Hi there,
We are using DM6467 for video coding and we are trying to make some not-too-demanding pre-processing in the ARM (like re-sizing, for example). We use CodeSourcery G++ Lite to compile the ARM part, but the code does not get optimized enough just by using the -O3 option. We tried to find a way to vectorize using "intrinsics" like in TI DSP chips, but it seems CodeSourcery do not support them for this ARM chip. Nevertheless, other compilers seem to support intrinsics indeed:
Realview:
http://www.keil.com/support/man/docs/armcc/armcc_CHDBJJFD.htm
Microsoft:
http://msdn.microsoft.com/en-us/library/ms924322.aspx
We wonder which of these options is the best way to optimize code in Davinci ARM chips:
1- Purchase Realview or other compiler that supports intrinsics and compile everything with it.
2- Compile the critical functions with Realview or CCS with CGTools (provided that there exist intrinsics in CGTools for ARM) and try to link the library with the rest of the application in Linux with CodeSourcery G++ Lite.
Should it be straightforward to compile critical functions with other compiler and then link the libraries with CodeSourcery G++? Or we will find some incompatibility problems on the way?
Thanks!
César