{"id":18711245,"url":"https://github.com/rocm/misa","last_synced_at":"2025-04-12T11:35:09.893Z","repository":{"id":41890915,"uuid":"254788916","full_name":"ROCm/MISA","owner":"ROCm","description":"Machine Intelligence Shader Autogen. AMDGPU ML shader code generator. (previously iGEMMgen)","archived":false,"fork":false,"pushed_at":"2025-03-20T17:49:14.000Z","size":3583,"stargazers_count":34,"open_issues_count":12,"forks_count":14,"subscribers_count":24,"default_branch":"develop","last_synced_at":"2025-03-26T06:22:33.919Z","etag":null,"topics":["amd","assembly","convolution","gpu","implicit-gemm","implicit-gemm-algorithm","python","tensor-contraction"],"latest_commit_sha":null,"homepage":"","language":"Python","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/ROCm.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.md","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":"2020-04-11T03:57:39.000Z","updated_at":"2025-03-20T17:49:22.000Z","dependencies_parsed_at":"2024-01-16T09:43:14.474Z","dependency_job_id":"35f8e488-b3fe-49f6-aca5-ef567d71cd2d","html_url":"https://github.com/ROCm/MISA","commit_stats":null,"previous_names":["rocm/misa"],"tags_count":7,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ROCm%2FMISA","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ROCm%2FMISA/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ROCm%2FMISA/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ROCm%2FMISA/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ROCm","download_url":"https://codeload.github.com/ROCm/MISA/tar.gz/refs/heads/develop","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248560702,"owners_count":21124706,"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":["amd","assembly","convolution","gpu","implicit-gemm","implicit-gemm-algorithm","python","tensor-contraction"],"created_at":"2024-11-07T12:38:07.306Z","updated_at":"2025-04-12T11:35:09.849Z","avatar_url":"https://github.com/ROCm.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"*under rapid development*\n\n# iGEMMgen\n\nCode generator for implicit gemm algorithm (generic tensor contraction)\n\n# Generate kernel\nsince f-string is utilized in python, require python \u003e= 3.6 to run.\n```\n# generate code based on tunable configuration, use one of following command to generate each direction\npython3 igemm_codegen.py config/igemm_fwd_gtc_gfx908.config\npython3 igemm_codegen.py config/igemm_bwd_gtc_gfx908.config\npython3 igemm_codegen.py config/igemm_wrw_gtc_gfx908.config\n\n# or auto generate code for all possible combinations, use one of following command to generate each direction\npython3 igemm_codegen.py config/igemm_fwd_gtc_gfx908_seq.config\npython3 igemm_codegen.py config/igemm_bwd_gtc_gfx908_seq.config\npython3 igemm_codegen.py config/igemm_wrw_gtc_gfx908_seq.config\n```\n\nThe output file will result in `out` directory. result in a assembly file `*.s` and several `*.inc` for different tile size, a codeobject `*.hsaco` and a host driver executable `conv_driver.exe`. This executable accept same cmdline argument as [MIOpenDriver](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/driver.html). e.g.\n```\n./conv_driver.exe  conv -n 128 -c 1024 -H 17 -W 17 -k 1024  -y 1 -x 1 -p 0 -q 0 -u 1 -v 1 -l 1 -j 1 -F 2 -V 1\n```\ncurrently this executable will run all the kernel configs one by one, the same as you used for kernel generation stage.\n\nsome environment variables may affect the behavior and printout of `conv_driver.exe`\n* `IGEMM_HSACO` : indicate the path of code object to use. default use the generated one in currentl directory.\n* `IGEMM_SCLK_MHZ` : current GPU sclk MHZ. used to calculate efficiency.\n* `IGEMM_LOG_FASTEST_CONFIG` : set to `1` to print the fastest config from current convolution. default is `0`\n\n*more description to be added*\n\n# Third party code for fp16 data type\n* `half.hpp` : When fp16 kernel is generated, `half.hpp` need to be installed, e.g.:\n``` shell\nwget https://github.com/pfultz2/half/archive/1.12.0.tar.gz\ntar -zvxf 1.12.0.tar.gz\ncp half-1.12.0/include/half.hpp /usr/local/include/\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Frocm%2Fmisa","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Frocm%2Fmisa","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Frocm%2Fmisa/lists"}