vdec_create_chn

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product / Atlas A3 inference product	√
Atlas A2 training product / Atlas A2 inference product	√
Atlas training product	x
Atlas inference product	√
Atlas 200I/500 A2 inference product	√

Description

Creates a decoding channel based on the configured channel attributes.

Prototype

C Prototype

        
             hi_s32hi_mpi_vdec_create_chn(hi_vdec_chnchn,consthi_vdec_chn_attr*attr)

Python Function

        
             ret=acl.himpi.vdec_create_chn(chn,attr)

Parameters

Parameter	Description
chn	Int, decoding channel ID. Atlas 350 Accelerator Card: The value range is [0, 256). The maximum number of JPEGD channels is 256. Atlas inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256. Atlas A2 training product / Atlas A2 inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256. The maximum number of JPEGD channels is 256, and the maximum number of VDEC channels is 32. Atlas 200I/500 A2 inference product : The value range is [0, 128). The JPEGD and VDEC functions share channels, and the maximum number of channels is 128. Atlas A3 training product / Atlas A3 inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256. The maximum number of JPEGD channels is 256, and the maximum number of VDEC channels is 32. NOTE: For Ascend virtual instances on the Atlas inference product , total number of VDEC and JPEGD channels = (Total number of allocated VDEC and JPEGD hardware units/Total number of VDEC and JPEGD hardware units) x 256. If the total number of channels is not an integer, round down the value. For Ascend virtual instances on the Atlas A2 training product / Atlas A2 inference product , number of VDEC channels = (Number of allocated VDEC hardware units/Total number of VDEC hardware units) x 32. If the total number of channels is not an integer, round down the value. The number of JPEGD channels is not affected by the computing power. However, the maximum number of JPEGD+VDEC channels is 256. For the Atlas 200I/500 A2 inference product , in the Ascend virtual instance scenario, the number of VDEC channels is calculated as follows: Number of VDEC channels = (Number of allocated VDEC hardware units/Number of VDEC hardware units) x 128. If the total number of channels is not an integer, round down the value. The number of JPEGD channels is not affected by the computing power. However, the maximum number of JPEGD+VDEC channels is 128. To view the computing power specifications of Ascend virtual instances in different scenarios, run the npu-smi info -t template-info command on the server where AI processor is installed.
attr	Dict, dictionary of decoding channel attributes. For details, see hi_vdec_chn_attr. If attr is left empty, the error code HI_ERR_VDEC_NULL_PTR is returned. When a setting in the channel attributes pointed by attr exceeds the decoding capability set, the error code HI_ERR_VDEC_ILLEGAL_PARAM is returned.

Parameter

Description

chn

Int, decoding channel ID.

Atlas 350 Accelerator Card: The value range is [0, 256). The maximum number of JPEGD channels is 256.
Atlas inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256.
Atlas A2 training product / Atlas A2 inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256. The maximum number of JPEGD channels is 256, and the maximum number of VDEC channels is 32.
Atlas 200I/500 A2 inference product : The value range is [0, 128). The JPEGD and VDEC functions share channels, and the maximum number of channels is 128.
Atlas A3 training product / Atlas A3 inference product : The value range is [0, 256). The JPEGD and VDEC functions share channels, and the maximum number of channels is 256. The maximum number of JPEGD channels is 256, and the maximum number of VDEC channels is 32.

NOTE:

For Ascend virtual instances on the Atlas inference product , total number of VDEC and JPEGD channels = (Total number of allocated VDEC and JPEGD hardware units/Total number of VDEC and JPEGD hardware units) x 256. If the total number of channels is not an integer, round down the value.

For Ascend virtual instances on the Atlas A2 training product / Atlas A2 inference product , number of VDEC channels = (Number of allocated VDEC hardware units/Total number of VDEC hardware units) x 32. If the total number of channels is not an integer, round down the value. The number of JPEGD channels is not affected by the computing power. However, the maximum number of JPEGD+VDEC channels is 256.

For the Atlas 200I/500 A2 inference product , in the Ascend virtual instance scenario, the number of VDEC channels is calculated as follows: Number of VDEC channels = (Number of allocated VDEC hardware units/Number of VDEC hardware units) x 128. If the total number of channels is not an integer, round down the value. The number of JPEGD channels is not affected by the computing power. However, the maximum number of JPEGD+VDEC channels is 128.

To view the computing power specifications of Ascend virtual instances in different scenarios, run the npu-smi info -t template-info command on the server where AI processor is installed.

attr

Dict, dictionary of decoding channel attributes. For details, see hi_vdec_chn_attr.

If attr is left empty, the error code HI_ERR_VDEC_NULL_PTR is returned.
When a setting in the channel attributes pointed by attr exceeds the decoding capability set, the error code HI_ERR_VDEC_ILLEGAL_PARAM is returned.

Return Value

Return Value	Description
ret	Int, error code. 0 indicates success. Other values indicate failure. For details, see Common Return Codes.

Restrictions

The channel ID of a single device must not be greater than the maximum channel ID.
Before this API call, ensure that a decoding channel has not been created yet or has been destroyed. Otherwise, a failure is returned.
If the OS memory is insufficient, the error code HI_ERR_VDEC_NO_MEM is returned. In this case, you can expand the OS memory as needed.
If the H.264 stream to be decoded has B frames or the H.265 stream to be decoded supports temporal motion vector prediction (sps_temporal_mvp_enabled_flag = 1), you need to set temporal_mvp_en to 1 when creating a channel. In addition, you need to allocate a video buffer (VB) for outputting the temporal motion vector (Tmv) information of each frame. The size of the VB is much smaller than that of the image VB. The number of required VBs is RefFrameNum + 1. The specific size can be obtained by calling the acl.himpi.vdec_get_tmv_buf_size function, otherwise, errors such as artifacts may occur during decoding.
For H.264 and H.265 decoding, the minimum VB allocation per decoding channel is the sum of the number of reference frames and the number of display frames plus 1. For JPEG decoding, the minimum VB allocation per decoding channel is the number of display frames plus 1. The requested VB allocation varies according to the decode protocol and can be queried by using the acl.himpi.vdec_get_pic_buf_size call.
To skip decoding B-frames in H.264 streams or skip decoding H.265 streams with temporal MVP support (sps_temporal_mvp_enabled_flag = 1), set temporal_mvp_en to 0 for the created channel. In this case, no TMV information is output, and therefore you do not need to allocate a TMV VB pool, which saves memory.
Set the requested frame VB size (frame_buf_size) and count (frame_buf_cnt) as well as the requested TMV VB size (tmv_buf_size) based on the stream to be decoded. The decoder allocates private VB pools based on these configurations.
As long as the allocated frame buffer and TMV buffer are large enough, the decoding channel can decode a stream with any resolution in the valid range. The channel width and height are related only to the size of the stream buffer and SCD buffer.

Reference

For details about the API call sequence, see JPEGE and VENC.

Parent topic: VDEC/JPEGD APIs