{"id":18525047,"url":"https://github.com/enp1s0/ozimmu","last_synced_at":"2025-04-09T12:31:02.824Z","repository":{"id":175138434,"uuid":"653399291","full_name":"enp1s0/ozIMMU","owner":"enp1s0","description":"FP64 equivalent GEMM via Int8 Tensor Cores using the Ozaki scheme","archived":false,"fork":false,"pushed_at":"2025-03-21T14:45:01.000Z","size":190,"stargazers_count":58,"open_issues_count":0,"forks_count":3,"subscribers_count":4,"default_branch":"main","last_synced_at":"2025-04-04T04:32:42.556Z","etag":null,"topics":["cuda","gemm","mixed-precision","tensorcore","tensorcores"],"latest_commit_sha":null,"homepage":"https://arxiv.org/abs/2306.11975","language":"Cuda","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/enp1s0.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2023-06-14T01:37:44.000Z","updated_at":"2025-04-04T02:26:19.000Z","dependencies_parsed_at":null,"dependency_job_id":"4d3b6851-fcb1-43f0-8ecc-c8a98f29d90b","html_url":"https://github.com/enp1s0/ozIMMU","commit_stats":null,"previous_names":["enp1s0/ozimmu"],"tags_count":2,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/enp1s0%2FozIMMU","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/enp1s0%2FozIMMU/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/enp1s0%2FozIMMU/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/enp1s0%2FozIMMU/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/enp1s0","download_url":"https://codeload.github.com/enp1s0/ozIMMU/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248040142,"owners_count":21037821,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cuda","gemm","mixed-precision","tensorcore","tensorcores"],"created_at":"2024-11-06T17:44:18.500Z","updated_at":"2025-04-09T12:31:01.784Z","avatar_url":"https://github.com/enp1s0.png","language":"Cuda","funding_links":[],"categories":[],"sub_categories":[],"readme":"# ozIMMU - DGEMM on Int8 Tensor Core\n\nThis library intercepts function calls for cuBLAS DGEMM functions and executes ozIMMU instead\n\n## Build\n```bash\ngit clone https://github.com/enp1s0/ozIMMU --recursive\ncd ozIMMU\nmkdir build\ncd build\ncmake ..\nmake -j4\n```\n\n## Usage\n\n1. Set an environmental variable to hijack the function calls\n```bash\nexport LD_PRELOAD=/path/to/ozIMMU/build/libozimmu.so\n```\n\n2. Set an environmental variable to choose the compute mode\n```bash\nexport OZIMMU_COMPUTE_MODE=fp64_int8_9\n```\nThe supported compute modes are [here](#supported-compute-mode).\n\n3. Execute the application\n\n### Supported compute mode\n| Mode          | Tensor Core type | Num splits |                         |\n|:--------------|:-----------------|:-----------|:------------------------|\n|dgemm          | --               | --         | Disable hijacking       |\n|sgemm          | --               | --         | Use SGEMM internally    |\n|fp64_int8_3    | Int8 TC          | 3          |                         |\n|fp64_int8_4    | Int8 TC          | 4          |                         |\n|fp64_int8_5    | Int8 TC          | 5          |                         |\n|fp64_int8_6    | Int8 TC          | 6          |                         |\n|fp64_int8_7    | Int8 TC          | 7          |                         |\n|fp64_int8_8    | Int8 TC          | 8          |                         |\n|fp64_int8_9    | Int8 TC          | 9          |                         |\n|fp64_int8_10   | Int8 TC          | 10         |                         |\n|fp64_int8_11   | Int8 TC          | 11         |                         |\n|fp64_int8_12   | Int8 TC          | 12         |                         |\n|fp64_int8_13   | Int8 TC          | 13         |                         |\n|fp64_int8_14   | Int8 TC          | 14         |                         |\n|fp64_int8_15   | Int8 TC          | 15         |                         |\n|fp64_int8_16   | Int8 TC          | 16         |                         |\n|fp64_int8_17   | Int8 TC          | 17         |                         |\n|fp64_int8_18   | Int8 TC          | 18         |                         |\n|fp64_int8_auto | Int8 TC          | AUTO       | fp64_int8_3..18 / dgemm |\n\n\n### Optional environmental variables\n```bash\n# Show info log\nexport OZIMMU_INFO=1\n\n# Show error and warning log\nexport OZIMMU_ERROR=1\n\n# Show CULiP ( https://github.com/enp1s0/CULiP ) log\nexport OZIMMU_ENABLE_CULIP_PROFILING=1\n\n# Choose malloc mode\nexport OZIMMU_MALLOC_ASYNC=1\n\n# Set AUTO mode mantissa loss threshold\nexport OZIMMU_AUTO_AVG_MANTISSA_LOSS_THRESHOLD=1.5\n\n# Set ozIMMU intercept threshold.\nexport OZIMMU_INTERCEPT_THRESHOLD_M=128\nexport OZIMMU_INTERCEPT_THRESHOLD_N=128\nexport OZIMMU_INTERCEPT_THRESHOLD_K=128\n# The ozIMMU gemm function is executed if `m`, `n`, and `k` are larger or equal to `OZIMMU_INTERCEPT_THRESHOLD_M`, `N`, and `K`.\n# Otherwise, the original cuBLAS function is executed.\n```\n\n## Citation\n```bibtex\n@article{ootomo2024dgemm,\n    author = {Hiroyuki Ootomo and Katsuhisa Ozaki and Rio Yokota},\n    title = {DGEMM on integer matrix multiplication unit},\n    journal = {The International Journal of High Performance Computing Applications},\n    year = {2024},\n    doi = {10.1177/10943420241239588},\n    URL = {https://doi.org/10.1177/10943420241239588}\n}\n```\n\n## License\nMIT\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fenp1s0%2Fozimmu","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fenp1s0%2Fozimmu","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fenp1s0%2Fozimmu/lists"}