2d517d270e
TensorRT OSS v8.2 Early Access Release
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-10-05 11:30:06 -07:00
0b65abfbd8
Update plugin directory structure
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-08-05 13:14:13 -07:00
aff45dd565
TensorRT OSS 8.0 release
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-07-02 16:35:44 -07:00
a2b3d3d5cc
Add switch for batch agnostic mode in NMS plugin
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-06-23 12:06:05 -07:00
d742af876d
Optimize TLT MaskRCNN plugins
...
- enable fp16 precision in multilevelCropAndResizePlugin and multilevelProposeROIPlugin
- Algorithms optimization for NMS kernels and ROIAlign kernel
- Fix invalid cuda config issue when bs is larger than 32
- Fix issues found on Jetson NANO
Signed-off-by: Tyler Zhu <tylerz@nvidia.com >
2021-06-23 12:06:05 -07:00
c5b8cf67fb
Fix numerical errors for floats in NMS/batchedNMS plugins.
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-06-23 12:06:05 -07:00
64fa61b415
Add varlen MHA fp16 slen=384 kernels for sm_86
...
1. add varlen mha fp16 slen=384 kernel for sm_86
2. referesh all sm_86 kernels now use NVCC -gencode=arch=compute_86,code=\"sm_86\"
3. use unfused kernel for fixed len s=384 fp16
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-04-12 14:56:06 -07:00
cc5dede848
clang-format plugins and samples
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-04-12 14:56:06 -07:00
0dcade2184
don't violate strict aliasing with reinterpret_cast
...
Signed-off-by: Tom Peters <thomas.d.peters@gmail.com >
2021-03-18 22:49:40 -07:00
19910a4c95
Bugfixes for decodeBBoxes kernel.
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-03-18 02:45:54 -07:00
d5b878a18e
optimized FP16 NMS/BatchedNMS plugins.
...
- enable n-bit radix sort for NMS_TRT plugin
- implement NMSDynamic_TRT plugin and enable n-bit radix sort for it
- enable n-bit radix sort for BatchedNMS_TRT and BatchedNMSDynamic_TRT plugins
- fixed a bug in configurePlugin() method of BatchedNMS_TRT
- other minor fixes for BatchedNMS_TRT/BatchedNMSDynamic_TRT plugins
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-03-18 02:45:54 -07:00
10a51bed54
Fix a computational problem of scaledSoftmax.
...
The original implementation results in wrong results of sum of softmax such that the results of BERT models (128 < seq_len < 384 and seq_len > 384) are very large or even 'nan'.
This implementation fix the computational problem such that the results of BERT models (128 < seq_len < 384 and seq_len > 384) become correct.
Signed-off-by: yuanzexi <hiyuanzexi@outlook.com >
2021-03-18 02:45:54 -07:00
183f891191
Update onnx-tensorrt and copyright headers (2021)
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2021-01-14 01:16:45 -08:00
0d538ddedc
Explicit casts to resolve compiler errors in __half type conversion
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
Signed-off-by: Paul Bridger <paul@paulbridger.com >
2021-01-13 17:56:41 -08:00
aa1b4b09d6
fix failing NMS on fp16 due to issue with box-size calculation
...
Signed-off-by: Paul Bridger <paul@paulbridger.com >
2021-01-13 17:56:41 -08:00
42dbbb0ef3
add fp16 capability to batchedNMSPlugins
...
Signed-off-by: Paul Bridger <paul@paulbridger.com >
2021-01-13 17:56:41 -08:00
492878b2df
TensorRT OSS release v7.2.1
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2020-10-20 16:48:23 -07:00
275eefce39
Restore lReLUPlugin in OSS
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2020-09-04 15:45:15 -07:00
3a68b0af8c
Fix maskrcnn kernel issue on Jetson Nano
...
Signed-off-by: Tyler Zhu <tylerz@nvidia.com >
2020-07-17 14:38:00 +08:00
c9b4e4dbc1
Review feedback #1 - remove redundant kernel
2020-07-11 03:11:52 -07:00
29498fd814
Re-enable GridAnchorRect_TRT plugin with rectangular feature maps
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2020-07-11 03:11:52 -07:00
805810b112
TensorRT 7.1 Open Source Release
...
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2020-07-01 16:31:50 -07:00
58bae5bfae
Commit for new MaskRCNN inference plugins and kernels:
2020-06-17 13:48:22 -07:00
eff9163466
BERT kernel improvements
...
- Softmax and batched GEMM tuning
- Increase tuning runs qkv
- Softmax and batched GEMM tuning
Signed-off-by: Rajeev Rao <rajeevrao@nvidia.com >
2020-06-17 13:48:22 -07:00
98ab844a6c
Adding support for building BERT plugins with GPU_ARCHS specified ( #255 )
2019-12-05 15:48:18 -08:00
77af534c21
Moves BERT plugins into main plugin library, removes demo (will be relocated to DeepLearningExamples shortly)
...
Signed-off-by: pranavm <pranavm@nvidia.com >
2019-11-13 13:48:15 -08:00
639d11abcc
TensorRT Open Source Release/6.0
2019-09-16 14:00:59 -07:00
0d36bbb297
TensorRT Open Source Release/5.1
2019-06-17 10:47:48 -07:00
Rajeev Rao