diff options
| author | Michael Sevakis <jethead71@rockbox.org> | 2007-02-21 07:06:58 +0000 |
|---|---|---|
| committer | Michael Sevakis <jethead71@rockbox.org> | 2007-02-21 07:06:58 +0000 |
| commit | 7c120fb0ced05419c825570ae08dbdced08178a4 (patch) | |
| tree | 0fe675e099b9cb62e2f4cb9b56f85b34d1c3ab2c /apps/dsp_cf.S | |
| parent | 8d7a364063c8ff47c1991a3c4e814e65e6ce5fd6 (diff) | |
SWCODEC-Coldfire: Small upsampling tweek improves speed and size a tiny bit and prevents unneeded reloading of the previous sample. I imagine it would help most with non-iram source buffers.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12425 a1c6a512-1295-4272-9138-f99709370657
Diffstat (limited to 'apps/dsp_cf.S')
| -rw-r--r-- | apps/dsp_cf.S | 40 |
1 files changed, 21 insertions, 19 deletions
diff --git a/apps/dsp_cf.S b/apps/dsp_cf.S index d832b9f0e8..b42e2b2aad 100644 --- a/apps/dsp_cf.S +++ b/apps/dsp_cf.S @@ -172,38 +172,41 @@ dsp_upsample: .uschannel_loop: move.l (%a0), %d5 | %d5 = phase = r->phase move.l -4(%a1, %d2.l*4), %a3 | %a3 = s = src[ch-1] + lea.l 4(%a0, %d2.l*4), %a4 | %a4 = &r->last_sample[ch-1] + lea.l (%a3, %d3.l*4), %a5 | %a5 = src_end = &src[count] + move.l (%a4), %d0 | %d0 = last = r->last_sample[ch-1] + move.l -4(%a5), %d1 | r->last_sample[ch-1] = s[count-1] + move.l %d1, (%a4) | move.l -4(%a2, %d2.l*4), %a4 | %a4 = d = dst[ch-1] - lea.l 4(%a0, %d2.l*4), %a5 | %a5 = &r->last_sample[ch-1] - move.l (%a5), %d0 | %d0 = last = r->last_sample[ch-1] - move.l -4(%a3, %d3.l*4), %d1 | r->last_sample[ch-1] = s[count-1] - move.l %d1, (%a5) | moveq.l #16, %d1 | %d0 = shift move.l %d5, %d6 | %d6 = pos = phase >> 16 lsl.l %d1, %d5 | swap phase to high word to use | carries to increment position lsr.l %d1, %d6 | pos == 0? - bne.b .usstart_1 | no? transistion from down - move.l (%a3), %d1 | %d1 = s[0] - sub.l %d0, %d1 | diff = s[pos] - last - bra.b .usloop_0 | jump to typical start point -.usstart_1: + beq.b .usstart_0 | no? transistion from down cmp.l %d3, %d6 | past end of samples? bge.b .usloop_skip | yes? skip loop + lea.l 4(%a3, %d6.l*4), %a3 | %a3 = s + pos + 1 + movem.l -8(%a3), %d0-%d1 | %d0 = *(s - 2), %d1 = *(s - 1) + .word 0x51fa | tpf.w - trap next instruction +.usstart_0: + move.l (%a3)+, %d1 | %d1 = *s++ + .word 0x51fb | tpf.l - trap next two instructions .usloop_1: - lea.l -4(%a3, %d6.l*4), %a5 | load s[pos-1] and s[pos] - movem.l (%a5), %d0-%d1 | + move.l %d6, %d0 | move previous sample to %d0 + move.l (%a3)+, %d1 | fetch next sample + move.l %d1, %d6 | save sample value sub.l %d0, %d1 | %d1 = diff = s[pos] - s[pos-1] .usloop_0: - move.l %d0, %acc0 | %acc0 = previous sample lsr.l #1, %d5 | make phase into frac - mac.l %d1, %d5, %acc0 | %acc0 += diff * frac - move.l %acc0, %d7 | + mac.l %d1, %d5, %acc0 | %acc0 = diff * frac + movclr.l %acc0, %d7 | %d7 = product lsl.l #1, %d5 | restore frac to phase - move.l %d7, (%a4)+ | *d++ = %d0 + add.l %d0, %d7 | %d7 = last + product + move.l %d7, (%a4)+ | *d++ = %d7 add.l %d4, %d5 | phase += delta bcc.b .usloop_0 | load next values? - addq.l #1, %d6 | increment position - cmp.l %d3, %d6 | pos < count? + cmp.l %a5, %a3 | src < src_end? blt.b .usloop_1 | yes? continue resampling .usloop_skip: subq.l #1, %d2 | ch > 0? @@ -212,9 +215,8 @@ dsp_upsample: move.l %d5, (%a0) | ...and save in r->phase move.l %a4, %d0 | return d - d[0] sub.l (%a2), %d0 | - asr.l #2, %d0 | convert bytes->samples movem.l (%sp), %d2-%d7/%a2-%a5 | restore non-clobberables - move.l %acc1, %acc0 | clear %acc0 + asr.l #2, %d0 | convert bytes->samples lea.l 40(%sp), %sp | cleanup stack rts | buh-bye .usend: |