Jacinto 7 TIDL Release Notes
Version: 08.01.00, Date: Dec 03, 2021
Contents
- Introduction
- Licensing
- What's New
- Documentation
- Upgrade and Compatibility Information
- Device Support
- Validation Information
- Fixed Issues
- Known Issues
- Change Request
- Technical Support
- Package Versioning
Introduction
The package consists of Texas Instrument's Deep learning solution on Jacinto 7 product family
Licensing
The licensing information of this product, as well as any third-party components included which are made available under a number of other open-source licenses are enumerated as part of the manifest. Note : Below Links would work only after installing the package
Documentation
1 | User's Guide | [HTML] |
2 | Software Manifest | [HTML] |
What's New
Here are a few of the new features supported in this release:
- Support to schedule multiple models with user specified priority and allow preemption of model inference
- Support of 3D Object detection model (PointPillars ) for Lidar input
- Support of associative embedding based human pose estimation models
- Performance improvement of ONNX Runtime
- Support for TIDL with FreeRTOS on C7x-MMA
Upgrade and Compatibility Information
NA
Device Support
Target SoC | Target Plaform | Build environment (OS) | Target environment (OS) |
J721E EVM | x86_64 Linux PC | x86_64 (Linux) | NA |
J721E EVM | x86_64 Windows PC | x86_64 Windows | NA |
Note: Support of x86_64 Windows would be removed from next release onwards and only x86_64 Linux will be maintained
Validation Information
This release was built and validated using the following tools:
Dependencies (included in Processor SDK RTOS):
- ivision - 01_00
- MMALIB - 02_02_00_03
- PDK - CORE SDK 08.01.00
- C7x CGT - ti-cgt-c7000_2.0.1.STS
Dependencies (NOT included in Processor SDK RTOS):
- Microsoft Visual Studio Verison 14.2
- GCC version 5.4.0 (In Ubuntu 18.04)
- OpenCV: 4.1.0 (for Test bench Only)
- Protobuf: 3.11.3 (for Import tool only)
- Flatbuffers : 1.12.0 (for Import tool only)
Fixed Issues
ID | Description | Module | Affected Versions | Affected Platforms |
---|---|---|---|---|
TIDL-1269 | TF maxpool avgpool functionality mismatch for k=3,s=2 | COMPUTE | 08.00.01, 08.00.00, 02.00.00, 01.04.00, 01.03.00 | J721E |
TIDL-1380 | OSR : Layers with 5 dimensional input not marked as unsupported in allowlisting/td> | TOOL | 08.00.01, 08.00.00, 02.00.00 | J721E |
TIDL-1533 | MSMC holding layers cause failure due to instance memory | TOOL | 08.00.01, 08.00.00, 02.00.00, 01.04.00 | J721E |
TIDL-1561 | Tflite import : Functional issue in 2x2 convolution computation | TOOL | 08.00.01, 08.00.00 | J721E |
TIDL-1598 | Output tensor scale smaller than expected with mixed precision | TOOL | 08.00.01, 08.00.00 | J721E |
TIDL-1601 | Global average pooling with 16bit input and 8 bit output hangs on EVM | TOOL, COMPUTE | 08.00.01 | J721E |
TIDL-1603 | ONNX models with asymmetric convolution padding result in incorrect output | TOOL | 08.00.01 | J721E |
TIDL-1606 | Target and Host emulation output mismatches when input is directly accessed by CPU | COMPUTE | 08.00.01 | J721E |
TIDL-1748 | TIDL InnerProductLayer layer input needs to be Flatten | TOOL, OSRT | 08.00.01, 08.00.00 | J721E |
TIDL-1749 | Object Detection : Incorrect box predictions for last object class | COMPUTE | 08.00.01 | J721E |
TIDL-1750 | Crop layer mismatching in 16 bit | COMPUTE | 08.00.01 | J721E |
TIDL-1762 | Segmentation fault with custom layers involving parameters | COMPUTE | 08.00.01, 08.00.00 | J721E |
TIDL-1763 | TIDL Sigmoid layer EVM output is not matching with reference output for 16-bit input | COMPUTE | 08.00.01, 08.00.00 | J721E |
TIDL-1768 | Incorrect offset calculation in TIDL_getLayerInPtrs function | COMPUTE | 08.00.01, 08.00.00 | J721E |
TIDL-1769 | Incorrect outConsumerCnt calculation in tidl_tfliteLayerUpdateConsumerCount function | TOOL | 08.00.01, 08.00.00 | J721E |
TIDL-1770 | Output of 2x2 max pool for 16-bit is not matching host emulation output | COMPUTE | 08.00.01, 08.00.00 | J721E |
TIDL-1782 | Input data scales get overwritten in case of multiple inputs to network with different tensor scales | TOOL | 08.00.01 | J721E |
TIDL-1784 | Concat across different axis not supported | TOOL | 08.00.01, 08.00.00 | J721E |
TIDL-1785 | TIDL Conv layer with pad = 0 output mismatch in the host emulation mode | TOOL, COMPUTE | 08.00.01, 08.00.00, 02.00.00 | J721E |
TIDL-1790 | Memory for TIDL_refConv2d accumulator is not sufficient | COMPUTE | 08.00.01, 08.00.00 | J721E |
Known Issues
ID | Description | Module | Reported in Release | Affected Platforms | Occurrence | Workaround in this release |
---|---|---|---|---|---|---|
TIDL-864 | Concat trace dump mismatch between REF & CN/CI | COMPUTE | 01.02.00 | J721E | Rare | NA |
TIDL-1197 | Model execution order for a branch is not optimal | COMPUTE | 01.02.00 | J721E | Rare | No fucntional issue, Only Performance drop would be obsrved |
TIDL-1350 | ShuffleNetV2 ONNX-RT accuracy is 2% below compared to TVM-RT | COMPUTE | 02.00.00 | J721E | Rare | NA |
TIDL-1530 | Processing on EVM gets stuck in case of "GEMM" layer in onnx model | OSRT | 02.00.00 | J721E | Rare | set reservedCtrl=1 of TIDL_CreateParams |
TIDL-1771 | Networks with multiple inputs may give wrong output if the order after importing changes | COMPUTE | 08.00.00 | J721E | Rare | Explicitly list all the input using inDataNamesList parameter in TIDL-RT import config file in the same order as it comes after importing the model |
TIDL-1794 | Network output is missing from TIDL imported network if the same is consumed by other layers | COMPUTE | 08.00.00 | J721E | Rare | Explicitly list all the output using outDataNamesList parameter in TIDL-RT import config file |
For latest status of the issues and any new issues found post this release, please refer TI External software Incident Report. Issues on this product can be filtered with search query as project = EXT_EP AND issuetype in (Bug, Enhancement) AND Product ~ "TI Deep Learning Library"
Change Request
ID | Description | Original Fix Version | New Fix Version |
---|---|---|---|
JACINTOREQ-1836 | Additional Performance Optimization for CNN model inference | 08.01.00 | 08.02.00 |
JACINTOREQ-2066 | Performance Optimization for CNN networks(RegNetx series)with grouped convolution | 08.01.00 | 08.02.00 |
Technical Support
For technical support, please post your questions on TI E2E Forum for Processors.
For additional assistance, contact local TI Field Application Engineer
Package Versioning
Each package version is composed of 4 period-delimited numbers - represented here by the letters M, m, p and b [M.m.p.b]
. The table below provides a descriptive reference regarding package version numbering.
Digit | Meaning | Description |
---|---|---|
1 (M=Major) | Major revision | Incremented when the new version is substantially different from the previous For example, a new module added or an existing module's algorithm significantly altered. |
2 (m=minor) | Minor revision | Incremented when the new version has changed but not in a major way. For example, some minor changes in the API or feature set. |
3 (p=patch) | Patch number | Incremented for all other source code changes. This include any packaging support code. |
4 (b=build) | Build number | Incremented for each release delivery to CM. Reset for any change to M, m or p |
© Copyright 1995-2021, Texas Instruments Incorporated. All rights reserved.
Trademarks | Privacy policy | Terms of use | Terms of sale