[PRIM][PIR]Migrate prim rules #57554
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… migrate_prim_custom_vjp_rules
… migrate_prim_custom_vjp_rules
… migrate_prim_custom_vjp_rules
… migrate_prim_custom_vjp_rules
…e/Paddle into migrate_prim_custom_vjp_rules
… migrate_prim_custom_vjp_rules
… migrate_prim_custom_vjp_rules
|
你的PR提交成功，感谢你对开源项目的贡献! |
… migrate_prim_custom_vjp_rules
xiaoguoguo626807
approved these changes
Sep 21, 2023
Charles-hit
approved these changes
Sep 21, 2023
Frida-a
pushed a commit
to Frida-a/Paddle
that referenced
this pull request
Oct 14, 2023
* fix bugs of generating Op::Build when Op has optional tensor * add default constructor for IrMetaTensor * fix bugs * polish guard * pir support prim gelu and rsqrt * support prim bwd ops * migrate vjp rules of cast,add,multiply,elementwise_pow * add cast as primitive op * fix bugs in elementwise_pow_grad * add test for cast_grad * add test for elementwise_add_grad * add test for elementwise_mul_grad * add test for elementwise_pow_grad * fix bugs * fix bugs * support pir prim backward ops * refien * fix bug * migrate layer_norm custom vjp rules to pir * fix bugs in ir_backward * fix backward , scope, and concat_grad prim * add layer_norm fwd decompose logic * fix pow * change _use_new_ir_api to in_pir_mode * add _static_guard * fix * fix executor cuda700 error caused by full and full_like * refine * add vjp rules * fix bugs * add scope * add test * add add op prim rules --------- Co-authored-by: YuanRisheng <yuanrisheng@baidu.com> Co-authored-by: cyber-pioneer <chenzhuo@tju.edu.cn> Co-authored-by: Charles-hit <wanghao107@baidu.com> Co-authored-by: zhangbo9674 <zhangbo54@baidu.com>
jiahy0825
pushed a commit
to jiahy0825/Paddle
that referenced
this pull request
Oct 16, 2023
* fix bugs of generating Op::Build when Op has optional tensor * add default constructor for IrMetaTensor * fix bugs * polish guard * pir support prim gelu and rsqrt * support prim bwd ops * migrate vjp rules of cast,add,multiply,elementwise_pow * add cast as primitive op * fix bugs in elementwise_pow_grad * add test for cast_grad * add test for elementwise_add_grad * add test for elementwise_mul_grad * add test for elementwise_pow_grad * fix bugs * fix bugs * support pir prim backward ops * refien * fix bug * migrate layer_norm custom vjp rules to pir * fix bugs in ir_backward * fix backward , scope, and concat_grad prim * add layer_norm fwd decompose logic * fix pow * change _use_new_ir_api to in_pir_mode * add _static_guard * fix * fix executor cuda700 error caused by full and full_like * refine * add vjp rules * fix bugs * add scope * add test * add add op prim rules --------- Co-authored-by: YuanRisheng <yuanrisheng@baidu.com> Co-authored-by: cyber-pioneer <chenzhuo@tju.edu.cn> Co-authored-by: Charles-hit <wanghao107@baidu.com> Co-authored-by: zhangbo9674 <zhangbo54@baidu.com>
danleifeng
pushed a commit
to danleifeng/Paddle
that referenced
this pull request
Nov 14, 2023
* fix bugs of generating Op::Build when Op has optional tensor * add default constructor for IrMetaTensor * fix bugs * polish guard * pir support prim gelu and rsqrt * support prim bwd ops * migrate vjp rules of cast,add,multiply,elementwise_pow * add cast as primitive op * fix bugs in elementwise_pow_grad * add test for cast_grad * add test for elementwise_add_grad * add test for elementwise_mul_grad * add test for elementwise_pow_grad * fix bugs * fix bugs * support pir prim backward ops * refien * fix bug * migrate layer_norm custom vjp rules to pir * fix bugs in ir_backward * fix backward , scope, and concat_grad prim * add layer_norm fwd decompose logic * fix pow * change _use_new_ir_api to in_pir_mode * add _static_guard * fix * fix executor cuda700 error caused by full and full_like * refine * add vjp rules * fix bugs * add scope * add test * add add op prim rules --------- Co-authored-by: YuanRisheng <yuanrisheng@baidu.com> Co-authored-by: cyber-pioneer <chenzhuo@tju.edu.cn> Co-authored-by: Charles-hit <wanghao107@baidu.com> Co-authored-by: zhangbo9674 <zhangbo54@baidu.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
Others
Description
Pcard-66975
问题修复：
fullapi 调用fill_constantapi中place参数在老ir下是采用未定义place在执行阶段做place选择，新ir跟动态图接口统一，采用的place是通过_current_expected_place获取，这儿会返回一个gpu place，但是静态图真正期望的设备是在执行阶段才知道的，这儿修改full place为未定义place，跟老ir形式一样，让其在执行的时候去自动推导，否则cpu执行器在执行的时候可能会通过full op影响下游op选到cuda kernel出现cuda700的错误。full_like api中动态图place参数直接传入x.place，但是新ir下value没有place属性，这儿传入的是_current_expected_place，可能会与执行器执行的设备不一致出现cuda700的问题，这儿修改为未定义place让其在执行期间根据设备自动去推导。后续这儿是否考虑可以删除pybind接口中place参数，内部调用c++api时直接传入x的place即可。concat_grad组合算子单侧core dump，是由于其拆解规则没考虑部分输出为空指针情况。grad->grad_value。