[BUG] [HIFI4] PeakVol.Quality issues for audio formats

[gbernatxintel]

Describe the bug
There is a problem processing signals of 11025, 22050, 44100, 88200, 176400 Hz by the HIFI4 module. The main problem is that, for example, for the frequency f=44.1kHz frame is 44 or 45. Some samples are not processed.

To Reproduce

1. Run the test:
   pytest -s "tests/avs/fw_01_base_fw_modules/test_09_peak_vol.py::TestPeakVolFull::test_01_09_peakvol_quality[44100Hz_16in16bit_1ch-0]"

Reproduction Rate
100%

Expected behavior
The entire signal should be processed correctly and no glitches should occur

Impact
It is not possible to use HIFI4, for MTL. Using HIFI3 does not use the full capabilities of the platform. The program runs sub optimally and uses several percent more CPU cycles than when using HIFI4

Environment

1. Branch name and commit hash of the 2 repositories: sof (firmware/topology) and linux (kernel driver).
   • Kernel: 53ddff639562ef68dc0a6f62b82f7505c75ebdce
   • SOF: 0bda13d
2. Name of the platform(s) on which the bug is observed.
   • Platform: MTL
3. Topology:

                +--------------------------------------------------+
        +----+  | +------+   +-------+   +----------+   +-------+  |
        |    |--->|Copier|-->|PeakVol|-->|PCM Copier|-->| MixIn |-----+
        |    |  | +------+   +-------+   +----------+   +-------+  |  |
        |    |  +--------------------------------------------------+  |
        |    |                                                        |
        |HOST|            +----------------------------+              |
        |    |            |  +--------+   +--------+   |              |
        |    | <-------------| Copier |<--| MixOut |<-----------------+
        |    |            |  +--------+   +--------+   |
        |    |            +----------------------------+
        +----+

Notes
A PR has been added that fixes the problem for frequencies of 88.2kHz and above. However, it does not solve the problem completely. You can find the link here:
#9158

Screenshots or console output

cc:

• [BUG] MFCC HiFi4 optimized version is broken #9209

[marc-hb]

@gbernatxintel can you please confirm that any other HiFi4 hardware other than MTL is or would be affected in the same way? In other words, a pure software bug?

EDIT: this is what #9208 seems to imply. Please confirm.

[gbernatxintel]

can you please confirm that any other HiFi4 hardware other than MTL is or would be affected in the same way? In other words, a pure software bug?

The problem is replicated on the LNL platform. The way it works is exactly the same as for MTL. The output signal looks the same for LNL when using HIFI4.

The fix 9208 that is here fixes the problem for the LNL platform

[marc-hb]

@gbernatxintel can this be closed?

[lgirdwood]

@singalsu can this be closed ?

[singalsu]

@singalsu can this be closed ?

No, I'm working with this issue. The merged fix is a workaround. @gbernatxintel has started a real fix and I'm contributing to that work.

[singalsu]

Small progress, I'm now able replicate the issue with Linux & SOF topology in LNL platform.

[singalsu]

@wszypelt Would it be possible to check if #9408 fixes the issue with test pytest -s "tests/avs/fw_01_base_fw_modules/test_09_peak_vol.py::TestPeakVolFull::test_01_09_peakvol_quality[44100Hz_16in16bit_1ch-0]"