Merge momentum ops/kernels #36380
Merged
Merge momentum ops/kernels #36380
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
|
Thanks for your contribution! |
sneaxiy
commented
Oct 13, 2021
| template <typename T, typename MT, bool kHasMasterParams, | ||
| uint32_t kParamNum = kHasMasterParams ? 55 : 110> | ||
| struct MergedMomentumKernelParam | ||
| : public MergedMomentumMasterParams<MT, kParamNum, kHasMasterParams> { |
There was a problem hiding this comment.
这个地方继承的目的是利用C++编译器的一个特性：继承空类不会影响该类的sizeof的大小。因此当不需要MasterParams时，sizeof(MergedMomentumKernelParam)会更小，可以容纳更多的kParamNum。
JamesLim-sy
reviewed
Oct 13, 2021
| auto master_params = ctx.MultiInput<framework::Tensor>("MasterParam"); | ||
| auto master_params_out = | ||
| ctx.MultiOutput<framework::Tensor>("MasterParamOut"); | ||
| auto multi_precision = ctx.Attr<bool>("multi_precision"); |
There was a problem hiding this comment.
这里的判断是不是表明了会在python端区分性的收集好AMP 和 非AMP optimizer，再分别传入执行optimizer计算
There was a problem hiding this comment.
是的。把AMP和非AMP分开有以下几个原因的考虑：
OperatorWithKernel::GetExpectedKernelType这个方法更好写。不然就得遍历所有Param的类型，如果有混合类型（FP16、FP32）得返回调用FP16的kernel，没有混合类型就调用单一类型的kernel。写起来会非常麻烦。MergedMomentumKernelParam这个类的sizeof会更小一些，不用装载bool multi_precision[N]，里面的实现也不用到处if-else判断是不是混合类型。- 如果有混合类型，
MergedMomentumKernelParam这个类的params和grads只能写成void *params[N]和void *grads[N]了，代码可读性比较差。
| static constexpr auto N = kParamNum; | ||
| size_t sizes[N]; | ||
| T *params[N]; | ||
| const T *grads[N]; |
There was a problem hiding this comment.
const T * 这类只读数据加入 __restrict__ 关键字会有一点性能提升，但是提升幅度可能不会明显，https://developer.nvidia.com/blog/cuda-pro-tip-optimize-pointer-aliasing/
| template <typename T, typename MT, bool kHasMasterParams, | ||
| uint32_t kParamNum = kHasMasterParams ? 55 : 110> | ||
| struct MergedMomentumKernelParam | ||
| : public MergedMomentumMasterParams<MT, kParamNum, kHasMasterParams> { |
sneaxiy
force-pushed
the
merge_momentum
branch
from
bd25e1b
to
ca2ecae
Compare
lanxianghit
approved these changes
Oct 14, 2021
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
Performance optimization
PR changes
OPs
Describe
Merge multiple momentum ops/kernels to be one momentum ops/kernels.