Spaces:
Sleeping
Sleeping
Ticket Name: Linux/TDA2: How does dual core C66 parallel the same CNN? | |
Query Text: | |
Part Number: TDA2 Other Parts Discussed in Thread: TDA3MA Tool/software: Linux The development board I use now is TDA2X, and SDK uses SDK_VISION_03_02_00_00. But we're going to use TDA3MA, which has two C66 and one EVE, and I don't know how these three cores work together. I have four CNNs to implement, and my idea is that I make a C66 link in which I will implement three CNNs (two C66 in parallel) and build an EVE LINK and the last CNN to implement in EVE, right? Then a LINK uses a single C66 to implement a CNN I know how to do it, but how about two C66 processes a CNN in parallel? Is there a sample use dual core parallel processing a algorithm in SDK? | |
Responses: | |
Hi, Refer below User guides of TIDL in VSDK, we had TIDL OD use case which uses multiple EVE cores and C66 for processing. 1. <install_dir>\vision_sdk\docs\FeatureSpecificUserGuides\VisionSDK_UserGuide_TI_DeepLearning.pdf <install_dir>\vision_sdk\vision_sdk\apps\src\rtos\usecases\tidl_OD\chains_tidlOD.jpg 2. <install_dir>\ti_components\algorithms\REL.TIDL.01.01.01.00\modules\ti_dl\docs\TIDeepLearningLibrary_UserGuide.pdf Thanks, Praveen | |
Hi Praveen: in my sdk have ti_SDK_VISION_03_02_00_00\vision_sdk\apps\src\rtos\usecases\tidl no vision_sdk\vision_sdk\apps\src\rtos\usecases\tidl_OD tidl is tidl_OD the same ??? in tidl have chains_tidlEveTda3x.jpg use eve1 only | |
in VisionSDK_UserGuide_TI_DeepLearning.pdf Entire input frame is processed on a single core and there are 2 processing pipelines which process the input frames alternatively: this mean the same frame is copy processed same time by DSP1 and DSP2 ,DSP1 is done cnn1 and cnn2 ? | |
Hi, Please refer to latest VSDK release ti_SDK_VISION_03_04_00_00 , the mentioned use case is not there in the ti_SDK_VISION_03_02_00_00 release. Thanks, Praveen | |
Hi:Praveen in VisionSDK_UserGuide_TI_DeepLearning.pdf Entire input frame is processed on a single core and there are 2 processing pipelines which process the input frames alternatively: this mean the same frame is copy processed same time by DSP1 and DSP2 ,DSP1 is done cnn1 and cnn2 ? Shuai | |
Hi Shuai, Yes, your understanding is right. Thanks, Praveen | |