the guest put it (#4540) by @mattias800 in #4576
\n– tools\/env: register every PROSPER_* switch and require new ones be classified by @ymatrix1 in #4543
\n– live compute: publish an RGB10A2 compute result into the renderer target by @ymatrix1 in #4571
\n– revision: see through a ccache\/sccache compiler wrapper in the dependency identity (#4356, #4187) by @mattias800 in #4581
\n– tools\/env: drop the stale PROSPER_AGC_SPLIT_FINAL_STATE registry row (main is red) by @ymatrix1 in #4582
\n– gpu: a clashed constant-buffer key is published per consumer, not per key by @mattias800 in #4584
\n– gpu: key a fold descriptor use the way the recompiler tags it (#4585) by @mattias800 in #4588
\n– docs: record that holding the EOP pulse does not stop DQ7’s label recycle (#2982) by @ymatrix1 in #4583
\n– gpu: an x2 load can be the latched source of a raw-wide register offset (#4578) by @mattias800 in #4587
\n– docs: blog Beast and DQ7 after the eliminate-fast-clear fix; record Kena’s refusal progress by @mattias800 in #4589
\n– feat(hle): implement Font StyleFrame cluster (15 exports) by @ymatrix1 in #4572
\n– docs: every document declares its kind; every relative link resolves by @ymatrix1 in #4567
\n– gpu: a register-offset raw-wide load’s pointer need only survive until the load; a branch before its source is harmless unless it lands inside by @mattias800 in #4593
\n– spv_validate: rdna2_proven_raw_register_wide_entry_loads is not an emitter (main red after #4593) by @mattias800 in #4595
\n– gpu: a base-zero T# with constant-zero selectors is the null image (#4592) by @mattias800 in #4596
\n– gpu: merged-NGG subgroup planner and launch values (#3135 P1) by @mattias800 in #4597
\n– gpu: fall back to the general CFG route when a counted loop’s prelude declines (Kena 0x5006fb0000) by @mattias800 in #4594
\n– revision: write the compiler-wrapper test in pytest conventions (Python quality red after #4581) by @mattias800 in #4601
\n– gpu: NGG evidence for the merged-NGG lowering (#3135 P0) by @mattias800 in #4599
\n– gpu: read PROSPER_DBG live for the null-image line (main red after #4596) by @mattias800 in #4604
\n– gpu: close three CFG-dispatcher Wave64 gaps so Kena’s compute 0x5008ec0000 recompiles by @mattias800 in #4603
\n– gpu: merged-NGG subgroup shell, compile and offline execution (#3135 P2) by @mattias800 in #4605
\n– gpu: publish an over-cap buffer window clamped to its mapped run by @ymatrix1 in #4566
\n– gpu: a plain copy of loaded words into EXEC, and a wide load into the special registers, are numeric uses in the raw-wide walk (#4574) by @mattias800 in #4579
\n– feat(hle): implement NGS2 system lock\/unlock\/sample-rate (3 exports) by @ymatrix1 in #4521
\n– feat(hle): implement Font kerning\/library\/rewind\/back-link (4 exports) by @ymatrix1 in #4520
\n– feat(hle): implement Font pixel-resolution\/cache\/library back-link (3 exports) by @ymatrix1 in #4548
\n– feat(hle): implement Font DPI\/glyph-form\/scale\/kerning queries (5 exports) by @ymatrix1 in #4570
\n– feat(hle): implement Font renderer-outline\/DPI\/rewind\/mask (6 exports) by @ymatrix1 in #4573
\n– recompiler: prove two-T# x16 bundles through image stores; disjoint stores keep the split-T# proof by @ymatrix1 in #4564
\n– feat(hle): implement Ult queues and semaphores (14 exports) by @ymatrix1 in #4544
\n– gpu: merged-NGG raster commit: pass-through vertex stage and layer routes (#3135 P3) by @mattias800 in #4611
\n– gpu: type CFG-dispatcher spill reloads, and keep a mask reloaded into its own register (#4600) by @mattias800 in #4607
\n– gpu: a 64-lane fragment launch is not sent to the owned-wave path for a load that is numeric only at 32 lanes (#4555) by @mattias800 in #4609
\n– docs: regenerate PROGRESS_TRACKER.md (MOUSE’s blockers after #4609) by @mattias800 in #4614
\n– gpu: lower ds_read_b32 … gds in compute, the read MOUSE’s per-frame counter copy was refused at (#4553) by @mattias800 in #4615
\n– docs: MOUSE: P.I. For Hire after #4609 and #4615: route, captures, and what is ruled out by @mattias800 in #4616
\n– vk: run merged-NGG draws in the backend: fill, shell dispatches, pass-through runs, no CPU wait (#3135 P4) by @mattias800 in #4617
\n– gpu: let replay admit a conditional branch out of the program (MOUSE frames could not be replayed) by @mattias800 in #4622
\n– gpu: MOUSE’s world renders: a DCC clear reaches the target a pass renders to, and a Z_16 depth plane keeps its size by @mattias800 in #4621
\n– docs: MOUSE’s second depth target is refused before gameplay only (correcting #4621’s docs and #4620) by @mattias800 in #4626
\n– gpu: replay asks whether a write lands in a retained depth plane by address (replayed frames lost every depth target) by @mattias800 in #4628
\n– gpu: reuse a draw’s render state while its registers are unchanged by @mattias800 in #4630
\n– gpu: BOOL64 predicated jumps run on a non-zero condition (DQ7 title clouds\/sea) by @mattias800 in #4610
\n– compute: honour a fast-cleared target’s control plane and a padded linear image’s row pitch by @ymatrix1 in #4586
\n– test: the savedata tmpfs check names the old fixed roots, not any \/tmp prefix by @mattias800 in #4640
\n– docs: Tomb Raider’s rung-3 manor is Tomb Raider II’s; Tomb Raider I’s does not render by @mattias800 in #4644
\n– gpu: borrow the readable start of a draw buffer declared past mapped memory by @mattias800 in #4633
\n– gpu: run merged ES+GS NGG draws live: admission, linked fold, cached producer (#3135 P5) by @mattias800 in #4634
\n– gpu: publish a renderer-claimed volume before compute reads it (#4625) by @mattias800 in #4651
\n– gpu: count draw-disposition seen at the pass entry so silent refusals raise unaccounted-draws by @mattias800 in #4655
\n– gpu: refuse a cached render target for a sampled view needing more bytes per texel (AC Black Flag red frame, #4197) by @ymatrix1 in #4648
\n– Sink vote-lowering diagnostic types to src\/diagnostics by @ymatrix1 in #4646
\n– windows: read the SSE4a-log and stub-dump switches through one helper each (getenv 14 to 10) by @ymatrix1 in #4639
\n– Move shared diagnostics helpers down to src\/diagnostics by @ymatrix1 in #4641
\n– ci: keep reviewer notes through arch-ratchet –update by @ymatrix1 in #4636
\n– Move COMPUTELOG diagnostics out of gpu_executor.cpp by @ymatrix1 in #4642
\n– compute: move execute_item’s post-wait tail into finish_dispatch_tail (preparatory for ADR 0009, proposed) by @ymatrix1 in #4650
\n– fix(prosper-app): list and boot dumps at drive roots by @ymatrix1 in #4658
\n– ADR 0009 Stage 1: design note and a tested retirement-order contract by @ymatrix1 in #4645
\n– docs: Black Flag’s measured frame cost and the bounded-buffer null result by @ymatrix1 in #4632
\n– hle: NetCtl\/Np state callbacks on every host; sceVoiceQoSInit refuses honestly by @ymatrix1 in #4577
\n– feat(prosper-app): Prosper window title and baked-in mark icon by @ymatrix1 in #4659
\n– ratchet: fix hle_kernel.cpp file-size row (5561 -> 5555) by @ymatrix1 in #4662
\n– ci: fix main’s red Docs and Python quality jobs (note frontmatter, stale switch row) by @ymatrix1 in #4661
\n– test: NGG shell-pipeline cache eviction no longer depends on native Wave64 by @mattias800 in #4663
\n– fix: retry transient Windows replace failures saving the pipeline cache by @ymatrix1 in #4667
\n– test: Windows cold-target repeat honours PROSPER_NO_SKIP_SEED by @ymatrix1 in #4666
\n– feat(prosper-app): game list with search, menus and controls view by @ymatrix1 in #4665
\n– ADR 0009 step 2: observe what pipelining would allow (PROSPER_PIPELINE_OBSERVE) by @ymatrix1 in #4664
\n– feat(prosper-app): persisted host settings for library-started boots by @ymatrix1 in #4670
\n– docs(prosper-app): add library view screenshot by @ymatrix1 in #4673
\n– compute: refresh a cached buffer by differing 64 KiB blocks (with a control switch) by @ymatrix1 in #4660
\n– compute: mirror R8Unorm storage results on the GPU (no CPU round trip) by @ymatrix1 in #4656
\n– feat(prosper-app): use Explorer folder picker on Windows by @ymatrix1 in #4672
\n– ci: exempt bench_*.cpp from the gtest policy; keep the NGG cache test’s push range valid by @mattias800 in #4675
\n– feat(prosper-app): add Game Log panel with bounded ring capture by @ymatrix1 in #4671
\n– compute: share one resident arena between overlapping read-only buffer windows (interim, not ADR 0010) by @ymatrix1 in #4635
\n– ci: spv_validate’s coverage scan exempts parent_scan.hpp’s two local vectors by @mattias800 in #4676
\n– test: compute buffer runtime tests expect #4660’s block-sync refresh, and detect a stale far block by @mattias800 in #4678
\n– feat(prosper-app): picker answers join the library; folders accumulate by @ymatrix1 in #4674
\n– perf: read hot-path diagnostic switches once per submit, not once per call by @ymatrix1 in #4677
\n– docs(kena): the title-menu world renders on main; #3835’s NaN half image was AGC’s helper rectangle by @mattias800 in #4679
\n– docs, tools: a path convention for screenshots, enforced for new captures by @ymatrix1 in #4629
\n– charter: point the architecture rules at docs\/spec and docs\/adr by @ymatrix1 in #4541
\n– docs, tools: architecture follow-ups from the PortPS5 docs comparison by @ymatrix1 in #4557
\n– diag: validation mapping census (which memory backs full compares) + GetWriteWatch ruled out by @ymatrix1 in #4681
\n– gpu: render layered volume passes with more than one colour target (#4643) by @mattias800 in #4685
\n– diag: always-on validation census with a size histogram; drop PROSPER_VALIDATION_MAPPING_CENSUS by @ymatrix1 in #4684
\n– cmake: prosper_add_gtest keeps list-valued test properties intact (#4682) by @mattias800 in #4690
\n– gpu: a resource-order refusal clears the colour-target stats too (#4687) by @mattias800 in #4692
\n– Kena: zero refused fragment programs (AGC helper no-effect before shader work; counted-loop VCC_HI counter) by @mattias800 in #4688
\n– perf: convert RGB10A2 compute results with a lookup table into a pooled buffer by @ymatrix1 in #4691
\n– present: read a display buffer’s guest bytes in its VideoOut pixel format (#4686) by @mattias800 in #4689
\n– present: guest-order DMA reads of ALT targets; name format declines; count the scanout swap (#4686 follow-up) by @mattias800 in #4695
\n– test: sequence schedstat_probe fixture on a virtual clock, drop RUN_SERIAL (#4142) by @mattias800 in #4697
\n– docs: restore a dropped table cell separator in RECOMPILER_REMAINING.md by @mattias800 in #4698
\n– test: keep the schedstat probe docstring out of the ctest gate by @mattias800 in #4702
\n– gpu: [t8-dropped] keeps the words of an image descriptor the fold declines (Kena #4700) by @mattias800 in #4701
\n– diag: census of compute publications that repeat the previous bytes (F8 window, no switch) by @ymatrix1 in #4705
\n– docs: propose ADR 0029, first-use graphics pipelines by library link (refines ADR 0014) by @ymatrix1 in #4722
\n– docs: propose ADR 0032, page-granular write tracking on Windows hosts by @ymatrix1 in #4724
\n– Kena past New Game: ldexp CLAMP and entry-M0 KILL; translator reject in the refused index by @mattias800 in #4711
\n– docs: propose ADR 0028, one per-shader policy for Wave64 programs on 32-lane hosts by @ymatrix1 in #4717
\n– docs: propose ADR 0030, a descriptor heap behind one binding interface by @ymatrix1 in #4721
\n– docs: propose ADR 0031, GPU-resident detile, retile and conversion by @ymatrix1 in #4723
\n– docs: propose ADR 0027, one phi-aware descriptor provenance analysis with a runtime fallback by @ymatrix1 in #4716
\n– docs: ADR 0028\/0030\/0031 review nits that missed their merges by @ymatrix1 in #4729
\n– gpu: implement the guest depth-bounds test (DB_DEPTH_CONTROL.DEPTH_BOUNDS_ENABLE) by @mattias800 in #4704
\n– texture: fast-cleared one- and two-component DCC surfaces sample the clear colour by @ymatrix1 in #4699
\n– docs: propose ADR 0033, GPU reads guest buffers in place (zero-copy) by @ymatrix1 in #4726
\n– docs: propose ADR 0034, guest async compute on a dedicated host queue by @ymatrix1 in #4727
\n– docs: propose ADR 0035, indirect arguments resolved on the GPU by @ymatrix1 in #4728
\n– recompiler: Wave64 fragment scalar-pair projection behind a definite-assignment proof (#4706 fix 2) by @mattias800 in #4725
\n– capture: record a draw buffer the live source gate refuses as a placeholder (#3807) by @mattias800 in #4707
\n– gpu: honour SPI_SHADER_COL_FORMAT and integer colour targets (Kena’s sun, #4703) by @mattias800 in #4715
\n– recompiler: refuse the projection in unreached dispatcher cases; compare results are ballots (Kena f1d1baa8 live) by @mattias800 in #4730
\n– perf: mirror packed R10G10B10A2 compute results into the renderer on the GPU by @ymatrix1 in #4709
\n– nid_census: name NIDs from a flat community database, not only the firmware dump by @ymatrix1 in #4731
\n– gpu: indexed and instanced merged-NGG draws (#3135 P6) by @mattias800 in #4733
\n– Kena title scene: fog gets its shadows (UNORM16 depth to compute) and foliage draws (POS_W is clip w) by @mattias800 in #4734
\n– gpu: admit Kena’s downstream merged-NGG producers: immediate-offset selector source, s0:s1 user-data address, nested EXEC saves (#3135) by @mattias800 in #4735
\n– recompiler: carry v_writelane spill slots through structured loops and if merges (Kena’s level-load device loss) by @mattias800 in #4740
\n– agc: implement sceAgcDriverSubmitMultiAcbs, folded into the queue’s compute context by @ymatrix1 in #4742
\n– console_oracle: check the HLE against a real PS5, and fix what it finds by @ymatrix1 in #4744
\n– Kena title scene: depth of field works (fold restores across compacted targets; narrow FP16 sampled natively) by @mattias800 in #4736
\n– agc: SubmitAcb keys its queue on 32 bits; SubmitMultiDcbs answers an empty batch by @ymatrix1 in #4748
\n– nid names: choose the verified flat-db candidate; label a missing source honestly by @ymatrix1 in #4756
\n– docs: accept ADR 0028 (Wave64 on 32-lane hosts) by @mattias800 in #4761
\n– console_oracle: measure the kernel and user service families, and fix what they find by @ymatrix1 in #4747
\n– diagnostics: route= field and per-route Wave64 counters (ADR 0028 step 2) by @ymatrix1 in #4754
\n– recompiler: ADR 0028 step 1, the s_cbranch_execz any-vote certificate by @ymatrix1 in #4755
\n– docs: propose ADR 0036, link allowlisted local-compute system libraries from the user’s firmware dump by @ymatrix1 in #4764
\n– recompiler: classify Wave64 compute cross-lane ops and name the route (ADR 0028 step 3, analysis) by @ymatrix1 in #4760
\n– console_oracle: measure the AGC packet-size queries, and answer the ones prosper left at 0 by @ymatrix1 in #4759
\n– recompiler: PROSPER_WAVE64_EXCHANGE retries a narrow-host-refused Wave64 compute module through the exact dispatcher (ADR 0028 step 3) by @ymatrix1 in #4763
\n– console_oracle: a second destroy is EINVAL, and sceUserServiceInitialize2 validates its priority by @ymatrix1 in #4771
\n– recompiler: drop the cached PROSPER_DBG tag line that reddened main by @ymatrix1 in #4777
\n– gpu: NGG without a GS through the subgroup shell; a layered draw needs every attachment proven one slice (#3135 P7) by @mattias800 in #4750
\n– recompiler: a MUBUF format store writes only its V# format’s components; Kena exposure settled by @mattias800 in #4773
\n– console_oracle: measure the AGC builders, libkernel, POSIX names and net\/http helpers by @ymatrix1 in #4780
\n– test: per-platform baseline row for kx_attr_getstack_default (fixes red macOS CI on main) by @mattias800 in #4785
\n– docs: Kena’s distance-field AO is empty because the mesh-SDF atlas upload is skipped (#4766, paused) by @mattias800 in #4788
\n– hle\/fs: a guest path outside every mount never reaches the host filesystem by @mattias800 in #4784
\n– tools: local save fixtures and a snapshot.py save_fixture field by @mattias800 in #4789
\n– gpu: replay a layered depth-only NGG draw once per depth slice (#3135, Kena shadow cubes) by @mattias800 in #4778
\n– gpu: memoize the compute nested wide-data inventory per program (Dawnwalker +44% presented fps) by @mattias800 in #4794
\n– fix(tools): report reconstruction size in bytes by @AaravMehta-07 in #4797
\n– tools: skip split_file’s CLI self-test arm on Windows (fixes red MinGW on main) by @mattias800 in #4798
\n– tests: materialize deep_scan’s pad frame so DmemCallerWalk.Contract passes under AppleClang -O3 by @esdras-kyo in #4800
\n– gpu: GPU-retile compute-written SW_64KB_R_X volumes (Kena compute writeback -20%); Stage 1 ceiling measured by @mattias800 in #4804
\n– test: multi-block-y volume shape for the GPU retile byte-equal test (#4804 review gap) by @mattias800 in #4807
\n– docs: 2026-10-09 status sweep \u2014 three new titles; PGA TOUR 2K25 and Beast of Reincarnation reach title screens by @mattias800 in #4806
\n– gpu: bind a null image for an unusable T# read only by a skippable instruction (deterministic admission) by @mattias800 in #4801
\n– gpu: lower image_sample_cd like image_sample_d; dump backend-declined compute programs (#4808) by @mattias800 in #4813
\n– tools: extract nested AMDGPU shader ELFs from game eboots by @ymatrix1 in #4812
\n– console_oracle: add offline coverage report tool and test suite by @ymatrix1 in #4809
\n– tools: add console_capture client and tests for ps5debug-NG by @ymatrix1 in #4817
\n– tools: console_capture breakpoint opt-in, bounded replies, restored status-doc text (#4817 review) by @mattias800 in #4820
\n– docs: escape a pipe that broke a GRIS_SONIC_COBRA table row (main Docs red) by @ymatrix1 in #4821
\n– gpu: split-T# proof keeps provenance across EXEC saves and analyses large programs once by @ymatrix1 in #4712
\n– recompiler: admit direct-EXEC-write wave-empty guards in counted loops by @ymatrix1 in #4818
\n– gpu: decode PM4 IT_LOAD_SH_REG and IT_LOAD_CONTEXT_REG indirect register packets by @ymatrix1 in #4822
\n– recompiler: compute Wave64 scalar-pair projection refuses a fabricated-zero word (#4714) by @ymatrix1 in #4749
\n– adr: 0037 amends Wave64 routes with what steps 1-3 measured (proposed) by @ymatrix1 in #4776
\n– gpu: admit Yakuza Kiwami’s Wave32 and The Pathless’s merged NGG draws (#4808 cause #3) by @mattias800 in #4819
\n– gpu: view a borrowed compute image in its own format, not the renderer’s folded host format (#4810) by @mattias800 in #4832
\n– gpu: lower the NSA image_sample_c, image_sample_o and image_get_lod forms in fragments (#4808) by @mattias800 in #4835
\n– gpu: decode hardware PM4 IT_ACQUIRE_MEM and IT_RELEASE_MEM synchronization packets by @ymatrix1 in #4823
\n– docs: rewrite root README using PortPS5 structure as reference by @ymatrix1 in #4824
\n– gpu: decode hardware PM4 IT_DISPATCH_DIRECT and IT_DISPATCH_INDIRECT compute packets by @ymatrix1 in #4825
\n– gpu: decode hardware PM4 IT_DRAW_INDEX_AUTO, IT_DRAW_INDIRECT, IT_SET_BASE, and IT_WRITE_DATA packets by @ymatrix1 in #4826
\n– gpu: decode hardware PM4 IT_WAIT_REG_MEM polling packets by @ymatrix1 in #4827
\n– gpu: decode hardware PM4 IT_CLEAR_STATE, IT_INDEX_BUFFER_SIZE, IT_ATOMIC_MEM, and IT_PFP_SYNC_ME packets by @ymatrix1 in #4828
\n– gpu: decode hardware PM4 IT_COND_EXEC, IT_DRAW_INDEX_INDIRECT_MULTI, and IT_WAIT_REG_MEM64 packets by @ymatrix1 in #4829
\n– docs: record hardware PM4 packet decode coverage for Black Flag by @ymatrix1 in #4830
\n– tests: pin Black Flag 32^3 R8 3D mip-pyramid T# decode and view refusals by @ymatrix1 in #4833
\n– build: gate movexec guard test on Vulkan availability by @ymatrix1 in #4834
\n– tools\/rdna2_ref: independent RDNA2 reference interpreter with LLVM cross-check by @ymatrix1 in #4836
\n– gpu: record s_bfm_b64 architectural scalar data alongside lane mask (#4694) by @ymatrix1 in #4837
\n– gpu: step over non-SCC writers in scc_branch_is_workgroup_uniform by @ymatrix1 in #4839
\n– gpu: admit live EXEC mask source in scalar_source_projectable by @ymatrix1 in #4841
\n– recompiler: place thick-3D tail views and sample them at explicit LOD by @ymatrix1 in #4843
\n– recompiler: flip 3D tail refusal arms and gate storage paths (#4814 review follow-up) by @ymatrix1 in #4844
\n– gpu: admit a Wave64 VCC_LO cselect that reads a scalar-data VCC_HI (#4808) by @mattias800 in #4853
\n– gpu\/recompiler: a Wave64 B32 VCC write after an unrelated merge keeps its SCC (#4846) by @mattias800 in #4854
\n– hle: implement AMM unmap and command-buffer destruction by @mattias800 in #4856
\n– video: publish Videodec2 decoded plane pitches by @mattias800 in #4859
\n– gpu: let the raw-wide load proof admit loops whose back edges land after the load (#4847) by @mattias800 in #4862
\n– test: make reallocf release checks independent of allocator reuse by @mattias800 in #4867
\n– gpu: preserve chroma in separately allocated tiled NV12 planes by @mattias800 in #4863
\n– gpu\/recompiler: admit loops whose direct break leaves VCC dead at the exit (#4848) by @mattias800 in #4864
\n– gpu: route indexed user-config loads to the UC register file by @mattias800 in #4868\n<\/div>\n
\n– gpu: preserve independent logical64 WAITCNT completion by @mattias800 in #4372
\n– docs: list all eleven ratchet rules in the charter and state the platform-arm rule by @ymatrix1 in #4367
\n– docs: clarify APR proposal and preserve historical Dragon Quest observations by @ymatrix1 in #4365
\n– fix(gpu): retain producing raster launch controls through capture replay by @mattias800 in #4376
\n– ci: enforce pytest policy on new Python tests by @ymatrix1 in #4314
\n– Bound native shader walkers by authentic readable source windows by @mattias800 in #4377
\n– docs: record current GTA Stage 2 ABBA checkpoint by @mattias800 in #4379
\n– docs: record Sonic deferred-wait world falsification by @mattias800 in #4380
\n– docs: record GTA warm target-count checkpoint by @mattias800 in #4381
\n– feat(gpu): connect original scalar-buffer reads to shared wave64 bank by @mattias800 in #4374
\n– test: use SDK-relative pointers in scalar-bank fixtures by @mattias800 in #4382
\n– test: adversarial pins for rdna2 closed-tail block admission by @ymatrix1 in #4386
\n– test: LLVM oracle pins for SOP2 destination widths by @ymatrix1 in #4390
\n– test: follow v71 capture tail and Wave32 eager routing so main’s Linux CI is green by @mattias800 in #4396
\n– test: pin mandatory-literal K capture for VOP2 mul-adds and SOPK setreg by @ymatrix1 in #4401
\n– feat(gpu): implement VOP3P integer dot product lowering (v_dot2, v_dot4, v_dot8) by @ymatrix1 in #4400
\n– test: pin SOPP direct-branch set against the ISA table by @ymatrix1 in #4405
\n– test: oracle pins for scalar opcode predicates by @ymatrix1 in #4408
\n– feat(gpu): implement v_min3_i32, v_max3_i32, and v_med3_i32 lowering by @ymatrix1 in #4404
\n– feat(gpu): implement subword SAD instructions (v_sad_u8, v_sad_hi_u8, v_sad_u16, v_msad_u8) by @ymatrix1 in #4388
\n– feat(gpu): implement v_perm_b32 byte permutation lowering by @ymatrix1 in #4389
\n– feat(gpu): implement image_gather4 texture gather lowering by @ymatrix1 in #4392
\n– test: ISA-table pins for memory-writer classification by @ymatrix1 in #4393
\n– feat(gpu): implement v_div_fixup_f32 division fixup lowering by @ymatrix1 in #4391
\n– feat(gpu): implement s_movrels_b32 relative SGPR move lowering by @ymatrix1 in #4394
\n– perf: emit opt-in per-flip stage timing samples by @ymatrix1 in #4406
\n– feat(gpu): implement v_mad_i64_i32 lowering by @ymatrix1 in #4387
\n– gpu: bound s_movrels_b32’s constant M0 before adding it to the base by @mattias800 in #4413
\n– test: migrate normal_magnitude_predicate to GTest by @ymatrix1 in #4414
\n– test: extract minmax3 kernels from rdna2_to_spirv main (pilot split) by @ymatrix1 in #4415
\n– gpu: recover GTA V’s #4270 bank-scene regression (2.9 -> 6.5 fps) by @mattias800 in #4421
\n– feat(hle): implement libSceAudio3d\/AudioPropagation\/Acm spatial audio stubs by @ymatrix1 in #4418
\n– recompiler: stop phantom raw-wide refusals of register-offset descriptor loads (#4422) by @mattias800 in #4425
\n– feat(hle): implement libSceNet sockets and libSceNp offline stubs by @ymatrix1 in #4417
\n– gpu: seed BGRA render targets into compute storage on the device (+ lift live compute’s storage codec) by @mattias800 in #4428
\n– gpu: read back the span before an ordered DMA only when the copy touches it (Alex Kidd holds 60 fps) by @mattias800 in #4459
\n– docs: Alex Kidd plateau finding and async pair outcome by @mattias800 in #4461
\n– test: migrate 8 host\/loader\/HLE suites to GoogleTest by @ymatrix1 in #4430
\n– test: migrate the fragment float-flag identity suites to GoogleTest by @ymatrix1 in #4441
\n– test: migrate the select\/pselect pure-sleep suite to GoogleTest by @ymatrix1 in #4446
\n– test: migrate the guest-facing HLE handler contracts to GoogleTest by @ymatrix1 in #4434
\n– test: migrate the diagnostics enabled-path and JSON suites to GoogleTest by @ymatrix1 in #4449
\n– test: extract scalar-ALU kernels from rdna2_to_spirv main by @ymatrix1 in #4431
\n– test: migrate the pthread-key destructor suite to GoogleTest by @ymatrix1 in #4443
\n– test: extract VOP3 batch kernels from rdna2_to_spirv main by @ymatrix1 in #4432
\n– test: migrate the clock and timed-wait HLE contracts to GoogleTest by @ymatrix1 in #4437
\n– test(gpu): pin f64 transcendental refusal with rcp_f32 control by @ymatrix1 in #4453
\n– test(gpu): pin div-fmas refusal with fixup control by @ymatrix1 in #4454
\n– test(gpu): pin scalar-misc refusal with movk\/mov controls by @ymatrix1 in #4451
\n– test(gpu): pin float-dot refusal with integer-dot control by @ymatrix1 in #4456
\n– test(gpu): pin x2 buffer-atomic refusal with 32-bit control by @ymatrix1 in #4460
\n– test(gpu): pin SOPP halt\/kill refusal with s_nop control by @ymatrix1 in #4445
\n– test(gpu): pin 64-bit DS refusal with ds_add_u32 control by @ymatrix1 in #4448
\n– test(gpu): pin WRXCHG2 DS refusal with write2 control by @ymatrix1 in #4450
\n– test(gpu): pin multi-level IMAGE_LOAD_MIP refusal with plain-load control by @ymatrix1 in #4438
\n– test(gpu): pin NSA IMAGE_STORE_MIP refusal with consecutive-store control by @ymatrix1 in #4444
\n– feat(hle): register remaining dialog entry points (Login\/MsgProgress\/WebBrowser) by @ymatrix1 in #4458
\n– feat(hle): register Trophy2 icon getters alongside the info queries by @ymatrix1 in #4440
\n– feat(hle): register Http option setters, epoll lifecycle and Https setup by @ymatrix1 in #4442
\n– gpu: bound MSAA\/unusual-swizzle colour targets so ordered DMAs skip the span readback (Summer Sports) by @mattias800 in #4464
\n– feat(hle): register sceSaveDataLoadIcon completing the icon round-trip by @ymatrix1 in #4436
\n– feat(hle): register UDS DestroyHandle\/SetInt32 and sceSaveDataDelete by @ymatrix1 in #4433
\n– feat(hle): register remaining UDS setters, lifecycle and stats (25 exports) by @ymatrix1 in #4455
\n– feat(hle): register libSceNpWebApi2 request and push-event surface by @ymatrix1 in #4435
\n– feat(hle): implement libSceRtc calendar arithmetic (22 exports) by @ymatrix1 in #4447
\n– hle: fix np.cpp splice that broke the build (GetStorageStat body) by @mattias800 in #4468
\n– feat(hle): register libSceNpManager remainder (14 exports) by @ymatrix1 in #4476
\n– feat(hle): register notice-skip setters, PowerTick and privacy\/Init2 by @ymatrix1 in #4478
\n– feat(hle): report voice unavailable across the port surface by @ymatrix1 in #4473
\n– feat(hle): register MouseClose and pad mic-in-use vibration quirk by @ymatrix1 in #4471
\n– feat(hle): register Net epoll\/introspection\/resolver and Rudp setup by @ymatrix1 in #4482
\n– test(gpu): pin float image-atomic refusal with add control by @ymatrix1 in #4470
\n– test(gpu): pin arithmetic x2 buffer-atomic refusal with 32-bit control by @ymatrix1 in #4474
\n– test(gpu): pin swappc\/code-end refusal with mov\/nop controls by @ymatrix1 in #4475
\n– test(gpu): pin bfe64\/mode-set refusal with mov\/nop controls by @ymatrix1 in #4477
\n– test(gpu): pin endpgm-variant refusal with endpgm control by @ymatrix1 in #4479
\n– test(gpu): pin setreg-imm32 refusal with flat-scratch control by @ymatrix1 in #4480
\n– test(gpu): pin deferred 32-bit atomic refusal with add control by @ymatrix1 in #4481
\n– test(gpu): pin image_atomic_fcmpswap and image_atomic_dec refusals (supersedes #4472) by @mattias800 in #4488
\n– memory: a fixed direct map replaces tracked direct memory on both platforms (replaces #4189) by @ymatrix1 in #4486
\n– feat(hle): implement NGS2 allocator-create\/destroy\/grain\/lock lifecycle (6 exports) by @ymatrix1 in #4487
\n– feat(hle): implement Fiber\/Ult remainder (7 exports) by @ymatrix1 in #4491
\n– feat(hle): implement AvPlayer trick-speed\/bandwidth\/printf remainder (4 exports) by @ymatrix1 in #4490
\n– feat(hle): implement Font render-state remainder (13 exports) by @ymatrix1 in #4494
\n– feat(hle): implement GameUpdate request lifecycle (5 exports) by @ymatrix1 in #4497
\n– gpu: restore the Unity UI vertex shader (Greak, Alex Kidd, Summer Sports menus and HUDs) + fix the macOS build by @mattias800 in #4495
\n– gpu: Summer Sports’ track is the right colour again (BC6H lightmaps clamp until the pipeline is linear) by @mattias800 in #4501
\n– feat(hle): register AJM direct-job shapes (14 exports) by @ymatrix1 in #4484
\n– feat(hle): implement Ime text-entry session (11 exports) by @ymatrix1 in #4503
\n– feat(hle): implement libSceCesCs conversion and libScePngDec decode by @ymatrix1 in #4452
\n– hle: sceRtcGetTick is the same on every host (fixes the Rtc tests on macOS and Windows) by @mattias800 in #4505
\n– host(win): keep the host’s allocations out of the guest’s address range \u2014 the UE4 launch hang (#4426) by @mattias800 in #4507
\n– hle: re-derive NGS2\/AvPlayer\/Font contracts from the .native modules PS5 titles link by @mattias800 in #4524
\n– feat(hle): implement NGS2 waveform parse\/calc\/reset-option (3 exports) by @ymatrix1 in #4510
\n– hle(apr): a counter completion is never delivered below one already delivered \u2014 Kena’s boot crash (#4504) by @mattias800 in #4517
\n– docs(kena): Windows -> Linux\/AMD handoff of the title-screen 3D bring-up by @mattias800 in #4500
\n– gpu: Space Adventure Cobra’s in-game 3D renders again (two recompiler defects in its per-light loop) by @mattias800 in #4522
\n– recompiler: accept UE4’s save-then-v_cmpx EXEC guard around a counted loop (Kena, #4427) by @mattias800 in #4534
\n– gpu: close two raw-wide classifier gaps found in review (wrong relative-move opcodes, implicit VCC readers) by @mattias800 in #4535
\n– gpu: a loop exit drops the scratch a mask overwrote, for every loop by @mattias800 in #4542
\n– tools\/ci: decode git and ruff I\/O as UTF-8 in check_python_quality by @ymatrix1 in #4539
\n– feat(hle): implement Ult pool\/runtime destroys (2 exports) by @ymatrix1 in #4528
\n– recompiler: admit several bottom-tested SCC do-while loops in one program (#4518) by @mattias800 in #4546
\n– tools: static HLE progress census by @ymatrix1 in #4532
\n– diagnostics: say whether Wave64 was the GPU’s doing or prosper’s by @mattias800 in #4531
\n– docs: architecture spec and decision records (ADRs 0001-0021), gated by check_arch_docs by @ymatrix1 in #4536
\n– gpu: s_movrels_b32 reads a range of registers, and the liveness proofs now know it (#4538) by @mattias800 in #4547
\n– tools\/docs: unbreak the Docs job (check_arch_docs’ \u00ab\u00a0ctest\u00a0\u00bb kind is not a ctest invocation) by @mattias800 in #4561
\n– docs: add Rayman 30th Anniversary Edition and MOUSE: P.I. For Hire with their first captures by @mattias800 in #4558
\n– gpu: two leftovers from the #4547 review (raw-x2 escape list; dispatcher SCC after a relative read) (#4559) by @mattias800 in #4562
\n– gpu: keep the program a draw lost to the owned-wave gate, and the gate’s reason (#4555) by @mattias800 in #4565
\n– gpu: AGC’s eliminate-fast-clear rectangle writes no colour (#1588) by @mattias800 in #4568
\n– gpu: four raw-wide classifier gaps, and one emitter fix they depend on, that put GTA V’s and MOUSE’s fragment programs on the owned-wave path (#4555) by @mattias800 in #4569
\n– hle\/agc: sceAgcCbBranch builds its packet; the branch runs where the guest put it (#4540) by @mattias800 in #4576
\n– tools\/env: register every PROSPER_* switch and require new ones be classified by @ymatrix1 in #4543
\n– live compute: publish an RGB10A2 compute result into the renderer target by @ymatrix1 in #4571
\n– revision: see through a ccache\/sccache compiler wrapper in the dependency identity (#4356, #4187) by @mattias800 in #4581
\n– tools\/env: drop the stale PROSPER_AGC_SPLIT_FINAL_STATE registry row (main is red) by @ymatrix1 in #4582
\n– gpu: a clashed constant-buffer key is published per consumer, not per key by @mattias800 in #4584
\n– gpu: key a fold descriptor use the way the recompiler tags it (#4585) by @mattias800 in #4588
\n– docs: record that holding the EOP pulse does not stop DQ7’s label recycle (#2982) by @ymatrix1 in #4583
\n– gpu: an x2 load can be the latched source of a raw-wide register offset (#4578) by @mattias800 in #4587
\n– docs: blog Beast and DQ7 after the eliminate-fast-clear fix; record Kena’s refusal progress by @mattias800 in #4589
\n– feat(hle): implement Font StyleFrame cluster (15 exports) by @ymatrix1 in #4572
\n– docs: every document declares its kind; every relative link resolves by @ymatrix1 in #4567
\n– gpu: a register-offset raw-wide load’s pointer need only survive until the load; a branch before its source is harmless unless it lands inside by @mattias800 in #4593
\n– spv_validate: rdna2_proven_raw_register_wide_entry_loads is not an emitter (main red after #4593) by @mattias800 in #4595
\n– gpu: a base-zero T# with constant-zero selectors is the null image (#4592) by @mattias800 in #4596
\n– gpu: merged-NGG subgroup planner and launch values (#3135 P1) by @mattias800 in #4597
\n– gpu: fall back to the general CFG route when a counted loop’s prelude declines (Kena 0x5006fb0000) by @mattias800 in #4594
\n– revision: write the compiler-wrapper test in pytest conventions (Python quality red after #4581) by @mattias800 in #4601
\n– gpu: NGG evidence for the merged-NGG lowering (#3135 P0) by @mattias800 in #4599
\n– gpu: read PROSPER_DBG live for the null-image line (main red after #4596) by @mattias800 in #4604
\n– gpu: close three CFG-dispatcher Wave64 gaps so Kena’s compute 0x5008ec0000 recompiles by @mattias800 in #4603
\n– gpu: merged-NGG subgroup shell, compile and offline execution (#3135 P2) by @mattias800 in #4605
\n– gpu: publish an over-cap buffer window clamped to its mapped run by @ymatrix1 in #4566
\n– gpu: a plain copy of loaded words into EXEC, and a wide load into the special registers, are numeric uses in the raw-wide walk (#4574) by @mattias800 in #4579
\n– feat(hle): implement NGS2 system lock\/unlock\/sample-rate (3 exports) by @ymatrix1 in #4521
\n– feat(hle): implement Font kerning\/library\/rewind\/back-link (4 exports) by @ymatrix1 in #4520
\n– feat(hle): implement Font pixel-resolution\/cache\/library back-link (3 exports) by @ymatrix1 in #4548
\n– feat(hle): implement Font DPI\/glyph-form\/scale\/kerning queries (5 exports) by @ymatrix1 in #4570
\n– feat(hle): implement Font renderer-outline\/DPI\/rewind\/mask (6 exports) by @ymatrix1 in #4573
\n– recompiler: prove two-T# x16 bundles through image stores; disjoint stores keep the split-T# proof by @ymatrix1 in #4564
\n– feat(hle): implement Ult queues and semaphores (14 exports) by @ymatrix1 in #4544
\n– gpu: merged-NGG raster commit: pass-through vertex stage and layer routes (#3135 P3) by @mattias800 in #4611
\n– gpu: type CFG-dispatcher spill reloads, and keep a mask reloaded into its own register (#4600) by @mattias800 in #4607
\n– gpu: a 64-lane fragment launch is not sent to the owned-wave path for a load that is numeric only at 32 lanes (#4555) by @mattias800 in #4609
\n– docs: regenerate PROGRESS_TRACKER.md (MOUSE’s blockers after #4609) by @mattias800 in #4614
\n– gpu: lower ds_read_b32 … gds in compute, the read MOUSE’s per-frame counter copy was refused at (#4553) by @mattias800 in #4615
\n– docs: MOUSE: P.I. For Hire after #4609 and #4615: route, captures, and what is ruled out by @mattias800 in #4616
\n– vk: run merged-NGG draws in the backend: fill, shell dispatches, pass-through runs, no CPU wait (#3135 P4) by @mattias800 in #4617
\n– gpu: let replay admit a conditional branch out of the program (MOUSE frames could not be replayed) by @mattias800 in #4622
\n– gpu: MOUSE’s world renders: a DCC clear reaches the target a pass renders to, and a Z_16 depth plane keeps its size by @mattias800 in #4621
\n– docs: MOUSE’s second depth target is refused before gameplay only (correcting #4621’s docs and #4620) by @mattias800 in #4626
\n– gpu: replay asks whether a write lands in a retained depth plane by address (replayed frames lost every depth target) by @mattias800 in #4628
\n– gpu: reuse a draw’s render state while its registers are unchanged by @mattias800 in #4630
\n– gpu: BOOL64 predicated jumps run on a non-zero condition (DQ7 title clouds\/sea) by @mattias800 in #4610
\n– compute: honour a fast-cleared target’s control plane and a padded linear image’s row pitch by @ymatrix1 in #4586
\n– test: the savedata tmpfs check names the old fixed roots, not any \/tmp prefix by @mattias800 in #4640
\n– docs: Tomb Raider’s rung-3 manor is Tomb Raider II’s; Tomb Raider I’s does not render by @mattias800 in #4644
\n– gpu: borrow the readable start of a draw buffer declared past mapped memory by @mattias800 in #4633
\n– gpu: run merged ES+GS NGG draws live: admission, linked fold, cached producer (#3135 P5) by @mattias800 in #4634
\n– gpu: publish a renderer-claimed volume before compute reads it (#4625) by @mattias800 in #4651
\n– gpu: count draw-disposition seen at the pass entry so silent refusals raise unaccounted-draws by @mattias800 in #4655
\n– gpu: refuse a cached render target for a sampled view needing more bytes per texel (AC Black Flag red frame, #4197) by @ymatrix1 in #4648
\n– Sink vote-lowering diagnostic types to src\/diagnostics by @ymatrix1 in #4646
\n– windows: read the SSE4a-log and stub-dump switches through one helper each (getenv 14 to 10) by @ymatrix1 in #4639
\n– Move shared diagnostics helpers down to src\/diagnostics by @ymatrix1 in #4641
\n– ci: keep reviewer notes through arch-ratchet –update by @ymatrix1 in #4636
\n– Move COMPUTELOG diagnostics out of gpu_executor.cpp by @ymatrix1 in #4642
\n– compute: move execute_item’s post-wait tail into finish_dispatch_tail (preparatory for ADR 0009, proposed) by @ymatrix1 in #4650
\n– fix(prosper-app): list and boot dumps at drive roots by @ymatrix1 in #4658
\n– ADR 0009 Stage 1: design note and a tested retirement-order contract by @ymatrix1 in #4645
\n– docs: Black Flag’s measured frame cost and the bounded-buffer null result by @ymatrix1 in #4632
\n– hle: NetCtl\/Np state callbacks on every host; sceVoiceQoSInit refuses honestly by @ymatrix1 in #4577
\n– feat(prosper-app): Prosper window title and baked-in mark icon by @ymatrix1 in #4659
\n– ratchet: fix hle_kernel.cpp file-size row (5561 -> 5555) by @ymatrix1 in #4662
\n– ci: fix main’s red Docs and Python quality jobs (note frontmatter, stale switch row) by @ymatrix1 in #4661
\n– test: NGG shell-pipeline cache eviction no longer depends on native Wave64 by @mattias800 in #4663
\n– fix: retry transient Windows replace failures saving the pipeline cache by @ymatrix1 in #4667
\n– test: Windows cold-target repeat honours PROSPER_NO_SKIP_SEED by @ymatrix1 in #4666
\n– feat(prosper-app): game list with search, menus and controls view by @ymatrix1 in #4665
\n– ADR 0009 step 2: observe what pipelining would allow (PROSPER_PIPELINE_OBSERVE) by @ymatrix1 in #4664
\n– feat(prosper-app): persisted host settings for library-started boots by @ymatrix1 in #4670
\n– docs(prosper-app): add library view screenshot by @ymatrix1 in #4673
\n– compute: refresh a cached buffer by differing 64 KiB blocks (with a control switch) by @ymatrix1 in #4660
\n– compute: mirror R8Unorm storage results on the GPU (no CPU round trip) by @ymatrix1 in #4656
\n– feat(prosper-app): use Explorer folder picker on Windows by @ymatrix1 in #4672
\n– ci: exempt bench_*.cpp from the gtest policy; keep the NGG cache test’s push range valid by @mattias800 in #4675
\n– feat(prosper-app): add Game Log panel with bounded ring capture by @ymatrix1 in #4671
\n– compute: share one resident arena between overlapping read-only buffer windows (interim, not ADR 0010) by @ymatrix1 in #4635
\n– ci: spv_validate’s coverage scan exempts parent_scan.hpp’s two local vectors by @mattias800 in #4676
\n– test: compute buffer runtime tests expect #4660’s block-sync refresh, and detect a stale far block by @mattias800 in #4678
\n– feat(prosper-app): picker answers join the library; folders accumulate by @ymatrix1 in #4674
\n– perf: read hot-path diagnostic switches once per submit, not once per call by @ymatrix1 in #4677
\n– docs(kena): the title-menu world renders on main; #3835’s NaN half image was AGC’s helper rectangle by @mattias800 in #4679
\n– docs, tools: a path convention for screenshots, enforced for new captures by @ymatrix1 in #4629
\n– charter: point the architecture rules at docs\/spec and docs\/adr by @ymatrix1 in #4541
\n– docs, tools: architecture follow-ups from the PortPS5 docs comparison by @ymatrix1 in #4557
\n– diag: validation mapping census (which memory backs full compares) + GetWriteWatch ruled out by @ymatrix1 in #4681
\n– gpu: render layered volume passes with more than one colour target (#4643) by @mattias800 in #4685
\n– diag: always-on validation census with a size histogram; drop PROSPER_VALIDATION_MAPPING_CENSUS by @ymatrix1 in #4684
\n– cmake: prosper_add_gtest keeps list-valued test properties intact (#4682) by @mattias800 in #4690
\n– gpu: a resource-order refusal clears the colour-target stats too (#4687) by @mattias800 in #4692
\n– Kena: zero refused fragment programs (AGC helper no-effect before shader work; counted-loop VCC_HI counter) by @mattias800 in #4688
\n– perf: convert RGB10A2 compute results with a lookup table into a pooled buffer by @ymatrix1 in #4691
\n– present: read a display buffer’s guest bytes in its VideoOut pixel format (#4686) by @mattias800 in #4689
\n– present: guest-order DMA reads of ALT targets; name format declines; count the scanout swap (#4686 follow-up) by @mattias800 in #4695
\n– test: sequence schedstat_probe fixture on a virtual clock, drop RUN_SERIAL (#4142) by @mattias800 in #4697
\n– docs: restore a dropped table cell separator in RECOMPILER_REMAINING.md by @mattias800 in #4698
\n– test: keep the schedstat probe docstring out of the ctest gate by @mattias800 in #4702
\n– gpu: [t8-dropped] keeps the words of an image descriptor the fold declines (Kena #4700) by @mattias800 in #4701
\n– diag: census of compute publications that repeat the previous bytes (F8 window, no switch) by @ymatrix1 in #4705
\n– docs: propose ADR 0029, first-use graphics pipelines by library link (refines ADR 0014) by @ymatrix1 in #4722
\n– docs: propose ADR 0032, page-granular write tracking on Windows hosts by @ymatrix1 in #4724
\n– Kena past New Game: ldexp CLAMP and entry-M0 KILL; translator reject in the refused index by @mattias800 in #4711
\n– docs: propose ADR 0028, one per-shader policy for Wave64 programs on 32-lane hosts by @ymatrix1 in #4717
\n– docs: propose ADR 0030, a descriptor heap behind one binding interface by @ymatrix1 in #4721
\n– docs: propose ADR 0031, GPU-resident detile, retile and conversion by @ymatrix1 in #4723
\n– docs: propose ADR 0027, one phi-aware descriptor provenance analysis with a runtime fallback by @ymatrix1 in #4716
\n– docs: ADR 0028\/0030\/0031 review nits that missed their merges by @ymatrix1 in #4729
\n– gpu: implement the guest depth-bounds test (DB_DEPTH_CONTROL.DEPTH_BOUNDS_ENABLE) by @mattias800 in #4704
\n– texture: fast-cleared one- and two-component DCC surfaces sample the clear colour by @ymatrix1 in #4699
\n– docs: propose ADR 0033, GPU reads guest buffers in place (zero-copy) by @ymatrix1 in #4726
\n– docs: propose ADR 0034, guest async compute on a dedicated host queue by @ymatrix1 in #4727
\n– docs: propose ADR 0035, indirect arguments resolved on the GPU by @ymatrix1 in #4728
\n– recompiler: Wave64 fragment scalar-pair projection behind a definite-assignment proof (#4706 fix 2) by @mattias800 in #4725
\n– capture: record a draw buffer the live source gate refuses as a placeholder (#3807) by @mattias800 in #4707
\n– gpu: honour SPI_SHADER_COL_FORMAT and integer colour targets (Kena’s sun, #4703) by @mattias800 in #4715
\n– recompiler: refuse the projection in unreached dispatcher cases; compare results are ballots (Kena f1d1baa8 live) by @mattias800 in #4730
\n– perf: mirror packed R10G10B10A2 compute results into the renderer on the GPU by @ymatrix1 in #4709
\n– nid_census: name NIDs from a flat community database, not only the firmware dump by @ymatrix1 in #4731
\n– gpu: indexed and instanced merged-NGG draws (#3135 P6) by @mattias800 in #4733
\n– Kena title scene: fog gets its shadows (UNORM16 depth to compute) and foliage draws (POS_W is clip w) by @mattias800 in #4734
\n– gpu: admit Kena’s downstream merged-NGG producers: immediate-offset selector source, s0:s1 user-data address, nested EXEC saves (#3135) by @mattias800 in #4735
\n– recompiler: carry v_writelane spill slots through structured loops and if merges (Kena’s level-load device loss) by @mattias800 in #4740
\n– agc: implement sceAgcDriverSubmitMultiAcbs, folded into the queue’s compute context by @ymatrix1 in #4742
\n– console_oracle: check the HLE against a real PS5, and fix what it finds by @ymatrix1 in #4744
\n– Kena title scene: depth of field works (fold restores across compacted targets; narrow FP16 sampled natively) by @mattias800 in #4736
\n– agc: SubmitAcb keys its queue on 32 bits; SubmitMultiDcbs answers an empty batch by @ymatrix1 in #4748
\n– nid names: choose the verified flat-db candidate; label a missing source honestly by @ymatrix1 in #4756
\n– docs: accept ADR 0028 (Wave64 on 32-lane hosts) by @mattias800 in #4761
\n– console_oracle: measure the kernel and user service families, and fix what they find by @ymatrix1 in #4747
\n– diagnostics: route= field and per-route Wave64 counters (ADR 0028 step 2) by @ymatrix1 in #4754
\n– recompiler: ADR 0028 step 1, the s_cbranch_execz any-vote certificate by @ymatrix1 in #4755
\n– docs: propose ADR 0036, link allowlisted local-compute system libraries from the user’s firmware dump by @ymatrix1 in #4764
\n– recompiler: classify Wave64 compute cross-lane ops and name the route (ADR 0028 step 3, analysis) by @ymatrix1 in #4760
\n– console_oracle: measure the AGC packet-size queries, and answer the ones prosper left at 0 by @ymatrix1 in #4759
\n– recompiler: PROSPER_WAVE64_EXCHANGE retries a narrow-host-refused Wave64 compute module through the exact dispatcher (ADR 0028 step 3) by @ymatrix1 in #4763
\n– console_oracle: a second destroy is EINVAL, and sceUserServiceInitialize2 validates its priority by @ymatrix1 in #4771
\n– recompiler: drop the cached PROSPER_DBG tag line that reddened main by @ymatrix1 in #4777
\n– gpu: NGG without a GS through the subgroup shell; a layered draw needs every attachment proven one slice (#3135 P7) by @mattias800 in #4750
\n– recompiler: a MUBUF format store writes only its V# format’s components; Kena exposure settled by @mattias800 in #4773
\n– console_oracle: measure the AGC builders, libkernel, POSIX names and net\/http helpers by @ymatrix1 in #4780
\n– test: per-platform baseline row for kx_attr_getstack_default (fixes red macOS CI on main) by @mattias800 in #4785
\n– docs: Kena’s distance-field AO is empty because the mesh-SDF atlas upload is skipped (#4766, paused) by @mattias800 in #4788
\n– hle\/fs: a guest path outside every mount never reaches the host filesystem by @mattias800 in #4784
\n– tools: local save fixtures and a snapshot.py save_fixture field by @mattias800 in #4789
\n– gpu: replay a layered depth-only NGG draw once per depth slice (#3135, Kena shadow cubes) by @mattias800 in #4778
\n– gpu: memoize the compute nested wide-data inventory per program (Dawnwalker +44% presented fps) by @mattias800 in #4794
\n– fix(tools): report reconstruction size in bytes by @AaravMehta-07 in #4797
\n– tools: skip split_file’s CLI self-test arm on Windows (fixes red MinGW on main) by @mattias800 in #4798
\n– tests: materialize deep_scan’s pad frame so DmemCallerWalk.Contract passes under AppleClang -O3 by @esdras-kyo in #4800
\n– gpu: GPU-retile compute-written SW_64KB_R_X volumes (Kena compute writeback -20%); Stage 1 ceiling measured by @mattias800 in #4804
\n– test: multi-block-y volume shape for the GPU retile byte-equal test (#4804 review gap) by @mattias800 in #4807
\n– docs: 2026-10-09 status sweep \u2014 three new titles; PGA TOUR 2K25 and Beast of Reincarnation reach title screens by @mattias800 in #4806
\n– gpu: bind a null image for an unusable T# read only by a skippable instruction (deterministic admission) by @mattias800 in #4801
\n– gpu: lower image_sample_cd like image_sample_d; dump backend-declined compute programs (#4808) by @mattias800 in #4813
\n– tools: extract nested AMDGPU shader ELFs from game eboots by @ymatrix1 in #4812
\n– console_oracle: add offline coverage report tool and test suite by @ymatrix1 in #4809
\n– tools: add console_capture client and tests for ps5debug-NG by @ymatrix1 in #4817
\n– tools: console_capture breakpoint opt-in, bounded replies, restored status-doc text (#4817 review) by @mattias800 in #4820
\n– docs: escape a pipe that broke a GRIS_SONIC_COBRA table row (main Docs red) by @ymatrix1 in #4821
\n– gpu: split-T# proof keeps provenance across EXEC saves and analyses large programs once by @ymatrix1 in #4712
\n– recompiler: admit direct-EXEC-write wave-empty guards in counted loops by @ymatrix1 in #4818
\n– gpu: decode PM4 IT_LOAD_SH_REG and IT_LOAD_CONTEXT_REG indirect register packets by @ymatrix1 in #4822
\n– recompiler: compute Wave64 scalar-pair projection refuses a fabricated-zero word (#4714) by @ymatrix1 in #4749
\n– adr: 0037 amends Wave64 routes with what steps 1-3 measured (proposed) by @ymatrix1 in #4776
\n– gpu: admit Yakuza Kiwami’s Wave32 and The Pathless’s merged NGG draws (#4808 cause #3) by @mattias800 in #4819
\n– gpu: view a borrowed compute image in its own format, not the renderer’s folded host format (#4810) by @mattias800 in #4832
\n– gpu: lower the NSA image_sample_c, image_sample_o and image_get_lod forms in fragments (#4808) by @mattias800 in #4835
\n– gpu: decode hardware PM4 IT_ACQUIRE_MEM and IT_RELEASE_MEM synchronization packets by @ymatrix1 in #4823
\n– docs: rewrite root README using PortPS5 structure as reference by @ymatrix1 in #4824
\n– gpu: decode hardware PM4 IT_DISPATCH_DIRECT and IT_DISPATCH_INDIRECT compute packets by @ymatrix1 in #4825
\n– gpu: decode hardware PM4 IT_DRAW_INDEX_AUTO, IT_DRAW_INDIRECT, IT_SET_BASE, and IT_WRITE_DATA packets by @ymatrix1 in #4826
\n– gpu: decode hardware PM4 IT_WAIT_REG_MEM polling packets by @ymatrix1 in #4827
\n– gpu: decode hardware PM4 IT_CLEAR_STATE, IT_INDEX_BUFFER_SIZE, IT_ATOMIC_MEM, and IT_PFP_SYNC_ME packets by @ymatrix1 in #4828
\n– gpu: decode hardware PM4 IT_COND_EXEC, IT_DRAW_INDEX_INDIRECT_MULTI, and IT_WAIT_REG_MEM64 packets by @ymatrix1 in #4829
\n– docs: record hardware PM4 packet decode coverage for Black Flag by @ymatrix1 in #4830
\n– tests: pin Black Flag 32^3 R8 3D mip-pyramid T# decode and view refusals by @ymatrix1 in #4833
\n– build: gate movexec guard test on Vulkan availability by @ymatrix1 in #4834
\n– tools\/rdna2_ref: independent RDNA2 reference interpreter with LLVM cross-check by @ymatrix1 in #4836
\n– gpu: record s_bfm_b64 architectural scalar data alongside lane mask (#4694) by @ymatrix1 in #4837
\n– gpu: step over non-SCC writers in scc_branch_is_workgroup_uniform by @ymatrix1 in #4839
\n– gpu: admit live EXEC mask source in scalar_source_projectable by @ymatrix1 in #4841
\n– recompiler: place thick-3D tail views and sample them at explicit LOD by @ymatrix1 in #4843
\n– recompiler: flip 3D tail refusal arms and gate storage paths (#4814 review follow-up) by @ymatrix1 in #4844
\n– gpu: admit a Wave64 VCC_LO cselect that reads a scalar-data VCC_HI (#4808) by @mattias800 in #4853
\n– gpu\/recompiler: a Wave64 B32 VCC write after an unrelated merge keeps its SCC (#4846) by @mattias800 in #4854
\n– hle: implement AMM unmap and command-buffer destruction by @mattias800 in #4856
\n– video: publish Videodec2 decoded plane pitches by @mattias800 in #4859
\n– gpu: let the raw-wide load proof admit loops whose back edges land after the load (#4847) by @mattias800 in #4862
\n– test: make reallocf release checks independent of allocator reuse by @mattias800 in #4867
\n– gpu: preserve chroma in separately allocated tiled NV12 planes by @mattias800 in #4863
\n– gpu\/recompiler: admit loops whose direct break leaves VCC dead at the exit (#4848) by @mattias800 in #4864
\n– gpu: route indexed user-config loads to the UC register file by @mattias800 in #4868\n<\/div>\n
<\/p>\n