DPUCVDX8H Feature Support - 1.1 English

DPUCVDX8H for Convolutional Neural Networks Product Guide (PG403)
Document ID
PG403
Release Date
2023-01-23
Version
1.1 English

The DPU IP provides a few fixed configurations by different XO files. The configuration includes the number of processing engines, the different kernel/filter size in element-wise, and pooling.

The deep neural network features and the associated parameters supported by the DPU are shown in the following table:

Table 1. Deep Neural Network Features and Parameters Supported by DPU
Features Description (channel_parallel=64, bank_depth=256)
Convolution Kernel Sizes W, H: [1, 16]
Strides W, H: [1, 4]
Pad_left/Pad_right [0, (kernel_w - 1) * dilation_w]
Pad_top/Pad_bottom [0, (kernel_h - 1) * dilation_h]
Input Size kernel_w * kernel_h * ceil(input_channel / channel_parallel) <= bank_depth
Output Size output_channel <= 256 * channel_parallel
Activation ReLU, LeakyReLU, ReLU6, Hard-Swish, Hard-Sigmoid
Dilation dilation * input_channel <= 256 * channel_parallell
depthwise-conv2d 1 Kernel Sizes W, H: [1, 8]
Strides W, H: [1, 4]
Pad_left/Pad_right [0, (kernel_w - 1) * dilation_w + 1]
Pad_top/Pad_bottom [0, (kernel_h - 1) * dilation_h + 1]
In Size kernel_w * kernel_h * ceil(input_channel / channel_parallel) <= bank_depth
Out Size output_channel <= 256 * channel_parallel
Activation ReLU, ReLU6
Dilation dilation * input_channel <= 256 * channel_parallell
transposed-conv2d Kernel Sizes kernel_w/stride_w, kernel_h/stride_h: [1, 16]
Strides
Pad_left/Pad_right [1, kernel_w-1]
Pad_top/Pad_bottom [1, kernel_h-1]
Out Size output_channel <= 256 * channel_parallel
Activation ReLU, LeakyReLU, ReLU6, Hard-Swish, Hard-Sigmoid
depthwise-transposed-conv2d 1 Kernel Sizes kernel_w/stride_w, kernel_h/stride_h: [1, 8]
Strides
Pad_left/Pad_right [1, kernel_w-1]
Pad_top/Pad_bottom [1, kernel_h-1]
Out Size output_channel <= 256 * channel_parallel
Activation ReLU, ReLU6

max-pooling/

average-pooling

(MISC unit in PL)

 Kernel Sizes

2/4/6pe: W,H: [1, 8] W==H

8pe_normal: W,H:{1,2,3,7} W==H
Strides W: [1, 8] H: [1, 8]
Pad_left/Pad_right [1, kernel_w-1]
Pad_top/Pad_bottom [1, kernel_h-1]
Activation Not supported
elementwise-sum

(MISC unit in PL)

 Input channel input_channel <= 256 * channel_parallel
Activation ReLU

max-pooling/

average-pooling

(MISC unit on AI Engine)

 Kernel Sizes 2/4/6/8pe: W,H: [1, 128]
Strides W: [1, 128] H: [1, 128]
Pad_left/Pad_right [1, kernel_w-1]
Pad_top/Pad_bottom [1, kernel_h-1]
Activation Not supported
elementwise-sum

(MISC unit on AI Engine)

 Input channel input_channel <= 128 * channel_parallel
Activation ReLU, Hard-Sigmoid
elementwise-multi

(MISC unit on AI Engine)

 Input channel input_channel <= 128 * channel_parallel
Activation ReLU, Hard-Sigmoid
The different configuration are as follows:
  1. 8pe_normal: Does not support depthwise and MISC unit in PL.
  2. 6pe_dwc: depthwise on AIE and MISC unit in PL.
  3. 6pe_misc: Does not support depthwise and MISC unit on AIE.
  4. 4pe_miscdwc/2pe_miscdwc: depthwise and MISC unit on AIE.