[SOT][Faster Guard] add FasterStringifiedExpression #69353
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
SigureMo
reviewed
Nov 13, 2024
| for (size_t i = 0; i < shape.size(); ++i) { | ||
| if (py::isinstance<py::int_>(shape[i])) { | ||
| _expected[i] = std::make_optional(shape[i].cast<int64_t>()); | ||
| } |
There was a problem hiding this comment.
这块主要是针对SombolcInt，现在只要不是整数都被当成动态shape了，没有做其他类型的检查
There was a problem hiding this comment.
那在构造的时候需要小心传入 -1，会产生奇怪的问题，或者直接不允许 -1（负数）
python/paddle/jit/sot/opcode_translator/executor/variables/basic.py
Outdated
Show resolved
Hide resolved
| self.faster_guard = faster_guard | ||
| if ENV_SOT_ENABLE_FASTER_GUARD: | ||
| original_expr_template = expr_template | ||
| guard_name = f"guard_check_{id(faster_guard)}" |
There was a problem hiding this comment.
名字里是否可以表明 guard 类型？这样生成的 guard 更具可读性些～
There was a problem hiding this comment.
[Cache]: Cache miss, Guard is
lambda frame: (___t_1 := (len((___t_0 := (frame.f_locals['input'])).shape) == 4)) and (___t_2 := (___t_0.shape[0] == 32)) and (___t_3 := (___t_0.shape[1] == 16)) and (___t_4 := (___t_0.shape[2] == 176)) and (___t_5 := (___t_0.shape[3] == 176)) and (___t_6 := (___t_0.dtype == paddle.core.DataType(10))) and (___t_7 := (___t_0.stop_gradient == False)) and (___t_15 := (id(type((___t_14 := ((___t_13 := ((___t_12 := ((___t_11 := ((___t_10 := ((___t_9 := ((___t_8 := (frame.f_locals['self']))._sub_layers))['0'])).__call__)).__globals__['in_to_static_mode'])).__globals__['global_var']))._in_to_static_mode_)))) == 94911584974496)) and (___t_16 := (___t_14 == False)) and (___t_18 := (isinstance((___t_17 := (___t_10._forward_pre_hooks)), dict))) and (___t_19 := (len(___t_17) == 0)) and (___t_21 := (isinstance((___t_20 := (___t_10._forward_post_hooks)), dict))) and (___t_22 := (len(___t_20) == 0)) and (___t_24 := (id(type((___t_23 := (___t_10._built)))) == 94911584974496)) and (___t_25 := (___t_23 == False)) and (___t_28 := (id(type((___t_27 := ((___t_26 := (___t_11.__globals__['in_profiler_mode'])).__globals__['_is_profiler_used'])))) == 94911584974496)) and (___t_29 := (___t_27 == False)) and (___t_32 := (isinstance((___t_31 := ((___t_30 := (___t_10.dw_conv))._forward_pre_hooks)), dict))) and (___t_33 := (len(___t_31) == 0)) and (___t_35 := (isinstance((___t_34 := (___t_30._forward_post_hooks)), dict))) and (___t_36 := (len(___t_34) == 0)) and (___t_38 := (id(type((___t_37 := (___t_30._built)))) == 94911584974496)) and (___t_39 := (___t_37 == False)) and (___t_41 := (id((___t_40 := (___t_30.conv))) == 139643550285056)) and (___t_42 := (___t_40.training == True)) and (___t_44 := (id((___t_43 := (___t_30.bn))) == 139643550284096)) and (___t_45 := (___t_43.training == True)) and (___t_47 := (id((___t_46 := (___t_30.act))) == 139643550285392)) and (___t_48 := (___t_46.training == True)) and (___t_50 := (id(type((___t_49 := (___t_10.use_se)))) == 94911584974496)) and (___t_51 := (___t_49 == False)) and (___t_54 := (isinstance((___t_53 := ((___t_52 := (___t_10.pw_conv))._forward_pre_hooks)), dict))) and (___t_55 := (len(___t_53) == 0)) and (___t_57 := (isinstance((___t_56 := (___t_52._forward_post_hooks)), dict))) and (___t_58 := (len(___t_56) == 0)) and (___t_60 := (id(type((___t_59 := (___t_52._built)))) == 94911584974496)) and (___t_61 := (___t_59 == False)) and (___t_63 := (id((___t_62 := (___t_52.conv))) == 139643550385152)) and (___t_64 := (___t_62.training == True)) and (___t_66 := (id((___t_65 := (___t_52.bn))) == 139643550385728)) and (___t_67 := (___t_65.training == True)) and (___t_69 := (id((___t_68 := (___t_52.act))) == 139643550385248)) and (___t_70 := (___t_68.training == True)) and (___t_71 := (isinstance(___t_9, dict))) and (___t_72 := (len(___t_9) == 1)) and (___t_73 := (id(___t_10) == 139643550283328)) and (___t_74 := (___t_10.training == True))
[Cache]: missed at frame.f_locals['input'].shape[1] == 16
最好 faster guard 仍然能有这样的体验，能够知道是在什么地方 cache miss 了（实现方式可能有所不同）
There was a problem hiding this comment.
现在是先加上了名字，后面再改进一下把参数也标注出来
There was a problem hiding this comment.
现在是先加上了名字，后面再改进一下把参数也标注出来
嗯嗯，这个 PR 不用考虑那么完善，后续设计考虑到这点即可（后续 lookup 不一定走这套（Python 端线性查找），在最终态（比如 C++ 端树形查找）实现这一点即可）
15 tasks
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 Category
Execute Infrastructure
PR Types
Performance
Description