-
Notifications
You must be signed in to change notification settings - Fork 21.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[AOTI] support freezing for MKLDNN #124350
Commits on Apr 18, 2024
-
Configuration menu - View commit details
-
Copy full SHA for fef3149 - Browse repository at this point
Copy the full SHA fef3149View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 39b354f - Browse repository at this point
Copy the full SHA 39b354fView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for ade4d60 - Browse repository at this point
Copy the full SHA ade4d60View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 61e8725 - Browse repository at this point
Copy the full SHA 61e8725View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 17c1c2a - Browse repository at this point
Copy the full SHA 17c1c2aView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for f69260b - Browse repository at this point
Copy the full SHA f69260bView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for fe42c1d - Browse repository at this point
Copy the full SHA fe42c1dView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 108f777 - Browse repository at this point
Copy the full SHA 108f777View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 69886ec - Browse repository at this point
Copy the full SHA 69886ecView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 8a40ae7 - Browse repository at this point
Copy the full SHA 8a40ae7View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 800f425 - Browse repository at this point
Copy the full SHA 800f425View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 2f55458 - Browse repository at this point
Copy the full SHA 2f55458View commit details
Commits on Apr 19, 2024
-
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for f8ce630 - Browse repository at this point
Copy the full SHA f8ce630View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for ad15e4e - Browse repository at this point
Copy the full SHA ad15e4eView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for bc8c5f7 - Browse repository at this point
Copy the full SHA bc8c5f7View commit details
Commits on May 6, 2024
-
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for ffaa8b0 - Browse repository at this point
Copy the full SHA ffaa8b0View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for fd7bd81 - Browse repository at this point
Copy the full SHA fd7bd81View commit details
Commits on May 8, 2024
-
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for a90c2b4 - Browse repository at this point
Copy the full SHA a90c2b4View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 5170d68 - Browse repository at this point
Copy the full SHA 5170d68View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 9f8ab4a - Browse repository at this point
Copy the full SHA 9f8ab4aView commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 32a3a95 - Browse repository at this point
Copy the full SHA 32a3a95View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 14e23a9 - Browse repository at this point
Copy the full SHA 14e23a9View commit details -
Update on "[AOTI] support freezing on CPU"
This PR builds upon the work done in #114451. cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for d038be4 - Browse repository at this point
Copy the full SHA d038be4View commit details
Commits on May 9, 2024
-
Update on "[AOTI] support freezing on CPU"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. The failure of "Check mergeability of ghstack PR" in the CI is because #122472 has been reverted on main. This mergeability issue could be fixed once it relanded. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 11c9a62 - Browse repository at this point
Copy the full SHA 11c9a62View commit details
Commits on May 15, 2024
-
Update on "[AOTI] support freezing on CPU"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. The failure of "Check mergeability of ghstack PR" in the CI is because #122472 has been reverted on main. This mergeability issue could be fixed once it relanded. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 9dfba9e - Browse repository at this point
Copy the full SHA 9dfba9eView commit details
Commits on May 17, 2024
-
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 2328330 - Browse repository at this point
Copy the full SHA 2328330View commit details
Commits on May 20, 2024
-
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for b229300 - Browse repository at this point
Copy the full SHA b229300View commit details -
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for ba12b18 - Browse repository at this point
Copy the full SHA ba12b18View commit details -
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for e7b5963 - Browse repository at this point
Copy the full SHA e7b5963View commit details -
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 085c0ef - Browse repository at this point
Copy the full SHA 085c0efView commit details
Commits on May 22, 2024
-
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 9a45dae - Browse repository at this point
Copy the full SHA 9a45daeView commit details
Commits on May 23, 2024
-
Configuration menu - View commit details
-
Copy full SHA for d608f9a - Browse repository at this point
Copy the full SHA d608f9aView commit details
Commits on May 24, 2024
-
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 88f100e - Browse repository at this point
Copy the full SHA 88f100eView commit details -
Configuration menu - View commit details
-
Copy full SHA for a08e858 - Browse repository at this point
Copy the full SHA a08e858View commit details
Commits on May 25, 2024
-
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for b1e285b - Browse repository at this point
Copy the full SHA b1e285bView commit details -
Update on "[AOTI] support freezing for MKLDNN"
## Description Fixes #114450. This PR builds upon the work from imzhuhl done in #114451. This PR requires #122472 to land firstly. We leverage the serialization and deserialization API from oneDNN v3.4.1 to save the opaque MKLDNN tensor during the compilation and restore the opaque tensor when loading the compiled .so. ideep version is updated so that we won't break any pipeline even if third_party/ideep is not updated at the same time. ### Test plan: ```sh python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_conv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_deconv_freezing_non_abi_compatible_cpu python -u test/inductor/test_aot_inductor.py -k AOTInductorTestNonABICompatibleCpu.test_linear_freezing_non_abi_compatible_cpu ``` ### TODOs in follow-up PRs 1. We found that using `AOTI_TORCH_CHECK` will cause performance drop on several models (`DistillGPT2`, `MBartForConditionalGeneration`, `T5ForConditionalGeneration`, `T5Small`) compared with JIT Inductor which uses `TORCH_CHECK`. This may need further discussion how to address (`AOTI_TORCH_CHECK` is introduced in #119220). 2. Freezing in non-ABI compatible mode will work with the support in this PR. While for ABI compatible mode, we need to firstly address this issue: `AssertionError: None, i.e. optional output is not supported`. https://github.com/pytorch/pytorch/blob/6c4f43f82675b5fcfe8cf3e5983d0c0f326408aa/torch/_inductor/codegen/cpp_wrapper_cpu.py#L2023-L2024 cc jgong5 mingfeima XiaobingSuper sanchitintel ashokei jingxu10 voznesenskym penguinwu EikanWang Guobing-Chen zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Configuration menu - View commit details
-
Copy full SHA for 39ed894 - Browse repository at this point
Copy the full SHA 39ed894View commit details