-#if defined(USE_SSE2)
- int i;
- union xmm_t *xmm_coeffs;
- union xmm_t *xmm_hist;
-
- fir->history[fir->curr_pos] = sample;
- fir->history[fir->curr_pos + fir->taps] = sample;
-
- xmm_coeffs = (union xmm_t *)fir->coeffs;
- xmm_hist = (union xmm_t *)&fir->history[fir->curr_pos];
- i = fir->taps;
- pxor_r2r(xmm4, xmm4);
- /* 16 samples per iteration, so the filter must be a multiple of 16 long. */
- while (i > 0) {
- movdqu_m2r(xmm_coeffs[0], xmm0);
- movdqu_m2r(xmm_coeffs[1], xmm2);
- movdqu_m2r(xmm_hist[0], xmm1);
- movdqu_m2r(xmm_hist[1], xmm3);
- xmm_coeffs += 2;
- xmm_hist += 2;
- pmaddwd_r2r(xmm1, xmm0);
- pmaddwd_r2r(xmm3, xmm2);
- paddd_r2r(xmm0, xmm4);
- paddd_r2r(xmm2, xmm4);
- i -= 16;
- }
- movdqa_r2r(xmm4, xmm0);
- psrldq_i2r(8, xmm0);
- paddd_r2r(xmm0, xmm4);
- movdqa_r2r(xmm4, xmm0);
- psrldq_i2r(4, xmm0);
- paddd_r2r(xmm0, xmm4);
- movd_r2m(xmm4, y);
-#elif defined(__bfin__)