diff options
Diffstat (limited to 'celt/x86/celt_lpc_sse.c')
-rw-r--r-- | celt/x86/celt_lpc_sse.c | 51 |
1 files changed, 10 insertions, 41 deletions
diff --git a/celt/x86/celt_lpc_sse.c b/celt/x86/celt_lpc_sse.c index 67e5592a..12a9b0e5 100644 --- a/celt/x86/celt_lpc_sse.c +++ b/celt/x86/celt_lpc_sse.c @@ -40,63 +40,32 @@ #if defined(FIXED_POINT) -void celt_fir_sse4_1(const opus_val16 *_x, +void celt_fir_sse4_1(const opus_val16 *x, const opus_val16 *num, - opus_val16 *_y, + opus_val16 *y, int N, int ord, - opus_val16 *mem, int arch) { int i,j; VARDECL(opus_val16, rnum); - VARDECL(opus_val16, x); __m128i vecNoA; opus_int32 noA ; SAVE_STACK; ALLOC(rnum, ord, opus_val16); - ALLOC(x, N+ord, opus_val16); for(i=0;i<ord;i++) rnum[i] = num[ord-i-1]; - for(i=0;i<ord;i++) - x[i] = mem[ord-i-1]; - - for (i=0;i<N-7;i+=8) - { - x[i+ord ]=_x[i ]; - x[i+ord+1]=_x[i+1]; - x[i+ord+2]=_x[i+2]; - x[i+ord+3]=_x[i+3]; - x[i+ord+4]=_x[i+4]; - x[i+ord+5]=_x[i+5]; - x[i+ord+6]=_x[i+6]; - x[i+ord+7]=_x[i+7]; - } - - for (;i<N-3;i+=4) - { - x[i+ord ]=_x[i ]; - x[i+ord+1]=_x[i+1]; - x[i+ord+2]=_x[i+2]; - x[i+ord+3]=_x[i+3]; - } - - for (;i<N;i++) - x[i+ord]=_x[i]; - - for(i=0;i<ord;i++) - mem[i] = _x[N-i-1]; #ifdef SMALL_FOOTPRINT for (i=0;i<N;i++) { - opus_val32 sum = SHL32(EXTEND32(_x[i]), SIG_SHIFT); + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); for (j=0;j<ord;j++) { - sum = MAC16_16(sum,rnum[j],x[i+j]); + sum = MAC16_16(sum,rnum[j],x[i+j-ord]); } - _y[i] = SATURATE16(PSHR32(sum, SIG_SHIFT)); + y[i] = SATURATE16(PSHR32(sum, SIG_SHIFT)); } #else noA = EXTEND32(1) << SIG_SHIFT >> 1; @@ -107,22 +76,22 @@ void celt_fir_sse4_1(const opus_val16 *_x, opus_val32 sums[4] = {0}; __m128i vecSum, vecX; - xcorr_kernel(rnum, x+i, sums, ord, arch); + xcorr_kernel(rnum, x+i-ord, sums, ord, arch); vecSum = _mm_loadu_si128((__m128i *)sums); vecSum = _mm_add_epi32(vecSum, vecNoA); vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT); - vecX = OP_CVTEPI16_EPI32_M64(_x + i); + vecX = OP_CVTEPI16_EPI32_M64(x + i); vecSum = _mm_add_epi32(vecSum, vecX); vecSum = _mm_packs_epi32(vecSum, vecSum); - _mm_storel_epi64((__m128i *)(_y + i), vecSum); + _mm_storel_epi64((__m128i *)(y + i), vecSum); } for (;i<N;i++) { opus_val32 sum = 0; for (j=0;j<ord;j++) - sum = MAC16_16(sum, rnum[j], x[i + j]); - _y[i] = SATURATE16(ADD32(EXTEND32(_x[i]), PSHR32(sum, SIG_SHIFT))); + sum = MAC16_16(sum, rnum[j], x[i+j-ord]); + y[i] = SATURATE16(ADD32(EXTEND32(x[i]), PSHR32(sum, SIG_SHIFT))); } #endif |