| 15:59 | dbsx | with gcc-4.8 are -mcpu=marvell-pj4 -mtune=iwmmxt |
| 15:59 | dbsx | reasonable options or do we take defaults? |
| 16:16 | dv_ | dbsx: is -mfpu=iwmmxt possible? |
| 16:32 | dbsx | dv_: not sure |
| 16:34 | dbsx | dv_: the answer is no. |
| 16:35 | dbsx | iwmmxt can be used with -mcpu= and -mtune according to gcc-4.8 --target-help |
| 16:40 | dv_ | I think mtune=iwmmxt means something different |
| 16:40 | dv_ | it sounds like the intel xscale stuff |
| 16:40 | dv_ | reminds me of the "iwmmxt" support in OpenEmbedded, which is actually tailored towards xscale, and expects an armv5te instruction set |
| 16:41 | dv_ | ask in the #gcc channel to be sure (I dont think its on freenode though) |
| 16:47 | dbsx | I will do that dv. My curiosity is a whether iwmmxt instructions on the PJ4 can be used for memcmp, memcpy, strlen etc and whether it is a faster on large strings and if so by how much. |
| 17:07 | dv_ | dbsx: I would be curious about your results |
| 17:08 | dv_ | dbsx: oh, I also wonder if by only -mcpu=marvell-pj4 alone the iWMMXt assembler instrinsics are enabled |
| 17:08 | dv_ | (relevant for ffmpeg / libav acceleration |
| 17:15 | dbsx | will let you know what I find out. jnettlet was saying that gcc-4.8 has iwmmxt optimizations by default |
| 18:02 | RandomPixels | hello |