> +INIT_XMM sse2 > +cglobal conv_s16p_to_flt_2ch, 3,4,6, dst, src0, len, src1 > + lea lenq, [2*lend] > + mov src1q, [src0q+gprsize] > + mov src0q, [src0q ] > + lea dstq, [dstq+4*lenq] > + add src0q, lenq > + add src1q, lenq > + neg lenq > + mova m5, [pf_s32_inv_scale] > + ALIGN 16 > +.loop: > + mova m2, [src0q+lenq] > + mova m3, [src1q+lenq] > + punpcklwd m0, m2, m3 > + punpckhwd m1, m2, m3
Avoid the extra mova, as in previous emails. > + pxor m2, m2 > + pxor m3, m3 > + punpcklwd m2, m0 > + punpckhwd m3, m0 > + pxor m0, m0 > + pxor m4, m4 > + punpcklwd m0, m1 > + punpckhwd m4, m1 Same xor comment as in previous emails. > + cvtdq2ps m2, m2 > + cvtdq2ps m3, m3 > + cvtdq2ps m0, m0 > + cvtdq2ps m4, m4 > + mulps m2, m5 > + mulps m3, m5 > + mulps m0, m5 > + mulps m4, m5 > + mova [dstq+4*lenq ], m2 > + mova [dstq+4*lenq+ mmsize], m3 > + mova [dstq+4*lenq+2*mmsize], m0 > + mova [dstq+4*lenq+3*mmsize], m4 > + add lenq, mmsize > + js .loop Any reason why .js instead of .jl? > + REP_RET Otherwise looks okay. Jason _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
