[dabekjakub]

Added buffer alignment API.
Needed to set alignment that different buffers require.
Example: Dma buffer has to be aligned to 128 due to hardware
configuration.

Signed-off-by: Jakub Dabek jakub.dabek@intel.com

[jajanusz]

ACK

[lgirdwood]

We should not force all malloc() callers to request an alignment (as most dont care). So we should have

void *malloc(size);
void *malloc_align(size, alignment);

where malloc() internally just calls malloc_align(size, 0)

[dabekjakub]

@lgirdwood With @tlauda we asumed that aligning every allocation to casheline will improve overall performance.

[jajanusz]

Please change order of commits to not break git bisect for host.

[dbaluta]

@dabekjakub Please use imperative mood in the commit subject line https://chris.beams.io/posts/git-commit/#imperative

e.g host: fixed memory function definitions -> host: Fix memory function definitions

Also, commit message:

Fixed alloc definitions for host so it does not relay on define being const free.

What does this even mean? Please try to make your commit messages more clear. People should get a feel of your change by reading the commit message.

[dbaluta]

@dabekjakub there is still one commit subject that needs to be fixed. Please be consistent :)

[jajanusz]

SOFCI TEST

[tlauda]

Could you add additional commit to this PR, where you are adding alignment also to rmalloc_sys and by default calling it with PLATFORM_DCACHE_ALIGN? Implementation itself is already in place, so it's just a minor tweak.

[dabekjakub]

@tlauda changes implemented , as far as i can see rmalloc_sys has selalign in it already

/* align address to dcache line size */
if (cpu_heap->info.used % PLATFORM_DCACHE_ALIGN)
	alignment = PLATFORM_DCACHE_ALIGN -
		(cpu_heap->info.used % PLATFORM_DCACHE_ALIGN);

[lgirdwood]

We need to use the ISA macro, #1826 has the ISA fix (pending on CI updates) - I will merge it later today since CI qemu updates have been merged.

[tlauda]

We need to use the ISA macro, #1826 has the ISA fix (pending on CI updates) - I will merge it later today since CI qemu updates have been merged.

We need to use the ISA macro, but also have our own additional macro for cache enablement. I think that right now the cache is only enabled on cAVS platforms, so there is no point in aligning to cache line in this case.

[lgirdwood]

We need to use the ISA macro, #1826 has the ISA fix (pending on CI updates) - I will merge it later today since CI qemu updates have been merged.

We need to use the ISA macro, but also have our own additional macro for cache enablement. I think that right now the cache is only enabled on cAVS platforms, so there is no point in aligning to cache line in this case.

Sorry, not following, do you mean there is no point aliging to cache line size on CAVS platforms ?

[tlauda]

We need to use the ISA macro, #1826 has the ISA fix (pending on CI updates) - I will merge it later today since CI qemu updates have been merged.

We need to use the ISA macro, but also have our own additional macro for cache enablement. I think that right now the cache is only enabled on cAVS platforms, so there is no point in aligning to cache line in this case.

Sorry, not following, do you mean there is no point aliging to cache line size on CAVS platforms ?

I'm saying there is no point aligning to cache line on non-cAVS platforms.

[lgirdwood]

@tlauda ok, but I'm hesistant here to just use an alignment that does not align with core-isa (even though cache is not implemented). I'm suspiscious that the cache line size value is defined as non zero and that may also be used by compiler for optimised text/data alignment, prefeching or some other undocumented hint used by the compiler to generate optimised code.

[dabekjakub]

SOFCI TEST

[dabekjakub]

@tlauda @lgirdwood it looks good to me now - bugs fixed

Stale review, PR uses correct alignment option

[jajanusz]

Checkpatch errors are related to code that was already there and are not effect of this PR

[paulstelian97]

Any reason why rebasing on top of the last commit breaks my platform on playback? I'm not seeing any error trace related to some buffer failing to allocate, however the actual alignment looks weird enough to me to need to take a closer look.

[paulstelian97]

Huh, I think you need to return with a fix for when CONFIG_HEAP_DEBUG is enabled. Currently (as-merged) rballoc macro is defined twice instead of defining rballoc and rballoc_align.

[dabekjakub]

@paulstelian97 when it comes to playback I saw no issues and CI went through with no problem.

DEBUG_HEAP Fix on the way.

[paulstelian97]

Yeah, I think something else actually caused issues with my SYS_RT heap being too small. I will continue to investigate my issues separately. Whenever I identify an upstream commit I will ask about it.