Attention

This version of the SDK has been superseded by the latest release of the SDK.

AMD Advanced Media Acceleration (AMA) Troubleshooting¶

Overview ¶

This section describes various troubleshooting methods and workarounds for some known issues such as out of memory errors, low frame-rate, etc.

Single Device Per Card ¶

If SDK has access to only a single device on a card, i.e.,:

lspci -d 10ee:

returns half the expected devices, then bifurication has not been enabled in the BIOS.

Workaround: Check your BIOS and enable 4x4 PCIe bifurcation on each slot with a MA35 card.

Memory Usage ¶

Messages such as:

ERROR: from element xxxxx: Internal data stream error.
ERROR [Channel] Cannot create channel: DeviceAllocate failed

...

, indicate memory pressure on the accelerator card.

Workaround: Other than the obvious over-subscription use-cases, these issues can be resolved by decrementing the lookahead buffer size.

AV1 Slow Playback ¶

An AV1 HLS stream may playback at slower than real time speeds.

Workaround: Recommend explicitly setting of FPS playback frame rate when using ffplay or use ffplay from FFmpeg 5.1.2 or 6.0.

AV1 MP4 Playback ¶

AV1 muxed into an MP4 container may not play back properly.

Workaround: Recommend playback of raw video using more recent versions of ffplay (n5.1.2 or later) or more recent versions of media players such as VLC v3.0.17.4. Alternatively, muxing the raw video into an IVF container format is also permitted.

30 FPS Density ¶

30 FPS transcodes and below can not run at full density with default lookahead depth.

Workaround: Decrease lookahead depth or density. Required decrease is dependent upon resolution/frame rate. See Performance Tables Section for details.

Gstreamer Variable Frame Rate ¶

Variable frame rate files are not supported by the Gstreamer AMA plugins.

Workaround: Turn XRM off and add a videorate filter and caps filter before the encoder to set a fixed frame rate.

400 Mbps Max Bit Rate ¶

Encoders do not support a target or max bitrate above 400 Mbps.

Workaround: This issue can be avoided by constraining the bitrate values below to 400 Mbps.

Many Parallel Encodes ¶

Encountering Cannot create channel: DeviceAllocate failed error, when running high number of encoding operations in parallel.

Workaround: Reduce the memory requirements by lowering the lookahead depth of the encoding operation. This can be controlled with the -lookahead_depth argument. Default values depend on the FPS and bitdepth of the input source. The following are some starting points:

30fps: -lookahead_depth 26 (reduce by steps of 2)

60fps: -lookahead_depth 46 (reduce by steps of 4)

HDR & spatialAqGain ¶

Encoding HDR content with a spatialAqGain of 1 or 2 is not supported.

Workaround: Using the default values of spatialAqGain is recommended for best video quality when possible. If it is desirable to reduce the spatialAqGain from the recommended range of 60 - 100, using a value of 3 or greater will avoid this issue.

Nonresponsive Devices ¶

It is possible that codec services and utilities may become unavailable, due to crashed SDK driver and as such mamgmt reset will not be able to reset any of the available devices.

Workaround: Execute the following commands:

sudo rmmod ama_transcoder
sudo modprobe ama_transcoder

If removal and reloading of the driver does not work, then system reboot is required.

Codec Issues in VM ¶

If codec operations are hanging in a VM, while the following sudo dmesg -w output log is observed:

...
[  323.436842] ama_transcoder0 0000:05:00.0 hdma: warning:hdma_link_rc2ep_xfer status is done, c:2,status=0x1,condition=0
[  323.440802] ama_transcoder0 0000:05:00.0 hdma: dir:rc2ep element_cnt=1 channel:2 link_table_pa:0x160000
[  323.443387] ama_transcoder0 0000:05:00.0 hdma: ctl:0x01 size:0x4 sh:0x2 sl:0x97e00000 dh:0x0 dl:0x20831000
[  323.445538] ama_transcoder0 0000:05:00.0 hdma: end ctrl:0x06 rsv:0x0 llp_h:0x0 llp_l:0x160000
[  323.447257] ama_transcoder0 0000:05:00.0 hdma: rc2ep PF c=2 0x500 = 0x1
...

, with host reporting IO_PAGE_FAULT in its dmesg logs, then this is an indication of a mis-configured VM.

Workaround: Ensure that VM has been properly created as per Virtualization.

Cannot Assign a VF to a VM ¶

If passing a VF to a VM fails, this could be due to VF not getting its own IOMMU group. To check this, run the following command:

for a in /sys/kernel/iommu_groups/*; do find $a -type l; done | sort --version-sort

, and ensure that VF is assign to a unique group. As an example, the following output shows that VF associated with device 0000:02:00.0 , i.e., 0000:02:00.1, has its own unique group 34:

...
/sys/kernel/iommu_groups/34/devices/0000:02:00.1
...

Workaround: Ensure that ACS has been enabled in the BIOS.

Cannot Reach Stated FPS ¶

While running typical pipelines, if the reported Frame Per Second (FPS) is below the reported one, ensure that Active State Power Management (ASPM) is disabled. To check the status of ASPM, issue the following command:

sudo lspci -vvvd 10ee: | grep -i aspm

and check that ASPM is disabled:

LnkCap: Port #0, Speed 16GT/s, Width x4, ASPM L1, Exit Latency L1 <64us
      ClockPM- Surprise- LLActRep- BwNot- ASPMOptComp+
LnkCtl: ASPM Disabled; RCB 64 bytes, Disabled- CommClk+

Workaround: Ensure that ASPM is disabled as per pcie_aspm.policy, in Chassis Setup.

Cannot Flash the Card ¶

While attempting to flash from satellite controller version 9.7.35, the following message may appear:

...
Device: 0000:e2:00.0
**** ERROR Programming BMC-MSP432.bin of type [SC] SC is accessing flash, pls try FW update after sometime.
...

Workaround: Wait for 10 minutes and try to flash again.