AcquireMagickMemory and multi thread

[vladimir_ff]

Hello!
I have test app with two threads. Each thread looks like:

Code: Select all

//reading image from file to buffer
while(true)
{
   Blob blob( imageData, size );
   image.ping( blob );
}

I run this app on a 24-core system and expect loading two cores at 100%.
But it doesn't happen. Each of two cores works at 80-90%.
I explored source code and found this call sequence: Image::ping->PingBlob->CloneImageInfo->AcquireImageInfo->AcquireMagickMemory
There is a special memory manager with blocking (memory_semaphore), so I can't use full power of cores.
Am I right?
Is there any way to have memory manager for each thread?
Or may be other solution?

IM 6.7.6.9
CentOs 6.3

[vladimir_ff]

No idea? Nobody knows?

[magick]

ImageMagick uses the standard memory methods, malloc(), free(), etc. You can use any memory manager you want. Just call MagickCore::SetMagickMemoryMethods() with three methods with the same signature as malloc(), realloc(), and free().

[vladimir_ff]

"Our custom memory allocation manager implements a best-fit allocation policy using segregated free lists..."

This manager uses semaphores.
How can I make ImageMagick to use pure malloc/free without special policies and blocks?

[magick]

Our custom memory allocation manager does implement a best-fit allocation policy. However, it only used if you use this configure script option: --enable-zero-configuration. Its likely you did not use this option, therefore ImageMagick is using the standard malloc(), realloc(), and free() memory methods.