Dear supporters:
Currently we are suffering performance deterioration due to some complex algorithm modification.
We have downscaled the image to 184x120 (~ 22KB in Y domain ) for better performance.
Because the algorithm has intensive access to this buffer, I 'd like to keep this buffer in the L2 (256K) or
L1(32K) at all time. I have tried to use
#pragma DATA_ALIGN(g_aucfBuffer, 128);
#pragma DATA_SECTION(g_aucfBuffer, ".Dmod_WorkBuffer");
unsigned char g_aucfBuffer[184*120];
and added cmd file, link_dsp.cmd
SECTIONS
{
...
.Dmod_WorkBuffer > L2
}
Got error below:
"D:/EAGLE-II/src/avsk_app/link_dsp.cmd", line 12: warning: memory range not
found: L2 on page 0
"D:/EAGLE-II/src/avsk_app/link_dsp.cmd", line 12: error: no valid memory
range(NULL) available for placement of ".Dmod_WorkBuffer"
"D:/EAGLE-II/src/avsk_app/link_dsp.cmd", line 12: error: program will not fit
into available memory. run placement with alignment fails for section
".Dmod_WorkBuffer" size 0x15900
error: errors encountered during linking;
"D:/EAGLE-II/src/outfiles/avsk_app/bin/ti814x-evm/avsk_app_dsp_debug.xe674"
not built
We use TI compiler thru makefile, not thru CSS. CSS is mainly for debug purpose.
My platform info: TMS320DM8148 (Vision-Mid)
600-MHz ARM® Cortex™-A8 RISC MPU
500-MHz C674x™ VLIW DSP
200-MHz M3-ISS/M3-HDVPSS
Best regards,
Joey from Altek