Replace the sinc4 resampler with cubic

Turns out the C version of the cubic resampler is just slightly faster than
even the SSE3 version of the FIR4 resampler. This is likely due to not using a
64KB random-access lookup table along with unaligned loads, both offseting the
gains from SSE.

1 files changed, 0 insertions, 69 deletions

diff --git a/Alc/mixer_sse41.c b/Alc/mixer_sse41.c
index 354d16b7..4f88d540 100644
--- a/Alc/mixer_sse41.c
+++ b/Alc/mixer_sse41.c
@@ -84,72 +84,3 @@ const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state),
     }
     return dst;
 }
-
-const ALfloat *Resample_fir4_SSE41(const InterpState *state,
-  const ALfloat *restrict src, ALsizei frac, ALint increment,
-  ALfloat *restrict dst, ALsizei numsamples)
-{
-    const ALfloat (*restrict filter)[4] = ASSUME_ALIGNED(state->sinc4.filter, 16);
-    const __m128i increment4 = _mm_set1_epi32(increment*4);
-    const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
-    union { alignas(16) ALint i[4]; float f[4]; } pos_;
-    union { alignas(16) ALsizei i[4]; float f[4]; } frac_;
-    __m128i frac4, pos4;
-    ALint pos;
-    ALsizei i;
-
-    InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
-
-    frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
-    pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
-
-    --src;
-    for(i = 0;numsamples-i > 3;i += 4)
-    {
-        const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]]);
-        const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
-        const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
-        const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
-        __m128 k0 = _mm_load_ps(filter[frac_.i[0]]);
-        __m128 k1 = _mm_load_ps(filter[frac_.i[1]]);
-        __m128 k2 = _mm_load_ps(filter[frac_.i[2]]);
-        __m128 k3 = _mm_load_ps(filter[frac_.i[3]]);
-        __m128 out;
-
-        k0 = _mm_mul_ps(k0, val0);
-        k1 = _mm_mul_ps(k1, val1);
-        k2 = _mm_mul_ps(k2, val2);
-        k3 = _mm_mul_ps(k3, val3);
-        k0 = _mm_hadd_ps(k0, k1);
-        k2 = _mm_hadd_ps(k2, k3);
-        out = _mm_hadd_ps(k0, k2);
-
-        _mm_store_ps(&dst[i], out);
-
-        frac4 = _mm_add_epi32(frac4, increment4);
-        pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS));
-        frac4 = _mm_and_si128(frac4, fracMask4);
-
-        pos_.i[0] = _mm_extract_epi32(pos4, 0);
-        pos_.i[1] = _mm_extract_epi32(pos4, 1);
-        pos_.i[2] = _mm_extract_epi32(pos4, 2);
-        pos_.i[3] = _mm_extract_epi32(pos4, 3);
-        frac_.i[0] = _mm_extract_epi32(frac4, 0);
-        frac_.i[1] = _mm_extract_epi32(frac4, 1);
-        frac_.i[2] = _mm_extract_epi32(frac4, 2);
-        frac_.i[3] = _mm_extract_epi32(frac4, 3);
-    }
-
-    pos = pos_.i[0];
-    frac = frac_.i[0];
-
-    for(;i < numsamples;i++)
-    {
-        dst[i] = resample_fir4(src[pos], src[pos+1], src[pos+2], src[pos+3], filter[frac]);
-
-        frac += increment;
-        pos  += frac>>FRACTIONBITS;
-        frac &= FRACTIONMASK;
-    }
-    return dst;
-}