On 05/15/2011 08:40 PM, Daniel Kang wrote:
On Sun, May 15, 2011 at 1:34 PM, Ronald S. Bultje <[email protected] <mailto:[email protected]>> wrote:Hi, On Sun, May 15, 2011 at 9:08 AM, Vitor Sessak <[email protected] <mailto:[email protected]>> wrote: [..] > + movss [outq+92], xmm7 > + addss xmm2, [outq+124] > + movss [outq+116], xmm2 The indenting of this last chunk is weird... Also, if you want to do AVX impl, shouldn't you use m instead of xmm, so you can automatically switch xmm->ymm by using INIT_AVX? INIT_AVX does not set m* to ymm registers in x86inc.asm. It was originally made to accommodate 3-instruction op-codes in an easy fashion (as there are no int instructions yet). Talk to Loren or Jason about this.
He probably meant INIT_YMM, as it is done in the FFT patch, but I found no way of sharing code between AVX and SSE, thus using this is not particular useful for this implementation.
-Vitor _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
