Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter… #651

Xreki · 2016-11-29T08:15:50Z

issue #434

修改dso相关实现，使其在cpu-only的情况下也可以使用。
添加WarpCTCLayer，其中集成了warp-ctc。
添加单测test_WarpCTCLayer，其中将loss、gradients和CTCLayer对比。

TODO:

将warp-ctc作为submodule添加到PaddlePaddle中，添加后pre-commit检查时，remove-crlf和detect-private-key不通过，错误信息为IOError: [Errno 21] Is a directory: 'warp-ctc'。
ci系统中添加对submodule的自动下载和编译功能。

…face.

wangkuiyi · 2016-11-29T08:50:20Z

CMakeLists.txt

@@ -94,6 +94,11 @@ endif()
 if(NOT WITH_GPU)
    add_definitions(-DPADDLE_ONLY_CPU)
    add_definitions(-DHPPL_STUB_FUNC)
+
+    if(WITH_DSO)
+        add_definitions(-DPADDLE_USE_DSO)


请问DSO是指？

DSO是动态装载so的简写， @gangliao 实现用来装载cuda相关的一些库，所以之前只在with_gpu的情况下使用。warp-ctc使用dso这种方式，由于warp-ctc存在cpu-gpu以及cpu-only版本，所以做了些修改，使得paddle在cpu-only的版本中也支持dso这个功能。

wangkuiyi · 2016-11-29T08:52:01Z

paddle/cuda/include/hl_dso_loader.h

+ * @param    **dso_handle   dso handler
+ *
+ */
+void GetWarpctcDsoHandle(void** dso_handle);


CTC是缩写，应该大写。DSO貌似也是一个缩写？所以是 GetWarpCTCDSOHandle

@gangliao DSO这个要统一改成大写吗？

不用改吧我觉得Dso还挺常见类似于tf loader, 当时我来的时候，由于只有linux，所以用了so命名, 而没有考虑DLL一类的名字。

如果是缩写，则应该都大写。

wangkuiyi · 2016-11-29T08:52:55Z

paddle/cuda/include/hl_warpctc_wrap.h

+#define HL_WARPCTC_WRAP_H_
+
+#include "hl_base.h"
+/// #include "hl_cuda.h"


如果是不要的代码，请删除，而不是注释掉。

gangliao · 2016-11-30T02:50:23Z

paddle/cuda/src/hl_warpctc_wrap.cc

+  struct DynLoad__##__name {                                           \
+    template <typename... Args>                                        \
+    __type operator()(Args... args) {                                  \
+      typedef __type (*warpctcFunc)(Args...);                          \


建议用这种风格, 使得编译器编译期间能自动推导函数返回类型，使得DYNAMIC_LOAD_WARPCTC_WRAP更加concise。 hl_cudart_wrap.cc之所以用__type, 是因为存在extern c 函数，无法进行推导。

wangkuiyi

我不是layer的专家，我的comments主要是关于代码风格的。 @Xreki 了解并修改了就好。我approve了。赞这个PR 集成 WarpCTC！

关于代码本身，请 @Xreki 主要征求其他同事的意见吧。

luotao1 · 2016-11-30T06:05:24Z

paddle/cuda/src/hl_cuda_sequence.cu

+        for (int i = threadIdx.x; i < sequenceWidth; i += blockDim.x) {
+          batch[batchBaseIdx + i] = sequence[sequenceBaseIdx + i];
+        }
+      }


可以考虑将469-477行都合并成一个for循环:（下同）
real scale = 1.0f / (real)sequenceLength if normByTimes else 1.0f；

luotao1 · 2016-11-30T06:10:23Z

paddle/gserver/tests/test_WarpCTCLayer.cpp

+
+P_DECLARE_bool(use_gpu);
+
+const real* getData(const Matrix& matrix) {


getData和checkError函数都已经有实现，可以直接调用

checkError这个函数确实有在多个文件中重复实现，比如paddle/math/tests/test_matrixUtil.h中的checkMatrixErr，paddle/gserver/tests/test_RecurrentLayer.cpp里面的checkError。
另外，paddle/math/tests/test_matrixUtil.h和paddle/gserver/tests/TestUtil.h也都定义了函数checkMatrixEqual。因此，test_WarpCTCLayer.cpp中不能同时引入TestUtil.h和test_matrixUtil.h，这里等单测重构了之后再统一修改吧。

zzsu · 2016-12-02T08:06:50Z

Looks good to me

gangliao · 2016-12-02T08:13:22Z

CMakeLists.txt

@@ -105,6 +105,11 @@ endif()
 if(NOT WITH_GPU)
    add_definitions(-DPADDLE_ONLY_CPU)
    add_definitions(-DHPPL_STUB_FUNC)
+
+    if(WITH_DSO)


和134行重复了，可以提出去单独写了。

gangliao · 2016-12-02T08:44:20Z

LGTM

* synchronize with develop (PaddlePaddle#642) * update_commitid1.3 (PaddlePaddle#641) * update inference c++ API doc (PaddlePaddle#634) * update inference c++ API doc * fix link * thorough clean for doc (PaddlePaddle#644) * thorough clean * delete_DS_Store * Cherrypick1.3 (PaddlePaddle#652) * thorough clean * delete_DS_Store * [Don't merge now]update_install_doc (PaddlePaddle#643) * update_install_doc * follow_comments * add maxdepth (PaddlePaddle#646) * upload_md (PaddlePaddle#649) * update_version (PaddlePaddle#650) * Translation of 16 new apis (PaddlePaddle#651) * fix_windows * Final update 1.3 (PaddlePaddle#653) * thorough clean * delete_DS_Store * update_1.3 * Deadlink fix (PaddlePaddle#654) * fix_deadlinks * update_docker * Update release_note.rst * Update index_cn.rst * update_Paddle (PaddlePaddle#658) * fix pic (PaddlePaddle#659) * [to 1.3] cn api debug (PaddlePaddle#655) (PaddlePaddle#661) * debug * fix 2 -conv2d * "锚" ==> anchor(s)

* synchronize with develop (PaddlePaddle#642) * update_commitid1.3 (PaddlePaddle#641) * update inference c++ API doc (PaddlePaddle#634) * update inference c++ API doc * fix link * thorough clean for doc (PaddlePaddle#644) * thorough clean * delete_DS_Store * Cherrypick1.3 (PaddlePaddle#652) * thorough clean * delete_DS_Store * [Don't merge now]update_install_doc (PaddlePaddle#643) * update_install_doc * follow_comments * add maxdepth (PaddlePaddle#646) * upload_md (PaddlePaddle#649) * update_version (PaddlePaddle#650) * Translation of 16 new apis (PaddlePaddle#651) * fix_windows * Final update 1.3 (PaddlePaddle#653) * thorough clean * delete_DS_Store * update_1.3 * Deadlink fix (PaddlePaddle#654) * fix_deadlinks * update_docker * Update release_note.rst * Update index_cn.rst * update_Paddle (PaddlePaddle#658) * fix pic (PaddlePaddle#659) * [to 1.3] cn api debug (PaddlePaddle#655) (PaddlePaddle#661) * debug * fix 2 -conv2d * "锚" ==> anchor(s) * Weekly cherrypick0302 (PaddlePaddle#668) * Update programming_guide.md (PaddlePaddle#664) * Update programming_guide.md * Update programming_guide_en.md * Update cn api to 1.3 (PaddlePaddle#663) * Update cn api to 1.3 fluid & layers * Rest to 1.3 * Weeklyupdate 0301 (PaddlePaddle#666) * Tables_rm_op * update_op * update_index * update_book_0302 (PaddlePaddle#667) * fix_format (PaddlePaddle#669) (PaddlePaddle#670) * fix_format * Update Tables.md * Update Tables_en.md * add dataset api_cn (PaddlePaddle#673) * rm fluid.core in desigin_idea (PaddlePaddle#674) * Update fluid_design_idea.md * Update fluid_design_idea_en.md * Fix array_read code example error. (PaddlePaddle#671) Signed-off-by: zhaoyuchen <zhaoyuchen01@baidu.com> * add data_reader_cn (PaddlePaddle#676) * fix doc error (PaddlePaddle#675) * update_book_commitid (PaddlePaddle#680) * update_book_commitid * commitid0309 * fix typo * book indexes (PaddlePaddle#677)

updated ernie milestone

…ation reset gpt2 generation logic, avoid using Engine.predict

Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter…

4d487c6

…face.

reyoung assigned gangliao Nov 29, 2016

wangkuiyi reviewed Nov 29, 2016

View reviewed changes

gangliao reviewed Nov 30, 2016

View reviewed changes

Merge branch 'develop' into warpctc

5a97c98

wangkuiyi approved these changes Nov 30, 2016

View reviewed changes

luotao1 reviewed Nov 30, 2016

View reviewed changes

gangliao assigned luotao1 Nov 30, 2016

Add submodule warp-ctc.

a816443

Xreki force-pushed the warpctc branch from 7cd1130 to a816443 Compare November 30, 2016 08:28

Xreki assigned zzsu Nov 30, 2016

Xreki added this to the 0.10.0 milestone Nov 30, 2016

Merge branch 'develop' into warpctc

25f1fbc

Xreki force-pushed the warpctc branch 2 times, most recently from f8989ae to cee7001 Compare December 1, 2016 07:39

Add a script to auto compile the warp-ctc submodule.

18b85e5

Xreki force-pushed the warpctc branch from cee7001 to 18b85e5 Compare December 1, 2016 08:28

gangliao reviewed Dec 2, 2016

View reviewed changes

Xreki added 3 commits December 2, 2016 09:20

Merge branch 'develop' into warpctc

46ef2bc

Simplify the CMakelist.txt and fix typos.

7bb7fed

Merge branch 'develop' into warpctc

78bdd32

Xreki merged commit 4823075 into PaddlePaddle:develop Dec 7, 2016

Xreki deleted the warpctc branch December 7, 2016 04:54

zhhsplendid pushed a commit to zhhsplendid/Paddle that referenced this pull request Sep 25, 2019

Translation of 16 new apis (PaddlePaddle#651)

335f5cd

Meiyim pushed a commit to Meiyim/Paddle that referenced this pull request May 21, 2021

Merge pull request PaddlePaddle#651 from yinweichong/develop

69e3d22

updated ernie milestone

heavengate pushed a commit to heavengate/Paddle that referenced this pull request Aug 24, 2022

Merge pull request PaddlePaddle#651 from GuoxiaWang/refine-gpt2-gener…

1fe4081

…ation reset gpt2 generation logic, avoid using Engine.predict

lizexu123 pushed a commit to lizexu123/Paddle that referenced this pull request Feb 23, 2024

Modify readme (PaddlePaddle#651)

466b587

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter… #651

Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter… #651

Xreki commented Nov 29, 2016 •

edited

Loading

wangkuiyi Nov 29, 2016

Xreki Nov 30, 2016

wangkuiyi Nov 29, 2016

Xreki Nov 30, 2016

gangliao Nov 30, 2016

wangkuiyi Nov 30, 2016

wangkuiyi Nov 29, 2016

Xreki Nov 30, 2016

gangliao Nov 30, 2016

wangkuiyi left a comment •

edited

Loading

luotao1 Nov 30, 2016

luotao1 Nov 30, 2016

Xreki Nov 30, 2016 •

edited

Loading

zzsu commented Dec 2, 2016

gangliao Dec 2, 2016

gangliao commented Dec 2, 2016


		P_DECLARE_bool(use_gpu);

		const real* getData(const Matrix& matrix) {

Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter… #651

Integrate warp-ctc as WarpCTCLayer, including unitest and layer inter… #651

Conversation

Xreki commented Nov 29, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wangkuiyi left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Xreki Nov 30, 2016 • edited Loading

Choose a reason for hiding this comment

zzsu commented Dec 2, 2016

Choose a reason for hiding this comment

gangliao commented Dec 2, 2016

Xreki commented Nov 29, 2016 •

edited

Loading

wangkuiyi left a comment •

edited

Loading

Xreki Nov 30, 2016 •

edited

Loading