aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/buffer_cache
Commit message (Collapse)AuthorAgeFilesLines
* nx2: fewer render pass splits on tiled GPUs, log failed GPU waits (Pokopia) ↵Alex21 hours1-1/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | (#361) * buffer_cache: don't end the render pass after mapped buffer uploads Every upload set any_buffer_uploaded, and the next draw or dispatch then recorded a transfer-to-shader barrier outside the render pass. On Vulkan the uploads are mapped: each copy either records its own barriers (CopyBuffer with barrier set) or goes to the upload command buffer, which ends with a barrier before the frame's commands run. The extra barrier only ended the render pass. On tiled GPUs every such split writes the attachments out and reads them back. Pokopia's first gameplay scene had ~50 of these splits a frame. Without them, an M3 MacBook Air (MoltenVK, 1x docked) goes from ~35 to ~29.5 ms of GPU time a frame and from 28.5 to 32 fps. OpenGL uploads immediately and keeps the barrier. Assisted-by: Claude Opus 5.5 (profiling, implementation and commit message) * vulkan: log a failed GPU wait before it ends the process When the device is lost, vkWaitSemaphores in MasterSemaphore::Wait returns VK_ERROR_DEVICE_LOST and the exception goes uncaught on the fence thread or the GPU thread, so the process aborts without a line in the log: ReportLoss is only reached from submit and present. A Pokopia crash on macOS left the log ending a second before the abort, with no hint that the GPU was lost. Log the wait's result before rethrowing. On macOS, MoltenVK prints the Metal reason (a GPU hang or page fault) to stderr only. Assisted-by: Claude Opus 5.5 (diagnosis, implementation and commit message)
* Merge pull request #334 from snowstormxd/codex/upstream-mkw-testJonas29 hours2-4/+150
|\ | | | | MKW performance improvements
| * nx2: rearm clean read areas when the GPU writes them againsnowstormxd30 hours1-2/+11
| | | | | | | | | | | | | | Port the coherency changes from alarok/Eglantine 77fc365f2 without its diagnostic framework or shutdown changes. Only GPU-dirty CPU reads become permanently preflushable, clean areas can be trapped on the next write, non-core readers avoid indexing core arrays, and collected CPU writes precede readback. Assisted-by: Codex (source review, selective integration and verification) (cherry picked from commit b64e211957180aa218ec6d9a3bcb020332279c3f)
| * video_core/buffer_cache: reuse unchanged small uniform buffers within ↵gabrielskaminski32 hours2-2/+139
| | | | | | | | | | | | | | command buffer (cherry picked from commit 4aa5cd89fd1862751b3297036f2d21a0dc924b23) (cherry picked from commit f1bdd5b13a1bdca81243b4aa1d8529f0485d3671)
* | Merge pull request #337 from Sssnowyyy/nx2-review-fixesJonas29 hours2-1/+47
|\ \ | | | | | | nx2: fix pokopia regressions, textures that handle shader computes, and more
| * | buffer_cache: fix the already-disabled check for uniform buffersSssnowyyy31 hours1-1/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | DisableGraphicsUniformBuffer tested !current.buffer_id to detect a binding that is already NULL_BINDING. SlotId converts to true unless it is the invalid id, and NULL_BUFFER_ID is BufferId{0}, so the check was false for NULL_BINDING and the early return never triggered. Compare against NULL_BUFFER_ID instead. Follow-up to 46b2577bf7. Assisted-by: Claude Sonnet 5.5 (implementation and commit message)
| * | vulkan: repack 69-byte vertices only when neededSssnowyyy31 hours2-0/+45
| |/ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | SetVertexRepackLayout returned old_mask | new_mask, which stays non-zero on every draw that uses a 69-byte binding. PrepareDraw then re-dirtied the vertex buffers and vertex input each time, so the repack compute pass ran and the render pass ended once per draw, which is costly on tiled GPUs. Return old_mask ^ new_mask so only a layout change re-dirties the bindings. The repacked copy only stays valid while the source data is unchanged, so the buffer cache now flags the vertex buffers overlapping a CPU write, inline or DMA write, or GPU write, and re-dirties the repacked ones before binding. Command buffer submission already invalidates all vertex buffer state, which keeps the staging copy within its tick. Also pass the source buffer size to the repack pass and clamp the word-rounded storage descriptor range to it; it could run up to three bytes past the end of the buffer. A binding that does not fit is skipped like one over the device limits. Follow-up to e38b738b84. Assisted-by: Claude Sonnet 5.5 (implementation and commit message)
* | Merge pull request #113 from ymatrix1/fix/storage-buffers-32Jonas29 hours1-1/+2
|\ \ | |/ |/| buffer_cache: allow 32 storage buffers per stage
| * buffer_cache: allow 32 storage buffers per stageYuri Olive36 hours1-1/+2
| | | | | | | | | | | | | | | | | | Native Switch 2 compute shaders (such as Donkey Kong Bananza) bind up to 18 global memory ranges, causing out-of-bounds binding errors: HW.GPU <Error> buffer_cache.h:BindComputeStorageBuffer: Storage buffer index 16 exceeds maximum storage buffer count HW.GPU <Error> buffer_cache.h:BindComputeStorageBuffer: Storage buffer index 17 exceeds maximum storage buffer count followed by GPU driver access violations (0xc0000005). The enabled and written storage buffer bitmasks (enabled_compute_storage_buffers and written_compute_storage_buffers) are already u32, so widening NUM_STORAGE_BUFFERS from 16 to 32 fits naturally within the existing 32-bit masks.
* | nx2: read GPU-written command lists from copies instead of drainingTakoroka35 hours2-0/+242
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The trees fix (#52) drains the split and flushes for every command list on a GPU-written page: in a Mario Kart World race ~13,500 lists and ~400 GPU round trips a second, during which the back has nothing to do. The replay ran at ~40 fps instead of ~45. The front now marks the pages those lists are on. The buffer cache notes writes to them; after a compute dispatch or DMA copy that writes one it copies the page into host memory and submits. The front waits for the back only up to the last compute or copy work it sent (not everything queued), then reads the copy once the GPU is done with it. Pages the GPU wrote once, long before, and the CPU now writes lists to have no copy: their first list drains and flushes the whole page, which is then read from guest memory until GPU work writes it again. CPU writes and inline uploads reach guest memory, so they keep such a page synced but make a copy stale. A page a draw writes always drains. Mario Kart World time-trial replay, Radeon 780M: ~40 -> ~45 fps, trees drawn, no list drains left.
* | Merge nx2 into nx2-experimentalkraibse38 hours1-3/+40
|\| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Resolved by area: - audio feature_support.h: nx2's native revision handling. - fiber.cpp: ours, which also drops the unused rewind stack. - host_memory.cpp: ours granule bitmap and HintCold, plus nx2's fallocate hole punching per whole dirty granule on Linux. - texture_cache.h: nx2's old-age eviction pass, then our Collect(). - kepler_compute.cpp: shared upscale slots, then Fortune's Weave's, then the Kirby Air Riders scan. - nx2_gsv.h: ours, plus nx2's GetBufferMaxResolution and graphics device ids (1-based). - nvhost_ctrl_gpu.cpp: engine support flag bits 40-44. - lower_to_maxwell.h: ours as the base, with nx2's seven lowering commits applied one at a time. Second texture descriptor buffers are read bindless in TEX, TLD.SCR (level zero) and SUST; FSEL keeps its operand modifiers; TXD takes any channel mask. lower_to_maxwell.h lowers 837 Fortune's Weave and 1183 Mario Kart World shader dumps with the same rejects as nx2 and ours (12 and 0). Assisted-by: Claude Sonnet 5.5 (conflict analysis and drafting)
| * Linux and AMD: BC3 flashing on RADV, ASTC decode stalls, lower RAM usage, a ↵Whovian6444 hours1-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | buffer cache crash (#260) * host_memory: punch holes in the backing on Linux instead of zero filling freed memory Freeing guest memory zero filled the memfd backing with memset, which commits every page it touches. fallocate(PUNCH_HOLE) hands the pages back to the kernel instead, and they read as zero afterwards. Only the partial pages at the ends are still cleared by hand. Mario Kart World on Linux (RX 9070 XT, 31 GB): peak RSS 23.1 -> 15.9 GB. EDEN_PUNCH_HOLE=0 restores the memset. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * fiber: don't zero fill the stacks, untouched pages stay unbacked The value initialised stack arrays were zero filled when each fiber was made, which commits the whole stack. Left uninitialised, the pages a fiber never touches stay unbacked. Together with the descriptor ring change, about 1.2 GB less RSS in Mario Kart World. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * vulkan: don't zero fill the descriptor update ring Same as the fiber stacks: make_unique value initialised the whole ring for every frame in flight. Entries are always written before they are read, so allocate it without initialising it. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * vulkan: wait for the GPU after an ASTC transcode decode like the other ASTC decodes TranscodeAstcToBcn (astc_recompression Bc1/Bc3 with GPU decode) was the only ASTC decode recorded with wait = false. On RADV (RX 9070 XT, Mesa) Mario Kart World with BC3 flashed heavily all over the screen, while the uncompressed GPU decode, which waits, did not. CPU decode with BC3 does not flash either. Waiting after the transcode removes the flashing; full pipeline barriers around its copy did not, so the wait hides a race somewhere else rather than fixing one in the transcode itself. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * vulkan: submit after an ASTC decode instead of waiting for the GPU by default Every ASTC decode ended with scheduler.Finish(). Over a 19 minute Mario Kart World Grand Prix that stalled the GPU thread 30.7 s in total, up to 1.2 s out of every 5 s while new textures stream in (track sections, race starts), which is where the hitches were. Flushing keeps the ordering that stops the BC3 flashing on RADV without the CPU waiting: no flashing in play, and frames over 100 ms went from dozens to 3 in about 7 minutes. EDEN_NX2_ASTC_FLUSH=0 restores the wait. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * texture_cache: over the expected use, evict costly images unused for 1800 frames first The collector never deletes CostlyLoad images outside aggressive mode, so once a cup moves on, the previous tracks' textures stay resident and the cache sits over its expected use deleting the cheap images still in use. When over the expected use, costly images unused for 1800 frames (half a minute) now go first. Mario Kart World Grand Prix on an RX 9070 XT: frames over 20 ms 8.8% -> 4.0%, 58.7 fps average. EDEN_NX2_TC_OLD_AGE=<frames> sets the age, 0 turns it off. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts) * buffer_cache: no buffer for device addresses past the page table FindBuffer indexed page_table, which covers 2^36 bytes, with any address. Mario Kart World crashed after about 4.5 minutes with a uniform buffer at 0x72253b2f2890 (FindBuffer <- UpdateUniformBuffers <- DrawEnd from a macro), reading far past the table. Treat such an address as no buffer. Where that address comes from is still unknown; this only stops the crash. Assisted-by: Claude Opus 5.5 (debugging, profiling and code drafts)
| * video_core/buffer_cache: fast-guard stage buffer updates and uniform buffer ↵ymatrix144 hours1-1/+18
| | | | | | | | disable (#264)
| * video_core: keep root tables and bind table 0 as constant bank 17Duck9122 days1-0/+20
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The 3D methods 0x504/0x508 write eight 256-byte root tables, which shaders read as constant banks 24-31; NVN2 keeps push constants in table 0 (bank 17 after the Ampere lowering). Fire Emblem: Fortune's Weave keeps its material constants there and never binds bank 17, and the blit constants pass replaced those reads with render-target sizes, flooding its 3D scenes red. Maxwell3D now keeps the tables, the buffer cache streams table 0 as bank 17 when the game has not bound one, and the blit constants pass leaves shaders that read bank 17 outside its parameter block alone. Assisted-by: Claude Opus 5.5 (debugging, implementation and drafting) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com>
* | Merge remote-tracking branch 'origin/fix/nx2-gpu-written-command-lists' into ↵Takoroka2 days1-0/+7
|\ \ | | | | | | | | | | | | | | | | | | test/linux-all-prs-plus # Conflicts: # src/core/hle/service/nvdrv/nvdrv.cpp
| * | nx2: read command lists the GPU wrote from the GPU's copyTakoroka4 days1-0/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Mario Kart World runs small command lists (20-byte entries) that its culling dispatches fill in each frame. Accurate GPU mode reads every command list through the caches; Fast mode reads guest memory directly (on the split's front thread too), where those entries were still the words of an older frame, so the draws behind them were left out: the trees around the courses vanished in Fast mode only. The buffer cache now also marks the device pages the GPU writes (shader stores, copies of GPU-written data) in a bitmap the split's front can read. A command list on such a page waits for the back to run what was queued before it and is flushed when its words are still only in host buffers. In a race that is ~5% of command lists (13,500 a second of 262,000) and ~400 flushes a second.
* | | Merge remote-tracking branch 'origin/fix/cold-hint-uploaded-memory' into ↵Takoroka2 days2-0/+13
|\ \ \ | | | | | | | | | | | | test/linux-all-prs
| * | | buffer_cache: mark guest memory large buffer uploads came from coldTakoroka3 days2-0/+13
| |/ / | | | | | | | | | | | | | | | | | | | | | | | | Like the texture cache's uploads: guest ranges of 256 KiB or more uploaded to the host GPU are mostly geometry the guest rarely reads again, so every 600 frames they are marked cold and the OS pages them out first when it needs RAM. (cherry picked from commit 5c942a5158a0f9769553eeba21b6121f277d0dc5)
* | / buffer_cache: collect Switch 2 buffers from 512 MiBTakoroka3 days1-3/+4
| |/ |/| | | | | | | | | | | | | | | | | | | The collector ran on Switch 2 titles only once their buffers reached 1 GiB (aggressively at 2 GiB), so geometry of scenes long gone stayed in VRAM: Mario Kart World's title screen grew from 0.2 to 0.9 GB of buffers in ten minutes, most of a ~1.4 GB climb in VRAM in use. It now starts at 512 MiB (aggressive at 1 GiB); it only evicts buffers unused for 120 frames or more. (cherry picked from commit fb5d797636f2269591b68fda7c8d65787a9e3a9f)
* | Gpu buffer readback to cpu. Fixing visible glitches such as missing ↵néo Luigi3 days2-2/+39
|/ | | | | | collision on some object, but can extend to more. Assisted-by: Claude Opus 5.5 (debugging & implementation)
* buffer_cache: collect on the buffers' own size for Switch 2 titlesTakoroka5 days2-5/+24
| | | | | | | | | | | The buffer collector compared the whole device's memory use with thresholds from the device size, so with title 1's textures filling an 8 GB GPU (~6.8 GB used, buffers ~200 MiB of it) it ran aggressively every frame: up to 64 buffers evicted and recreated later, the GPU-written ones downloaded first with a wait for the GPU (~5% of the GPU thread). For Switch 2 titles it now runs when the buffers themselves pass 1 GiB (aggressive past 2 GiB); below the aggressive mode it also keeps buffers holding GPU-written data, which it would have to download. Other titles keep the device-wide thresholds. Buffer downloads by the collector: 5.4% -> 0%; the cache settles around 540 MiB.
* buffer_cache: one GPU wait for all buffers the collector evictsTakoroka5 days2-6/+66
| | | | | | | The buffer collector downloaded each evicted buffer's GPU-written data with its own wait for the GPU. Without per-fence downloads (d530bc4) most of title 1's evicted buffers are GPU-written, and on an 8 GB GPU near its budget it evicts some every frame. Their ranges now go to one staging buffer, copied with one wait.
* common: pass-through hash for integer and pointer keys; RangeSet::IntersectsTakoroka5 days1-3/+3
| | | | | | | | | | | Common::unordered_map/set defaulted to std::hash, a byte-wise FNV-1a on MSVC for integers and pointers, which boost's flat containers then mix again: the GPU thread's page maps (shader and query cache invalidation, per inline upload) spent ~1% in it. Integer and pointer keys now go through as they are and boost mixes them. BufferCache::IsRegionGpuModified visited every GPU-written range inside the query just to answer yes or no; without per-fence downloads those ranges stay fragmented and it was ~4% of title 1's GPU thread (inline uploads). RangeSet::Intersects answers with one lookup. Inline uploads: 10.5% -> 7.5% of the GPU thread.
* buffer_cache: no buffer downloads at every fence for Switch 2 titlesTakoroka5 days2-2/+25
| | | | | | | | | | | | | Eden copied every GPU-written buffer range back to guest memory at the next fence. Title 1 signals a fence per compute dispatch, so that was 250-330 MB/s in 13,000-20,000 ranges a second, with the fencing thread holding the buffer cache while it wrote them. Measured: the native MME's DMA reads never touched those pages (they read query words written on the CPU side), and with reactive flushing the guest CPU read GPU-written memory at most 64 times in 200 s. Without them, for Switch 2 titles (native tiling): fences ~1.1 us instead of ~5, draw setup 6.1 us instead of 6.4-6.7, macro calls 160 ns instead of 167-170, GPU-thread lock spinning 1.8% (was 5.5%). Regions stay marked GPU written and are still flushed on demand. EDEN_NX2_BUFFER_DOWNLOADS=0/1 forces either way. Reactive flushing reads of GPU-written memory are now counted in the log.
* video_core: compact the macro memo arena; download profiling countersTakoroka5 days1-0/+13
| | | | | | | | | Calls stored again with another shape leave their old record in the memo's arena; title 1 filled its 16 MiB every ~2 s and clearing it made the next ~20,000 calls interpret again. The live records are now copied to a new arena (a clear only when they alone fill three quarters of it). The memo statistics count hits on the last call's slot and compactions. With EDEN_NX2_GPU_PROFILE the GPU thread also reports how much downloaded buffer data lands on pages the native MME has DMA-read.
* video_core: skip empty fences; one lock per cache per fence; single-pass ↵Takoroka5 days1-0/+3
| | | | | | | | | | | | | overlap subtract Switch 2 titles signal a fence for each compute dispatch's release (5,000-7,000 a second in title 1). With fast GPU fences the callback runs at once, so a fence with no downloads, queries or queued operations does nothing: those (~20%) are no longer made (one in 64 still is, for the operations fences carry). Each cache's lock is now taken once per fence (the flush check took both together first). The overlap range set's Subtract restarted its search after every erase (quadratic); the fencing thread runs it for every downloaded range while holding the buffer cache. The GPU profiler counts fences, skipped fences and buffer download bytes.
* video_core: less waiting on the GPU caches' locksTakoroka6 days1-1/+2
| | | | | | | | | | | | | | | - The texture and buffer caches' mutexes spin briefly before sleeping (Common::SpinRecursiveMutex): the GPU thread, the fencing thread and CPU write traps take them tens of thousands of times a second for microseconds, and a sleeping waiter's wake-up took longer than that. Title 1's GPU thread slept 7.7% of its time there, now 3.7%. - Fences commit and pop texture and buffer downloads separately, and only when there are any (one fence per compute dispatch in Switch 2 titles): the fencing thread no longer holds both caches while it copies buffer downloads. - Inline uploads and pre-draw invalidation take the caches one at a time again: holding one while std::lock waited for the other waited longer.
* video_core: cheaper small GPU reads, buffer usage tracking and compute uniformsTakoroka6 days2-1/+25
| | | | | | | | | | | | | | - MemoryManager::ReadBlockUnsafe reads within one 4 KiB page through a single translation (as Read<T> does) instead of the page walk: texture and sampler descriptors, several per draw (2.6% -> 0.1% of title 1's GPU thread). - UsageTracker skips ranges it tracked just before (until Reset): draws bind the same multi-megabyte storage buffers again and again, and each Track wrote a word per 4 KiB (2.0% -> 0.7%). - Small compute uniform buffers the GPU has not written are streamed, as graphics ones are; synchronizing them re-protected pages an inline upload had just unprotected, two VirtualProtect calls per dispatch (1.9% -> 0.7%, page protection 3.2% -> 2.3%).
* ASTC: re-compress to BC3 except huge arrays, which stay RGBA8 and trimmedTakoroka6 days1-0/+16
| | | | | | | | | | | | | | | | Title 1 has ~700 ASTC 8x8 textures (~280 MB as stored); decoded to RGBA8 they took ~4.5 GB, pushing Eden past the 8 GB of an RTX 4070 Laptop (8.4-9.8 GB in use) and the texture cache into permanent aggressive collection. With astc_recompression=Bc3 (CPU, asynchronous) they fit (4.6-5.6 GB). The two 8192x8192x68 terrain arrays would still not fit as BC3, so ASTC 2D arrays over 1 GiB as RGBA8 always use the GPU decoder with trimmed levels (RGBA8 host image, UNORM/sRGB views), and the decoder pass now exists whenever the GPU lacks native ASTC. Also logs every 600 frames: device memory against the collector thresholds, the largest images, totals by format, and buffer cache totals. Title screen 65-150 s: 10.1 -> 16.0 fps.
* video_core: 64 GiB device address spaceTakoroka6 days2-2/+3
| | | | | | Switch 2 titles map most of their memory twice plus GPU-only regions; title 1 ran out of the 16 GiB SMMU space. The buffer cache page table and memory tracker cover the wider space.
* Switch 2: Turing macros, native compute, Ampere shader lowering, native tilingTakoroka7 days1-1/+7
| | | | | | | | | | | | | | | | | From changes.patch: the TU104/GA10x MME interpreter (shared macro context for the process's channels, also on the compute class), native compute with QMD 3.0 dependency scheduling, the Ampere-to-Maxwell shader lowering, native vertex stream sizes, and the console's GOB swizzle for textures, DMA copies and the host unswizzle shaders. Switch 2 channels turn all of it on at init. The lowering is widened for title 4 (all 32 of its programs lower): general TEX (any dimension/LOD/mask, operand split 1+1/2+1/2+2/4+n, any but consistent descriptor buffer), cube/array, SULD channel masks, 128-bit STG, indexed IPA, VOTE, SHFL with any immediate or predicate output, R2P/P2R byte select, uniform ULOP3/ULEA/UBREV/UFLO.SH/UISETP signed/UMOV URZ, IMNMX cbuf, IADD3 and FFMA uniform operands, FMUL /4 /8 x8, and BSSY/BSYNC/BREAK/WARPSYNC as no-ops. The capture stops at unmapped memory. EDEN_NX2_SHADER_DUMP dumps every native program; tools in C:\dev\nx2work test the lowering offline.
* [buffer_cashe] use old logic when Optimized Vertex Buffers is disabled (#4528)PavelBARABANOV8 days1-27/+55
| | | | | | | | | | | | | - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- Fixes vertex explosions on Turnip. Fixes Celeste on QCOM. Fixes new issues discovered after #3579 Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4528 Reviewed-by: CamilleLaVey <camillelavey99@gmail.com> Reviewed-by: Shinmegumi <shinmegumi@eden-emu.dev>
* [*] Remove Fix{Static,Small}VectorADL workaround for GCC 12 (#4386)lizzie2026-10-121-1/+1
| | | | | | | | | | | | | | | 1. We demand at least, minimum, gcc 14 2. Debian stable boost uses a newer version https://packages.debian.org/trixie/libboost-all-dev, so fix isn't longer required - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4386 Reviewed-by: MaranBr <maranbr@eden-emu.dev> Reviewed-by: crueter <crueter@eden-emu.dev>
* [buffer_cache] Simplify GPU fence synchronization and remove GPU buffer ↵MaranBr2026-09-262-23/+20
| | | | | | | | | | | | | | | | | | | readback (#4477) - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- This is understood to improve GPU synchronization within the buffer cache in certain edge cases. GPU Fence Strict is no longer necessary. We now apply the stronger synchronization only in the specific case that actually requires it. The GPU Buffer Readback has been removed, as it is no longer needed following PR #4473. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4477 Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
* [video_core] cpu buffer fix + kepler uploads / maxwell macro dirty tracking ↵xbzk2026-09-241-8/+15
| | | | | | | | | | | | | | | | | | | | | | | | | fixes (#4473) - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- Aimed to fix two known UE5 crashes: Kepler uploads and Maxwell macros, both caused by CPU/GPU races due dirty tracking issues. Kepler ComputeInline: preserved dirty tracking across dma continuations and async readback. Maxwell macros: preserved gpu owned subranges during page granular cpu uploads. DiscardWrite: stopped clearing neighboring macro arguments by rounding up ranges. DMA Step: improved continuation aware dirty sampling. To the Ender Magnolia crew (maybe 1 or 2 persons): This will fix the dash crash, and the random / shackled beast vaper crashes. There are some more UE5 issues to go next. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4473 Reviewed-by: lizzie <lizzie@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [shader_recompiler, vulkan] Virtual buffer pages with multi-range + sparse ↵CamilleLaVey2026-09-083-28/+331
| | | | | | | | | | | | buffers binding (#4362) Storage buffers can span GPU pages that aren't contiguous in host memory, but the buffer cache assumed one contiguous buffer per binding, so anything crossing a mapping boundary read the wrong bytes. Multi-range binding resolves the real segments and presents them to the shader as one buffer, aliasing their memory into a sparse VkBuffer or gathering them into a copy, all of this was mostly fixed by making a path to use sparse on buffers (the same way as texture cache has it) and retrieve properly the mapping ranges to the virtual pages, including the actual structure on the use of binding sparse. Meanwhile this fixes Monster Hunter Sunbreak z-fighting, bad performance and mostly vertex explosions (coming from the contiguous buffer read from the shader's game) it's only the basis for a further investigation towards how are we gonna treat phi calculations, optimized them and mostly eradicate the currently excess of exposure on the textures where the lightning trace should not being reflected. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4362 Reviewed-by: lizzie <lizzie@eden-emu.dev> Reviewed-by: Maufeat <sahyno1996@gmail.com>
* [video_core/buffer_cache] fix PAGE_SHIFT/BSD macro clash on Linux, *BSD (#4329)lizzie2026-08-311-17/+20
| | | | | | | | | | | | | | | | | | | Signed-off-by: lizzie <lizzie@eden-emu.dev> - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- PAGE_SHIFT is a macro (sys/user.h on linux, machine/params.h on FBSD), dont name stuff that https://stackoverflow.com/questions/77611721/get-page-shift-value-in-user-space BSD is also a macro Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4329 Reviewed-by: CamilleLaVey <camillelavey99@gmail.com> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [common] Switch to boost::unordered_flat containers (#4326)crueter2026-08-312-3/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | Replaces all instances of ankerl's unordered map/set with boost's `unordered_flat_*` classes. This uses std::hash since boost::hash is actually a lot slower. Also adds an abstraction layer in `Common` so future changes are quicker and easier. Other implementation details: - ankerl provided hash specializations for tuple and pair, so those were ported here - std::erase_if doesn't work on boost, so just used the ADL'd erase_if This should be about equal or superior performance as unordered_dense for everything except iteration. Signed-off-by: crueter <crueter@eden-emu.dev> - [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions). - [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md) - [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability. ------------------- Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4326 Reviewed-by: Lizzie and Samuel <lizzie@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [video_core] Improve synchronization and refactor buffer cache timing logic ↵MaranBr2026-07-103-17/+43
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | (#4182) This simplifies the GPU accuracy setting by removing the intermediate Balanced mode and setting High as the default on desktop platforms. It introduces a dedicated setting for GPU fence behavior, allowing the synchronization policy to be configured independently of GPU accuracy. The Vulkan buffer cache now tracks GPU recording timeline ticks and waits only when necessary, reducing unnecessary synchronization while maintaining correctness for hard-to-trace graphical bugs. GPU buffer readback has also been refined to synchronize only the affected upload regions when needed, and default DMA behavior has been updated to align with the new GPU accuracy model. ### TL;DR The fix for particles freezing and unfreezing in mid-air in `Super Mario Odyssey` has been improved, resulting in less of a performance hit. This game requires the new `Enable GPU Buffer Readback` option to be enabled to fix this issue. The vertex explosions that occurred in `Super Mario Bros. Wonder`, especially in World 4, have been completely eliminated. You can now enjoy a smooth experience without graphical glitches exploding across the screen. This game requires the new `GPU Fence Behavior` option to be set to `Strict` to fully fix this issue. The flickering issue inside certain Shrines in `The Legend of Zelda: Tears of the Kingdom` has also been fixed. For now, this game requires the new `GPU Fence Behavior` option to be set to `Accurate` to fully fix this issue. These options are intended to fix graphical bugs in games that require better synchronization behavior between CPU and GPU, so other games may be affected as well. Co-authored-by: xbzk <xbzk@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4182 Reviewed-by: Lizzie <lizzie@eden-emu.dev>
* [buffer_cache] Add option to control GPU buffer readback (#4126)MaranBr2026-06-271-9/+11
| | | | | | | | Added an option to control the GPU buffer readback, as it causes issues if the hardware cannot keep up with the additional workload. Some games require this to render certain effects properly. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4126
* [buffer_cache] Refactor buffer synchronization to use bounding-box download ↵MaranBr2026-06-221-26/+22
| | | | | | | | | | region (#4094) Implement bounding-box buffer synchronization to minimize downloads and preserve GPU-modified regions during uploads. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4094 Reviewed-by: Lizzie <lizzie@eden-emu.dev> Reviewed-by: Shinmegumi <shinmegumi@eden-emu.dev>
* [buffer_cache] Fix Xenoblade Chronicles series (#4066)MaranBr2026-06-131-2/+0
| | | | | | | | This fixes a bug introduced in the Xenoblade Chronicles series where some small elements would occasionally flicker on the screen. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4066 Reviewed-by: crueter <crueter@eden-emu.dev> Reviewed-by: Lizzie <lizzie@eden-emu.dev>
* [*] basic in-house cpp linting (#4039)lizzie2026-06-041-1/+1
| | | | | | | | | | | | - add `#pragma once` to remainder files - "correcter" defines (ANDROID), see https://groups.google.com/g/android-ndk/c/cf9_f1SLXls - extra miscelly fixups Signed-off-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4039 Reviewed-by: MaranBr <maranbr@eden-emu.dev> Reviewed-by: crueter <crueter@eden-emu.dev>
* [video,buffer] rewrite storagebufferbinding size() lambda to be full data ↵xbzk2026-06-021-15/+9
| | | | | | | | | | | | driven instead of cbuf_index based (#4041) fixes an yxzx era bug where eden assumed NVN-style packed SSBO descriptors (u64 gpu_addr followed by size) only happen in cbuf_index == 0. Mega Man Star Force Collection proves that assumption is false. It has an unbiased/global-memory SSBO descriptor in cbuf_index == 3, offsets 0x100/0x300, but the descriptor still appears to be { address, size }. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4041 Reviewed-by: MaranBr <maranbr@eden-emu.dev> Reviewed-by: crueter <crueter@eden-emu.dev>
* [buffer_cache] Fix buffer upload overwriting GPU-modified regions (#4000)MaranBr2026-05-291-13/+27
| | | | | | | | | | This fixes a bug in Super Mario Odyssey, in Bowser's Kingdom, where particles rapidly freeze and unfreeze in midair. It also fixes vertex explosions in Super Mario Party Jamboree. Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4000 Reviewed-by: Lizzie <lizzie@eden-emu.dev> Reviewed-by: Maufeat <sahyno1996@gmail.com>
* [video_core] fix std::bitset<> dirty tracker OOB, fix slightly wrong index ↵lizzie2026-05-271-3/+2
| | | | | | | | | | | | | | format estimate (#4006) u8 may have been 0xff, (aka. 255), but bitset was only 255 elements, so doing bitset[255] is technically OOB additionally the max size estimate for index formats was not correct, there can be up to 256 elements with a u8 format index, not just 255 Signed-off-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4006 Reviewed-by: crueter <crueter@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [*] Re-fix clang-cl building (#3940)lizzie2026-05-091-2/+2
| | | | | | | | Signed-off-by: lizzie <lizzie@eden-emu.dev> Co-authored-by: crueter <crueter@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3940 Reviewed-by: crueter <crueter@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [video_core/engines/maxwell3d] memory inline DrawState to reduce indirection ↵lizzie2026-04-282-22/+16
| | | | | | | | | | | | | | | | | | on hot paths (#3758) usual indirection removal helps very slightly to codegen the idea is basically to reduce the amount of pointer deference overall in the code, and use idiomatic std::variant<>-isms to not rely on vtables/unique_ptr overhead this should allow the compiler to emit better code of course it's a tiny optimisation and only CPU side, but allows us to reduce indirection which is almost always a good thing "but youre passing more parameters to the function!!!" its literally memoized into a register my friend Signed-off-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3758 Reviewed-by: crueter <crueter@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
* [vulkan] 2nd Vulkan Global Maintenance (#3853)CamilleLaVey2026-04-241-15/+18
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This pr is a sequel to the one merged some days ago (#3839); which aims to improve stability, graphical accuracy and better Vulkan implementation and coherency among all platforms, contains the next changes: -> Removal of VK_EXT_unified_image_layouts: The removal of this ext was for cleaning purposes since the only part of this extension implemented was the activator; meanwhile a proper structure of use for this extension was not implemented, currently it's not viable to keep following an idea of a proper implementation due to complexity of this feature and the state of buffer cache and texture cache, which it's task that we must do near in the future, when this happens a better oportunity will arise to properly set layouts along a proper implementation of VK_EXT_descriptors_indexing, practically this feature was dead code. -> Adjustment of VK_EXT_custom_border_color: The implementation of this feature was handled poorly and worsened during the first tries of making ExtendedDynamicState stable, by gating it's use to the slider of EDS (dyna_state) if the counter was at least in 1, even tho this entered in a bug with the RemoveUnsuitableExtension, when is not a requirement for enabling in Vulkan's documentation and was my mistake, some time later in ExtendedDynamicState refactor (#3074) I tried to make the implementation more robust in comparison the Yuzu's implementation which had bans on vendor drivers, the new handling was requesting if extension was available and what kind of support feature it had, enabling what it was available and wiring an adequate path for said available feature; which leads us to today's change, after reading carefully how certain paths weren't triggered or caused mostly issues on how extension should work I did the next changes: - I removed the forced disabling with ExtendedDynamicState setting - Resolved the bug with RemoveUnsuitableExtension + dyna_state - Removed comments of explanation + log_debug warning - Set extension to be disabled if customBorderColorWithoutFormat is not available - Helps to solidify the removal of bans in vendor drivers This changes fixes the VUID 04015 for the handling with undefined format and made the usage of the extension more near to what Vulkan specification expects, yet there is still cases where we can't emulate properly samplers and some translucid black boxes will still appear, yet, now alleviated by allowing extension choose the proper custom available in Vulkan or degrade into a fallaback of solid colors. -> Adjustment of VK_EXT/KHR_robustness2: This feature was introduced in ExtendedDynamicState refactor (#3074), as safety measure for descriptors during the Write of buffers, providing robustness with an upgraded access to image, buffers and proper discard of null data in descriptors, however, despite the configuration the logs during debug sessions never stopped to bring the next VUID-VkWriteDescriptorSet-descriptorType-00324 and VUID-VkWriteDescriptorSet-descriptorType-00325, being the first one, the most constant issue plaguing logs; the approach was not only ensuring device can access between each of the version of this feature, whether is an EXT or KHR (drivers can report one of them or both, yet, if we call the one of them and it's not the version supported, driver would not load the feature, there's a priority to the KHR version) with a simplified configuration of the extension to use only nullDescritor to deflect properly buffers and other trash data outside of descriptors bound; ensuring to wire the path when it's and not available and also with BindVertexBuffers2EXT when it's or not available; fixing both VUID's. This changes helps to save some CPU resources and memory on binding routes. - Fixes VUID-VkWriteDescriptorSet-descriptorType-00324 - Fixes VUID-VkWriteDescriptorSet-descriptorType-00325 - Fixes VUID-vkCmdBindVertexBuffers-pBuffers-00621 (alongisde a latter adjustment for pStrides) -> Adjustment VK_EXT_image_robustness: As other features, this was implemented during the ExtendedDynamicState refactor (#3074), currently this change it's just to ensure more drivers are accessing this feature by changing the modality from extension to an explicit feature, some other redundant code was cleaned within this change. -> Restored gating flush operation on removed gpu accuracy: An issue report from an user called CaptFaraday in https://github.com/eden-emulator/Issue-Reports/issues/425, posted a behavior appearing after the rework of gpu accuracy levels (#3129), which broke the rendering in Paper Mario - The-Thousand-Year Door where some graphical issues such as black flash and missing rendering from many animations through the game thanks to the removal of the flush inside FlushAndInvalidateRegion gated with IsGPULevelExtreme and suggested a possible fix with resting the missing gating and flush; which I did and properly restoring the complete behavior of this functionality + wiring to the new IsGPULevelHigh for a better semantic correctness, the change was tested and didn't affected Yoshi's Crafted World graphical problems and main reason behind the deletion of this function, fixed in fcfcee7247. - Solves https://github.com/eden-emulator/Issue-Reports/issues/266 - Solves https://github.com/eden-emulator/Issue-Reports/issues/425 - Fixes Paper Mario - The-Thousand-Year Door - Keeps Yoshi's Crafted World issue still fixed -> Adjustment VK_EXT_conditional_rendering: Yuzu inherited us in their Vulkan backend multiple flaws which got worsened with the time as game and drivers changed, aside that, with the time studying this source code and especially the Vulkan-side of Eden, I started to learn and recognize some extensions that required a wide and robust modification to ensure the logic of the extension works as intended; currently ConditionalRendering had a lot of minimal modification: - Reordering the functions from "_NotifySegment_" to avoid a masive ram leak coming from query_cache (@weakboson) (#131) - Removing a function "_NotifySegment_" from rasterizer to ensure Metroid Prime 4 stopped crashing due to serious ram leaks in query_cache (@Maufeat) (#3142) And other intents to make ConditionalRendering fully working, such as happened in ExtendedDynamicState refactor (#3074) but only patched an horrible situation with how the extension was truly working, after spending more than 3 months studying how this and other sub-sequential and essential extensions touched in this PR worked in Vulkan, I dived once again to make it work properly; one of the first changes was to fix an invalid reference lookup of queries, which fixed the removal of "NotifySegment" inside rasterizer and start to adjusting other parts of the implementation of ConditionalRendering minimally and switching with heavy tests to ensure not a single game gets broken among the changes; yet the initial benefits from fixing the indirection in the lookups to query cache, was to reduce the amount of time of GPU was spending in the constant state of queries, which proved to reduce flickering in Pokémon ZA among others, with also an small increase of performance but more noticeable stability, starting to reduce stutering bit by bit. This advances allowed me to fix one of the the functions of IsGPULevelHigh, where existed a bypass to accelerate conditional rendering without the proper checks if the extension was truly supported, freeing QCOM from the flag of a fixated presync workaround; which also improved the usage of QCOM driver for 8 Elite devices and Unreal Engine 4 - 5 games, such as Dragon Ball Z - Sparking Zero; but that's not the only benefit from the current tries to make ConditionalRendering implementation more robust and accurate to specifications, but also started to show key points of where VK_EXT_transform_feedback was also failing to work properly. -> Adjustment VK_EXT_transform_feedback: Like many other features in this PR, this one was also adjusted minimally in ExtendedDynamicState refactor (#3074), with the ConditionalRendering refactor going on, the solutions for the usage of this extensions started by ensuring each key function is properly gated by a getter which would only be enabled if the extension was already being loaded in the virtual device, if wasn't the case I was making sure to wire the fallback correctly, which wasn't in place and didn't had a robust handling since ever, this way games started to not only improve graphical accuracy, like some games such as Zelda - Echoes Of Wisdom where the lightning and dark border moves/ reacts dynamically. Besides that, this change brought the possibility to finally get rid of the indirection of the buffers synchronization which often take a non-synchornization path to ensure a faster reply but with higher possibilities to cause graphical issues in among several games; aside that, also helped to ensure "_query_cache.CounterEnable(VideoCommon::QueryType::StreamingByteCount, false)_;" function is properly allocated and reset in a different place than where this was placed. - Maintain Metroid Prime 4 fixes even after returning the lines that caused the game to be unplayable, whether was an instant crash or crash after some minutes of gameplay, fixes that were introduced in other work (#3142). Within the first step in the refactor of this extension, this work was reviewed by @wildcard which made me notice of an issue of handling inside buffers, there existed a mismatch on the tracking of feedback buffers and since we're treating them as buffer_slot, counterstream was still tracking and consuming stream_buffers and not where data was really passing through; derivating the counter selection of counterBufferCount by stream indexes and not by slot, which could cause cases of _Stream =! slot_, a solution proposed for me was to add stream mapper function where the stream slot were located + updating UpdateBuffers() to calculate buffer counts per slot and not stream, allowing to fully map the map funtion of stream mapper; along other changes on the WriteBuffers + ProduceBufferCounter to avoid any misaligment. -> Other minimal adjustments: Alongside these important adjustment, others were also made to ensure logical coherency to this recent changes, such as ensuring FullSynchronization of buffers path, since GPU has mostly a syncing issue with certain type of textures and vertex calculations, the original behavior of jumping into a non-synchronize path of buffers let GPU ran without proper awareness of the textures being loaded, ensuring more performance if all the textures reached properly inside GPU, but with no safety provided for buffers, even we added a cases were dummies and mostly buffer trash data gets discarded by nullDescriptors, this won't ensure graphical artifacts or a bad calculation on the range of lightning/ gfx could appear. I dare to think this was thought to be implemented due to the original heavy costs on Yuzu's time, this along the removal of QCOM's drivers from Query's Presync funtions. A small adjustment to the mutable functions inside the CreateImageView structure to add the extended usage: "Because Switch's GPU creates incompatible views (sRGB and UNORM) on the same image. A sRGB image can't be used as storage but it is in a UNORM view. Which is exactly the use case of these flags." - @weakboson During all of this changes inside queries, we started to get in some devices "Device Loss" warning from Vulkan along 2 specific warnings: - [ 102.384895] Render.Vulkan <Critical> video_core/vulkan_common/vulkan_debug_callback.cpp:69:DebugUtilCallback: vkDeviceWaitIdle(): THREADING ERROR : object of type VkQueue is simultaneously used in current thread 517024609280 and thread 517864567808 - [ 114.003530] Render.Vulkan <Critical> video_core/vulkan_common/vulkan_debug_callback.cpp:69:DebugUtilCallback: vkCmdBeginQuery(): VkQueryPool 0x15e400000015e4 and query 172: query not reset. After query pool creation, each query must be reset (with vkCmdResetQueryPool or vkResetQueryPool) before it is used. Queries must also be reset between uses. Since before all of this adjustments, original GPU thread often take it's time to stop and look for a moment to synchronize with CPU (non-TimelineSemaphore), the whole flow data was improved that we were producing more data stale than we could really take due to the lack of a Reset to avoid pools being filled with old data, in order to get rid of this, another try to implement ResetQueryPool's appeared which was intented to be implemented some months ago and got removed in #3270, this time aligning the vkDeviceWaitIdle + ResetQueryPool was proved to be effective than first implementation and didn't caused major issues, now GPU can Vulkan can reset staling data, which can catalogued as old once they were used and displayed in frame, keeping a more fluid exchance and discard of data. - Fixes VUID-vkCmdBeginQuery-None-00807 - Fixes multithreading error with vkDeviceWaitIdle and data allocation We have some other changes to the coherency of ExtendedDynamicState2 and the feature of restart primitives, which now patches topologies once are processed if they pass through ExtendedDynamicState2 enabled and get reset before every draw to prevent another topology VUID; also I ensured refresh, reset, clamp and overall improve the math inside the Viewport/ Scissor feature operations inside DynamicState and later upgrades. --------------------------- UPDATE (23/04/2026): After passing a heavy testing phase, an issue was encountered with AMD drivers on Windows which based on the commit: c07dfa6fb4, Super Mario Odissey started to show vertex glitches on the the waterfall + water fog being rendered incorrectly, if VertexInputDynamicState was disabled caused black screen on ExtendedDynamicState (1 - 3) and hard crash if ExtendedDynamicState it was disabled; this situation was caused to the vertex input dynamic tied to ExtendedDynamicState1, AMD driver didn't allowed fast access to BindVertexBuffers2EXT without binding strides first, which caused a syncing problem between the binded vertex and the missing buffer in the same chain, this got fixed by removing the conditions for vertex input dynamic. Aside that; another pair of issues were addressed in the meantime of refining this PR, one of them was to solve the failing BGR565 formats to swizzle into RGBA5 which allows to swap between red and blue; solving the inverse situation of blue icons on Mario Kart 8 Deluxe for older QCOM drivers and SoC's, such as Snapdragon 855 - 870; which will also help some Exynos processors to render properly. This solution was converted into a toggle/hack because it's use it's very conditional on older hardware; newer SoC's such as 8 Elite won't longer require this handling to convert properly BGR565 texture even if the support for the format is not available. --------------------------- Here an small preview of what this pr has been fixed so far, but our testing range may be more limited than what this can actually do: - Allow to display new effect on games 1. Jump Force: New particle on stages and main menu. 2. The Legend Of Zelda - Echoes of Wisdom: darkness post processing effect on screen filter such as game intro and smokes on houses (8 Elite). 3. Reduce texture flickering on games such as EoW, Monster Hunter Rise. 4. Improved performance stability on various games with Android. 5. Improved Xenoblade games rendering with QCOM stock drivers by improve viewpoint handling (8 Elite). 6. Fixes vertex explosion on Xenoblade 3 with AMD GPU with extended dynamic state enabled. 7. Fixed Mario Kart 8 Deluxe rendering with VK_EXT_vertex_input_dynamic_state enabled. 8. Fixes certain angle of Pokemon Legend Z-A would look mono color with vertex input dynamic state. 9. Fixed graphical issue with VK_EXT_vertex_input_dynamic_state on mobile drivers. 10. Fixed vertex explosion with Turnip (8 Elite) on The Legend Of Zelda- Breath Of The Wild during loading screen. 11. Fixed issue of vertex on Pokémon Legends ZA with VK_EXT_vertex_input_dynamic_state enabled. 12. Improved rendering and stability of Inmortal Fenyx Rising, including QCOM drivers being able to reach into gameplay. 13. Fixes Paper Mario - The-Thousand-Year Door missing rendering on animations through the whole game. 14. Fixed Mario Kart 8 Deluxe blue tint icon on Snapdragon 855 - 870 (by enabling Emulated BGR565 hack toggle). 15. Fixed Naruto Ultimate Ninja Storm issue rendering on characters like Naruto being blue on older QCOM SoC's and Exynos (by enabling Emulated BGR565 hack toggle) 16. Fixed Dangaronpa Killing Harmony v3 issue rendering on characters with blue tint on older QCOM SoC's. --------------------------- _**Special Thanks - Credits**_ -> @Gidoly for being able to keep track of the intensive testing phase this pr required and the will to keep helping in development, you're a good friend and very useful. -> @CaptFaraday for the suggestion of the fix for Paper Mario. -> @wildcard for the review during the refactor of VK_EXT_transform_feedback, without this comment I would probably ran into many untrackable issues. -> @weakboson for the suggestion into the solution for sRGB's and UNORM's in the incompatible views. Co-authored-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3853 Reviewed-by: crueter <crueter@eden-emu.dev>
* [vk] use static_vector instead of small_vector for TFB and other bindings ↵lizzie2026-03-061-5/+8
| | | | | | | | | | | | | | (#3641) MK8D is a big offender, taking up lots of time memcpy'ing and memmov'ing small_vector<> AND to add salt to the wound it doesn't even do heap allocations (no game does I think) - so basically useless waste of compute time in hot path for NO reason :^) Signed-off-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3641 Reviewed-by: CamilleLaVey <camillelavey99@gmail.com> Reviewed-by: DraVee <chimera@dravee.dev> Co-authored-by: lizzie <lizzie@eden-emu.dev> Co-committed-by: lizzie <lizzie@eden-emu.dev>
generated by cgit v1.2.3 (git 2.39.1) at 2026-11-04 03:16:34 +0000