I am using DMA in a custom video processing algorithm, based on universal_copy.
I don't want to use codec engine to manage caching for my codec, both before and after the process call, both for input and output buffers (because of DMA)
What should I set the skel scaching policy in the server to? (NONE might cause issues with other codecs, so might not be an option)
I checked on the universal_skel.c file (below).
Will the line in red , where all inBufs and outBufs are invalidated before process call, cause performance issues?
if (SKEL_cachingPolicy == SKEL_LOCALBUFFERINVWB) {
/* invalidate cache for all input buffers */
for (i = 0, numBufs = 0; i < XDM_MAX_IO_BUFFERS; i++) {
if (inBufs.descs[i].buf != NULL) {
/* valid member of sparse array, manage it */
Memory_cacheInv(inBufs.descs[i].buf,
inBufs.descs[i].bufSize);
if (++numBufs == inBufs.numBufs) {
break;
}
}
}
Is there manageInBufsCache for Iuniversal type algorithms, to manage cache policy before process call? Or is Iuniversal not recommended for algorithms that use DMA?
