Together with the disabling of the assert in the loop, this improves the performance of s16 2ch resampling by 4.5x. --- src/pulsecore/resampler.c | 21 +++++++++++++-------- 1 files changed, 13 insertions(+), 8 deletions(-) diff --git a/src/pulsecore/resampler.c b/src/pulsecore/resampler.c index c3b6df1..3ca0a2f 100644 --- a/src/pulsecore/resampler.c +++ b/src/pulsecore/resampler.c @@ -1369,7 +1369,7 @@ static int speex_init(pa_resampler *r) { static void trivial_resample(pa_resampler *r, const pa_memchunk *input, unsigned in_n_frames, pa_memchunk *output, unsigned *out_n_frames) { size_t fz; - unsigned o_index; + unsigned i_index, o_index; void *src, *dst; pa_assert(r); @@ -1383,18 +1383,23 @@ static void trivial_resample(pa_resampler *r, const pa_memchunk *input, unsigned dst = (uint8_t*) pa_memblock_acquire(output->memblock) + output->index; for (o_index = 0;; o_index++, r->trivial.o_counter++) { - unsigned j; + i_index = ((r->trivial.o_counter * r->i_ss.rate) / r->o_ss.rate); + i_index = i_index > r->trivial.i_counter ? i_index - r->trivial.i_counter : 0; - j = ((r->trivial.o_counter * r->i_ss.rate) / r->o_ss.rate); - j = j > r->trivial.i_counter ? j - r->trivial.i_counter : 0; - - if (j >= in_n_frames) + if (i_index >= in_n_frames) break; pa_assert_fp(o_index * fz < pa_memblock_get_length(output->memblock)); - memcpy((uint8_t*) dst + fz * o_index, - (uint8_t*) src + fz * j, (int) fz); + /* Directly assign some common sizes, use memcpy as fallback */ + if (fz == 2) + ((uint16_t *) dst)[o_index] = ((uint16_t *) src)[i_index]; + else if (fz == 4) + ((uint32_t *) dst)[o_index] = ((uint32_t *) src)[i_index]; + else if (fz == 8) + ((uint64_t *) dst)[o_index] = ((uint64_t *) src)[i_index]; + else + memcpy((uint8_t *) dst + fz * o_index, (uint8_t *) src + fz * i_index, (int) fz); } pa_memblock_release(input->memblock); -- 1.7.4.1