[DNM] [DRAFT] heap: Increase virtual heap from 1MB to 1.5MB#10499
[DNM] [DRAFT] heap: Increase virtual heap from 1MB to 1.5MB#10499serhiy-katsyuba-intel wants to merge 1 commit intothesofproject:mainfrom
Conversation
TEST Signed-off-by: Serhiy Katsyuba <serhiy.katsyuba@intel.com>
| { 4096, 11}, | ||
| { 8192, 10}, | ||
| { 65536, 3}, | ||
| { 32768, 8}, |
There was a problem hiding this comment.
BTW, this is not just an increase in size but also a different configuration of memory blocks. Is this the optimal split?
There was a problem hiding this comment.
Yes, new extra 0.5 MB are used to introduce new 32 kB and add more 64 kB blocks. Before this change practical maximum buffer size was 8 kB, as we could only allocate max 5 buffers larger then 8 kB. That number of buffer is usually not enough even for simple topology.
Is this the optimal split?
That's actually the problem with our current implementation: we have to define blocks configuration at compile time and as we have to support formats from 8 kHz to 192 kHz, 1 to 8 channels, 1 ms buffers to 100 ms buffers, there could be a problem to define optimal split. Now half of the memory is sort of "reserved" for KPB buffer -- that seems quite un-optimal.
Maybe the proper solution would be to re-implement virtual heap allocator to eliminate the need to define any blocks at compile time. Similarly like this is currently done for Zephyr heap. But that's not a quick fix.
2 participants
*** Do not merge ***
TEST