How to get a byte aligned memory for SSE instructions.

[am28836718]

I'd like to get a aligned memory for SSE instruction on AE CS5.5. 

Does any one know the way to get an aligned memory in the AE memory management framework?

Very best regards.

[shachar carmi]

AE dispenses memory chunks by size, specified in bytes.

if i understand what you mean by "aligned" and "SSE" correctly, then isn't

it an issue of compiling structures with the correct alignment, rather than

the memory allocation?

perhaps i don't understand your question fully...

[am28836718]

Thank you for your quick reply.

I mean how to realize 16 byte memory alignment  for SSE instructions on AE.

According to the AE user manual, 16 byte memory alignment is not guaranteed .

"

The pixels in a PF_EffectWorld are not guaranteed to be 16-byte-aligned. An effect may get a subregion of a larger PF_EffectWorld. Users of Apple’s sample code for pixel processing optimization, you have been warned.

"

_aligned_malloc‌‌ can be used for this purpose but I think allocated memory is deviated from AE memory management framework.

Best regards.

[shachar carmi]

the quote from the docs refers to the way AE handles it's buffer pixels in memory. i don't see how memory you allocate has anything to do with it.

it basically says that is you get the base address of a buffer (the address of the upper left pixel) you can't rely on the next lines to start at (base + y *  width). hence you can't transfer that that pixel's address to processes that rely on that assumption.

i don't know of a way to ask AE for the buffer data at any other way...

you'll just need to copy the buffer pixels into a new and controlled memory chunk.

am i completely off here? what exactly are you trying to do, and where did you get stuck?

[am28836718]

Sorry for lack of explanation.

In my plugin, I usually allocate memory by following way.

int bufferSize = imageWidth * imageHeight * 4;

AEGP_SuiteHandler suites(in_data->pica_basicP);

PH_Handle handler = (PF_Handle)suites.HandleSuite1()->host_new_handle(sizeof(float) * bufferSize);

float *p = reinterpret_cast<T*>(suites.HandleSuite1()->host_lock_handle(handler));

// The code below does not work

for(int h = 0; h < imageHeight; h++)

{

  for(int w = 0; w < imageWidth; w++)

  {

  index = imageWidth * h * 4 + w * 4;

  __m128 s = _mm_loadu_ps(p + index);

  }

}

// dispose memory

suites.HandleSuite1()->host_unlock_handle(handler);

suites.HandleSuite1()->host_dispose_handle(handler);

The allocated memory is not aligned to a 16-byte boundary, right?

I'd like to know whether we can allocate a memory (suitable for SSE) with 16-byte boundary using AE SDK or not.

Thanks a lot.