78:T12c9, ## Description With NVENC on Linux, streaming through the GL→CUDA path (`gl_cuda_vram_t`, used for DMA-BUF frames from KMS, wlroots and PipeWire capture) costs other GPU work throughput, and the loss grows with the stream framerate. `convert()` called `cuGraphicsMapResources` and `cuGraphicsUnmapResources` on every frame. Each pair is a synchronizing GL↔CUDA handshake, so the GPU sits idle once per frame. The textures are now mapped once in `set_frame()` and stay mapped until the device is destroyed. The destructor unmaps them before the resources are unregistered. Because mapping no longer orders GL and CUDA access, `convert()` waits on a GL fence before the CUDA copies, and records a CUDA event that it waits on before GL writes the textures for the next frame. NvFBC and software encoding use other paths and are unchanged. RTX 5070 Ti, driver 615.78.08, 3024x1898 at 120 FPS, HEVC, KWin capture, with a game in a static scene on the same GPU (`nvidia-smi`, 45 s per build): | Build | GPU utilization | Power | |----------|-----------------|-------| | `master` | 87% | 201 W | | This PR | 100% | 223 W | ### Screenshot ### Issues Fixed or Closed - Fixes #4567 #### Roadmap Issues ## Testing Arch Linux, GCC 15, CUDA 13.4, `-DBUILD_TESTS=ON -DBUILD_WERROR=ON -DSUNSHINE_ENABLE_CUDA=ON`: builds without warnings. clang-format 23.1.0 reports no changes. No regression test: `gl_cuda_vram_t` is internal to `cuda.cpp` and needs an NVIDIA GPU and an EGL context, and the bug is lost throughput rather than wrong output. - Streamed with this build on the machine above and benchmarked it against `master` (table above). No errors in the log, and the fence never timed out. Not tested: KMS and wlroots capture, 10-bit/HDR, YUV 4:4:4. ## Type of Change - [ ] **feat**: New feature (non-breaking change which adds functionality) - [x] **fix**: Bug fix (non-breaking change which fixes an issue) - [ ] **docs**: Documentation only changes - [ ] **style**: Changes that do not affect the meaning of the code (white-space, formatting, missing semicolons, etc.) - [ ] **refactor**: Code change that neither fixes a bug nor adds a feature - [x] **perf**: Code change that improves performance - [ ] **test**: Adding missing tests or correcting existing tests - [ ] **build**: Changes that affect the build system or external dependencies - [ ] **ci**: Changes to CI configuration files and scripts - [ ] **chore**: Other changes that don't modify src or test files - [ ] **revert**: Reverts a previous commit - [ ] **BREAKING CHANGE**: Introduces a breaking change (can be combined with any type above) ## Checklist - [x] Code follows the style guidelines of this project - [x] Code has been self-reviewed - [x] Code has been commented, particularly in hard-to-understand areas - [x] Code docstring/documentation-blocks for new or existing methods/components have been added or updated - [ ] Tests have been added or updated for new or modified functionality, including regression tests for bug fixes ### AI Usage See our [AI usage policy](https://docs.lizardbyte.dev/latest/developers/contributing.html#ai-usage). - [x] AI tools were used while preparing this PR An AI agent investigated the stall, wrote the fix, ran the builds and benchmarks, and drafted this description. I reviewed the diff and the results.79:Tc09,## [![Quality Gate Passed](https://cdn.trylle.com/media/images/9c421b538b25f3b09f4b2fe2b0372ec2?key=AcQi5aACo4g%2FYo5l5IjHF3xVl6%2FNs4GJqvnQmK1WYQSdtSdRF26f%2FHvQaxVtYfwO5Zmn8dluvsUfc7rMx%2B%2FgQY2OVykvJejXvd9AGYYByqXGzgFAz8gNAGxDHrnoHKtqik9jZGG8QAdOiWU9mFubBHZXaF9DzWU0BmQyy0MI3IUIg8dPDcYRSwg1fpo1NafBvh5XrY6twS%2B7azc0BlfnOuF%2F1fUWs5kIgW%2Fyqnll0eMqf1%2BqbdHuUxive7nB4cFYbX1HRO8%3D "Quality Gate Passed")](https://sonarcloud.io/dashboard?id=LizardByte_Sunshine&pullRequest=6010) **Quality Gate passed** Issues\ ![](https://cdn.trylle.com/media/images/1f80bcb90d79b2b11bb739f48a001eba?key=AUC4JpfjThVa%2Bs177EUoO7i8HgbzCRl8Z01fv0xPDZPLrH%2BDQmUptp2xqyXz6Ry5UVz15ge3hfUJfcrNsco49HOerENB6eluOhBp2S6N6eD%2BR8aS31oqgsuMjumS%2FmD8yI8EO6a6pU6kCSlBJSXMdzHQdpi2f29zd6J3eWkDgK27I67k4Q%2FGj8yrJg24QPoLAitYTBYgMAOhTN%2FUCC%2FmZ0m0c0v6pN10QUhZhOgt%2BmAQ "") [0 New issues](https://sonarcloud.io/project/issues?id=LizardByte_Sunshine&pullRequest=6010&issueStatuses=OPEN,CONFIRMED&sinceLeakPeriod=true)\ ![](https://cdn.trylle.com/media/images/2dc6921822f23cc8f5227994feb4f3a1?key=Ac0A%2B0GOiKtG7s6yUOkdAw93Omn2d6UW4CeD2L0U%2BJBRKNXLhiTbhcAYui4CWhVpOt5sS%2FIpI5GuMUAr%2BR7rVdKH%2BhxG6gKozNa21fwbQxSdpTU%2FjZ2iFl9kDiolBU4S%2BRFIeFGghV92lOKfCd6%2BYiGPTNPcoNmsgSbkTS3%2FbidmrAdQOrK9LWMRC%2FNHb3Q89TG%2B6T73DUYy9yDpnhpnR17g1F78itSkkEQd60W2pU9pWj4%3D "") [0 Accepted issues](https://sonarcloud.io/project/issues?id=LizardByte_Sunshine&pullRequest=6010&issueStatuses=ACCEPTED) Measures\ ![](https://cdn.trylle.com/media/images/1f80bcb90d79b2b11bb739f48a001eba?key=AUC4JpfjThVa%2Bs177EUoO7i8HgbzCRl8Z01fv0xPDZPLrH%2BDQmUptp2xqyXz6Ry5UVz15ge3hfUJfcrNsco49HOerENB6eluOhBp2S6N6eD%2BR8aS31oqgsuMjumS%2FmD8yI8EO6a6pU6kCSlBJSXMdzHQdpi2f29zd6J3eWkDgK27I67k4Q%2FGj8yrJg24QPoLAitYTBYgMAOhTN%2FUCC%2FmZ0m0c0v6pN10QUhZhOgt%2BmAQ "") [0 Security Hotspots](https://sonarcloud.io/project/security_hotspots?id=LizardByte_Sunshine&pullRequest=6010&issueStatuses=OPEN,CONFIRMED&sinceLeakPeriod=true)\ ![](https://cdn.trylle.com/media/images/1f80bcb90d79b2b11bb739f48a001eba?key=AUC4JpfjThVa%2Bs177EUoO7i8HgbzCRl8Z01fv0xPDZPLrH%2BDQmUptp2xqyXz6Ry5UVz15ge3hfUJfcrNsco49HOerENB6eluOhBp2S6N6eD%2BR8aS31oqgsuMjumS%2FmD8yI8EO6a6pU6kCSlBJSXMdzHQdpi2f29zd6J3eWkDgK27I67k4Q%2FGj8yrJg24QPoLAitYTBYgMAOhTN%2FUCC%2FmZ0m0c0v6pN10QUhZhOgt%2BmAQ "") [0.0% Coverage on New Code](https://sonarcloud.io/component_measures?id=LizardByte_Sunshine&pullRequest=6010&metric=new_coverage&view=list)\ ![](https://cdn.trylle.com/media/images/1f80bcb90d79b2b11bb739f48a001eba?key=AUC4JpfjThVa%2Bs177EUoO7i8HgbzCRl8Z01fv0xPDZPLrH%2BDQmUptp2xqyXz6Ry5UVz15ge3hfUJfcrNsco49HOerENB6eluOhBp2S6N6eD%2BR8aS31oqgsuMjumS%2FmD8yI8EO6a6pU6kCSlBJSXMdzHQdpi2f29zd6J3eWkDgK27I67k4Q%2FGj8yrJg24QPoLAitYTBYgMAOhTN%2FUCC%2FmZ0m0c0v6pN10QUhZhOgt%2BmAQ "") [0.0% Duplication on New Code](https://sonarcloud.io/component_measures?id=LizardByte_Sunshine&pullRequest=6010&metric=new_duplicated_lines_density&view=list) [See analysis details on SonarQube Cloud](https://sonarcloud.io/dashboard?id=LizardByte_Sunshine&pullRequest=6010) 62:["$","$L75",null,{"entries":[{"key":["platform",{"userId":"anon"},"pulls","page",{"owner":"LizardByte","repo":"Sunshine","pullNumber":6010}],"data":{"repositoryId":"3992","detail":{"author":{"avatarUrl":"https://cdn.trylle.com/media/images/8f6fa84e0d0901ab187126de053875a3?key=AdaJfppxO7tPBpqmQbo7A1mWOShrX1Ddc%2BK1MslraR7iaLgccDQpwaSVVAn88XKj%2Bnzyv4r8iY62UDytxeQWchXb2tl3g%2FN1U7Y1cLdF65cfy2qqfAoq1qbnIW3wpkXbxlVzUt8SLHp9PL1ZstCv5%2FrF4trW4LZ3TaOxwinnEzQ%2F3QPR1DjIJPB90Q%3D%3D","externalProvider":"github","login":"Niewiarowski","type":"User","url":"https://github.com/Niewiarowski"},"authorId":"21334271","baseRefName":"master","closedAt":null,"comments":1,"createdAt":"2026-10-28T20:45:03Z","headRefName":"Niewiarowski/fix/linux-cuda-persistent-gl-map","id":"4958361","isDraft":false,"labels":[],"number":6010,"repository":{"name":"Sunshine","owner":"LizardByte","provider":"forgejo"},"state":"open","title":"fix(linux/cuda): map GL textures into CUDA once instead of per frame","updatedAt":"2026-10-28T21:16:57Z","url":"https://trylle.com/LizardByte/Sunshine/pulls/6010","body":"$78","additions":98,"deletions":14,"changedFiles":1,"commits":2,"reviewComments":0,"headSha":"f3f88db87e6aefe8957caec3354b48bb67a46cf0","baseSha":"a679b059096bcdd502e2192ce4f6a3c3732fbaae","headRepoOwner":"LizardByte","headRepoName":"Sunshine","isMerged":false,"mergeCommitSha":null,"mergedBy":null,"mergeable":true,"mergeableState":"clean","conflictingFiles":null,"requestedTeams":[],"requestedReviewers":[],"graphqlId":"issue:6010"},"comments":[{"id":27026619,"graphqlId":"27026619","body":"$79","createdAt":"2026-10-28T21:16:56Z","author":{"avatarUrl":"https://cdn.trylle.com/media/images/ffba2eae3878387d344176f3a6e03729?key=ASDUdHAAsxJIAzAhbbYTxbs1BE8boDosy7WO7CO6xXE%2Fxi90ansATxCFdbDT9%2FASXlT6s5z9DHN7VIv965csx7KvK%2FxHWyAwh%2FLfiaj%2BZjXYlqI1smSZf%2FKXF8%2BA8Gg4x9FgZ9IJx3aycXtKcw%2BjBlX4b8sfj%2Fcj6oy6PGJ3EzgTdt5YNoI9FE0b%2Fw%3D%3D","externalProvider":"github","login":"sonarqubecloud[bot]","type":"Bot","url":"https://github.com/apps/sonarqubecloud"}}],"commits":[{"sha":"f3f88db87e6aefe8957caec3354b48bb67a46cf0","message":"refactor(linux/cuda): group persistent mapping state into cu_mapping\n\nKeeps gl_cuda_vram_t within Sonar's 20-field limit (cpp:S1820).\n","createdAt":"2026-10-28T17:02:43-04:00","author":{"avatarUrl":"","login":"","name":"Niewiarowski","type":"User","url":""},"verification":{"verified":false,"reason":"gpg.error.not_signed_commit"}},{"sha":"b232be3de49efad95388437fbf56a7000c39b8aa","message":"fix(linux/cuda): map GL textures into CUDA once instead of per frame\n\ngl_cuda_vram_t::convert() mapped and unmapped the GL textures with\ncuGraphicsMapResources/cuGraphicsUnmapResources on every frame. Each pair\nis a synchronizing GL<->CUDA handshake, so the GPU idles once per frame and\nthroughput lost by other GPU workloads grows with the stream framerate.\n\nMap the textures once in set_frame(), cache the CUarray handles, and keep\nthem mapped for the lifetime of the encode device. The destructor unmaps\nthem before the registered resources are unregistered.\n\nBecause mapping no longer orders GL and CUDA access, convert() does it\nexplicitly: it waits on a GL fence after the GL conversion before issuing\nthe CUDA copies, and records a CUDA event after the copies that is waited\non before GL renders into the textures again.\n","createdAt":"2026-10-28T16:39:01-04:00","author":{"avatarUrl":"","login":"","name":"Niewiarowski","type":"User","url":""},"verification":{"verified":false,"reason":"gpg.error.not_signed_commit"}}],"events":[],"reviews":[],"stackEvents":[],"commentPagination":{"totalCount":1,"perPage":30,"loadedPages":[1]},"eventPagination":{"loadedPages":[1],"hasMore":false},"headRefDeleted":false},"fetchedAt":1793417197460}],"children":"$L7a"}] 64:null 6b:[["$","title","0",{"children":"Commits · fix(linux/cuda): map GL textures into CUDA once instead of per frame · LizardByte/Sunshine | Trylle"}],["$","meta","1",{"name":"description","content":"Commits on fix(linux/cuda): map GL textures into CUDA once instead of per frame (#6010) in LizardByte/Sunshine."}],["$","meta","2",{"name":"application-name","content":"Trylle"}],["$","link","3",{"rel":"manifest","href":"/manifest.json","crossOrigin":"$undefined"}],["$","meta","4",{"name":"keywords","content":"Commits · fix(linux/cuda): map GL textures into CUDA once instead of per frame · LizardByte/Sunshine, LizardByte/Sunshine, LizardByte/Sunshine repository, Sunshine, C++ repository, source code, trylle, agent-native git platform, git platform for agents and humans, git platform for ai agents, coding agent pull requests, ai code review, repository automation, stacked pull requests, two-way github sync, github-compatible actions, try cli"}],["$","meta","5",{"name":"robots","content":"index, follow, max-image-preview:large, max-snippet:-1, max-video-preview:-1"}],["$","meta","6",{"name":"mobile-web-app-capable","content":"yes"}],["$","meta","7",{"name":"googlebot","content":"index, follow, max-image-preview:large, max-snippet:-1, max-video-preview:-1"}],["$","meta","8",{"name":"twitter:domain","content":"trylle.com"}],["$","meta","9",{"name":"twitter:url","content":"https://trylle.com/LizardByte/Sunshine/pulls/6010/commits"}],["$","link","10",{"rel":"canonical","href":"https://trylle.com/LizardByte/Sunshine/pulls/6010/commits"}],["$","meta","11",{"name":"format-detection","content":"telephone=no"}],["$","meta","12",{"name":"mobile-web-app-capable","content":"yes"}],["$","meta","13",{"name":"apple-mobile-web-app-title","content":"Trylle"}],["$","meta","14",{"name":"apple-mobile-web-app-status-bar-style","content":"black-translucent"}],["$","meta","15",{"property":"og:title","content":"Commits · fix(linux/cuda): map GL textures into CUDA once instead of per frame · LizardByte/Sunshine | Trylle"}],["$","meta","16",{"property":"og:description","content":"Commits on fix(linux/cuda): map GL textures into CUDA once instead of per frame (#6010) in LizardByte/Sunshine."}],["$","meta","17",{"property":"og:url","content":"https://trylle.com/LizardByte/Sunshine/pulls/6010/commits"}],["$","meta","18",{"property":"og:site_name","content":"Trylle"}],["$","meta","19",{"property":"og:locale","content":"en_US"}],["$","meta","20",{"property":"og:image","content":"https://trylle.com/api/og.png?kind=repo&owner=LizardByte&repo=Sunshine"}],["$","meta","21",{"property":"og:image:secure_url","content":"https://trylle.com/api/og.png?kind=repo&owner=LizardByte&repo=Sunshine"}],["$","meta","22",{"property":"og:image:type","content":"image/png"}],["$","meta","23",{"property":"og:image:width","content":"1200"}],["$","meta","24",{"property":"og:image:height","content":"630"}],["$","meta","25",{"property":"og:image:alt","content":"LizardByte/Sunshine repository preview on Trylle"}],["$","meta","26",{"property":"og:type","content":"website"}],["$","meta","27",{"name":"twitter:card","content":"summary_large_image"}],["$","meta","28",{"name":"twitter:site","content":"@trylle"}],["$","meta","29",{"name":"twitter:title","content":"Commits · fix(linux/cuda): map GL textures into CUDA once instead of per frame · LizardByte/Sunshine | Trylle"}],["$","meta","30",{"name":"twitter:description","content":"Commits on fix(linux/cuda): map GL textures into CUDA once instead of per frame (#6010) in LizardByte/Sunshine."}],["$","meta","31",{"name":"twitter:image","content":"https://trylle.com/api/og.png?kind=repo&owner=LizardByte&repo=Sunshine"}],["$","meta","32",{"name":"twitter:image:alt","content":"LizardByte/Sunshine repository preview on Trylle"}],["$","link","33",{"rel":"icon","href":"/favicon.png","type":"image/png"}],["$","link","34",{"rel":"icon","href":"/favicon-white.png","type":"image/png","media":"(prefers-color-scheme: dark)"}],"$L7b","$L7c"]

Commits on Wed, Oct 28, 2026