[XPU] feat: add xpu async memory copy to enable zero cost checkpoint #71168
Conversation
This file contains hidden or 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! |
tizhou86
force-pushed
the
flashcheckpoint
branch
2 times, most recently
from
033a925 to
406fb2d
Compare
tizhou86
force-pushed
the
flashcheckpoint
branch
from
406fb2d to
dd8b145
Compare
XiaociZhang
reviewed
Mar 6, 2025
| std::shared_ptr<distributed::XpuAsyncLoad::Task>>( | ||
| *m, "XpuAsyncLoadTask") | ||
| .def("is_completed", | ||
| &distributed::XpuAsyncLoad::Task::IsCompleted, |
dynamicheart
reviewed
Mar 7, 2025
| return; | ||
| } | ||
| // platform::MemcpySyncH2D(dst, src, num, dst_place); | ||
| xpu_memcpy_async(dst, src, num, XPU_HOST_TO_DEVICE, stream); |
| return; | ||
| } | ||
| // platform::MemcpySyncD2H(dst, src, num, src_place); | ||
| xpu_memcpy_async(dst, src, num, XPU_DEVICE_TO_HOST, stream); |
dynamicheart
reviewed
Mar 7, 2025
| // (but let's store a CPU event just so we can return a reference). | ||
| // In a real design, you might do a separate approach. | ||
|
|
||
| phi::Place event_place = is_xpu_place(place) ? phi::CPUPlace() : place; |
| data1 = paddle.randn([10, 10]) | ||
| print_debug_info(data1, "data1 (for compute)") | ||
|
|
||
| # Offload data0 -> pinned memory (usually on CPU) |
There was a problem hiding this comment.
似乎没有看到哪里有指定是CPU pinned memory的类型？
| void XpuAsyncLoad::SyncCalcuStream(const Place& place, | ||
| phi::XPUContext* offload_ctx, | ||
| platform::DeviceEvent* calc_event) { | ||
| if (is_xpu_place(place)) { |
There was a problem hiding this comment.
这个place似乎是offload的src place，也就是XPU place，为啥不需要插入event wait？
| src.place(), | ||
| src_ptr, | ||
| size, | ||
| /*stream=*/nullptr); |
There was a problem hiding this comment.
这里使用nullptr stream，而task->UpdateWaitChain(*load_ctx_)使用的是load_ctx_的stream，可能会有同步问题，即task.wait返回成功了，但是copy可能还未成功
| src.place(), | ||
| src_ptr, | ||
| size, | ||
| /*stream=*/nullptr); |
| return task; | ||
| } | ||
|
|
||
| /* ------------ Reload (CPU -> XPU) ------------ */ |
There was a problem hiding this comment.
Reload的CPU如果不是pinned memory，xpu_memcpy_async可能会退化成同步xpu_memcpy
YqGe585
pushed a commit
to YqGe585/Paddle
that referenced
this pull request
May 7, 2025
…addlePaddle#71168) * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint * [XPU] feat: add xpu async memory copy to enable zero cost checkpoint
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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
Custom Device
PR Types
New features
Description
Add xpu async memory copy to enable zero cost checkpoint