[CustomOps] TensorRT Gather Topk Ops

[grimoire]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily receiving feedbacks. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Might be disabled after NVIDIA provide a fix.

Motivation

The dynamic batch detector failed because NVIDIA/TensorRT#2299.

Modification

Add custom ops for gather topk.

old:

-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                   Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                         Memcpy DtoD (Device -> Device)         0.00%       0.000us         0.00%       0.000us       0.000us     614.000us        32.32%     614.000us       1.535us           400  
(anonymous namespace)::kernelShapeCopyH2D(int*, std:...         0.00%       0.000us         0.00%       0.000us       0.000us     374.000us        19.68%     374.000us       1.247us           300  
void cuTopKLayer::independentTopK<(nvinfer1::TopKOpe...         0.00%       0.000us         0.00%       0.000us       0.000us     301.000us        15.84%     301.000us       3.010us           100  
                                  __myl_bb0_1_MulAddGat         0.00%       0.000us         0.00%       0.000us       0.000us     204.000us        10.74%     204.000us       2.040us           100  
void cuReduceLayer::tailReduceFast<32u, (nvinfer1::R...         0.00%       0.000us         0.00%       0.000us       0.000us     200.000us        10.53%     200.000us       2.000us           100  
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
Self CPU time total: 481.878ms
Self CUDA time total: 1.900ms

new:

-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
                                                   Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg     Self CUDA   Self CUDA %    CUDA total  CUDA time avg    # of Calls  
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
void cuTopKLayer::independentTopK<(nvinfer1::TopKOpe...         0.00%       0.000us         0.00%       0.000us       0.000us     300.000us        42.80%     300.000us       3.000us           100  
void cuReduceLayer::tailReduceFast<32u, (nvinfer1::R...         0.00%       0.000us         0.00%       0.000us       0.000us     201.000us        28.67%     201.000us       2.010us           100  
void gather_topk_kernel<float>(float const*, int con...         0.00%       0.000us         0.00%       0.000us       0.000us     200.000us        28.53%     200.000us       2.000us           100  
                                        cudaEventRecord         7.96%     104.000us         7.96%     104.000us       1.040us       0.000us         0.00%       0.000us       0.000us           100  
                                       cudaLaunchKernel        91.35%       1.194ms        91.35%       1.194ms       3.980us       0.000us         0.00%       0.000us       0.000us           300  
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------  
Self CPU time total: 1.307ms
Self CUDA time total: 701.000us 

Add shape info, ut, docs, etc.

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a dependency on downstream projects of a newer version, this PR should be tested with all supported versions of downstream projects.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[irexyc]

convert:
faster_rcnn_r50_fpn_1x_coco.py with min/opt/max shape of 1,2,4. (batch)

test:
with --batch-size 1, the bbox_mAP is 0.373.
with --batch-size 2, the bbox_mAP is 0.207

[grimoire]

@irexyc fixed.

[irexyc]

test with faster-rcnn and retinanet

[codecov]

Codecov Report

Attention: Patch coverage is 34.09091% with 29 lines in your changes missing coverage. Please review.

Project coverage is 49.51%. Comparing base (a1a19f0) to head (267570d).
Report is 206 commits behind head on master.

Files with missing lines	Patch %	Lines
mmdeploy/codebase/mmdet/deploy/utils.py	23.07%	20 Missing ⚠️
...oy/codebase/mmdet/core/post_processing/bbox_nms.py	0.00%	5 Missing ⚠️
...odebase/mmdet/models/dense_heads/reppoints_head.py	60.00%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1033      +/-   ##
==========================================
- Coverage   49.59%   49.51%   -0.09%     
==========================================
  Files         305      305              
  Lines       10732    10758      +26     
  Branches     1602     1609       +7     
==========================================
+ Hits         5323     5327       +4     
- Misses       5030     5051      +21     
- Partials      379      380       +1     

Flag	Coverage Δ
unittests	49.51% <34.09%> (-0.09%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.