Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/apps/dsp_cf.S

506 lines
25 KiB
ArmAsm
Raw Blame History

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 Thom Johansen
* Portions Copyright (C) 2007 Michael Sevakis
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
/****************************************************************************
* void apply_crossfeed(int count, int32_t *src[])
*/
.section .text
.global apply_crossfeed
apply_crossfeed:
lea.l -44(%sp), %sp
movem.l %d2-%d7/%a2-%a6, (%sp) | save all regs
movem.l 48(%sp), %d7/%a4 | %d7 = count, %a4 = src
movem.l (%a4), %a4-%a5 | %a4 = src[0], %a5 = src[1]
lea.l crossfeed_data, %a1
move.l (%a1)+, %a6 | a6 = direct gain
movem.l 12(%a1), %d0-%d3 | fetch filter history samples
move.l 132(%a1), %a0 | fetch delay line address
movem.l (%a1), %a1-%a3 | load filter coefs
/* Register usage in loop:
* %a0 = delay_p, %a1..%a3 = b0, b1, a1 (filter coefs),
* %a4 = src[0], %a5 = src[1], %a6 = direct gain,
* %d0..%d3 = history
* %d4..%d6 = temp.
* %d7 = count
*/
.cfloop:
mac.l %a2, %d0, 4(%a0), %d0, %acc0 | acc = b1*dr[n - 1] d0 = dr[n]
mac.l %a1, %d0 , %acc0 | acc += b0*dr[n]
mac.l %a3, %d1, (%a4), %d4, %acc0 | acc += a1*y_l[n - 1], load L
move.l %acc0, %d1 | get filtered delayed sample
mac.l %a6, %d4, %acc0 | acc += gain*x_l[n]
movclr.l %acc0, %d6 |
move.l %d6, (%a4)+ | write result
mac.l %a2, %d2, (%a0), %d2, %acc0 | acc = b1*dl[n - 1], d2 = dl[n]
mac.l %a1, %d2 , %acc0 | acc += b0*dl[n]
mac.l %a3, %d3, (%a5), %d5, %acc0 | acc += a1*y_r[n - 1], load R
movem.l %d4-%d5, (%a0) | save left & right inputs to delay line
move.l %acc0, %d3 | get filtered delayed sample
mac.l %a6, %d5, %acc0 | acc += gain*x_r[n]
lea.l 8(%a0), %a0 | increment delay pointer
movclr.l %acc0, %d6 |
move.l %d6, (%a5)+ | write result
cmpa.l #crossfeed_data+136, %a0| wrap a0 if passed end
bge.b .cfwrap |
.word 0x51fb | tpf.l - trap the buffer wrap
.cfwrap:
lea.l -104(%a0), %a0 | wrap
subq.l #1, %d7 | --count < 0 ?
bgt.b .cfloop |
lea.l crossfeed_data+16, %a1 | save data back to struct
movem.l %d0-%d3, (%a1) | ...history
move.l %a0, 120(%a1) | ...delay_p
movem.l (%sp), %d2-%d7/%a2-%a6 | restore all regs
lea.l 44(%sp), %sp
rts
.cfend:
.size apply_crossfeed,.cfend-apply_crossfeed
/****************************************************************************
* int dsp_downsample(int count, struct dsp_data *data,
* in32_t *src[], int32_t *dst[])
*/
.section .text
.global dsp_downsample
dsp_downsample:
lea.l -40(%sp), %sp | save non-clobberables
movem.l %d2-%d7/%a2-%a5, (%sp) |
movem.l 44(%sp), %d2/%a0-%a2 | %d2 = count
| %a0 = data
| %a1 = src
| %a2 = dst
movem.l 4(%a0), %d3-%d4 | %d3 = ch = data->num_channels
| %d4 = delta = data->resample_data.delta
moveq.l #16, %d7 | %d7 = shift
.dschannel_loop:
move.l 12(%a0), %d5 | %d5 = phase = data->resample_data.phase
move.l -4(%a1, %d3.l*4), %a3 | %a3 = s = src[ch-1]
move.l -4(%a2, %d3.l*4), %a4 | %a4 = d = dst[ch-1]
lea.l 12(%a0, %d3.l*4), %a5 | %a5 = &data->resample_data.ast_sample[ch-1]
move.l (%a5), %d0 | %d0 = last = data->resample_data.last_sample[ch-1]
move.l -4(%a3, %d2.l*4), (%a5) | data->resample_data.last_sample[ch-1] = s[count-1]
move.l %d5, %d6 | %d6 = pos = phase >> 16
lsr.l %d7, %d6 |
cmp.l %d2, %d6 | past end of samples?
bge.b .dsloop_skip | yes? skip loop
tst.l %d6 | need last sample of prev. frame?
bne.b .dsloop | no? start main loop
move.l (%a3, %d6.l*4), %d1 | %d1 = s[pos]
bra.b .dsuse_last_start | start with last (last in %d0)
.dsloop:
lea.l -4(%a3, %d6.l*4), %a5 | load s[pos-1] and s[pos]
movem.l (%a5), %d0-%d1 |
.dsuse_last_start:
sub.l %d0, %d1 | %d1 = diff = s[pos] - s[pos-1]
move.l %d0, %acc0 | %acc0 = previous sample
move.l %d5, %d0 | frac = (phase << 16) >> 1
lsl.l %d7, %d0 |
lsr.l #1, %d0 |
mac.l %d0, %d1, %acc0 | %acc0 += frac * diff
add.l %d4, %d5 | phase += delta
move.l %d5, %d6 | pos = phase >> 16
lsr.l %d7, %d6 |
movclr.l %acc0, %d0 |
move.l %d0, (%a4)+ | *d++ = %d0
cmp.l %d2, %d6 | pos < count?
blt.b .dsloop | yes? continue resampling
.dsloop_skip:
subq.l #1, %d3 | ch > 0?
bgt.b .dschannel_loop | yes? process next channel
asl.l %d7, %d2 | wrap phase to start of next frame
sub.l %d2, %d5 | data->resample_data.phase =
move.l %d5, 12(%a0) | ... phase - (count << 16)
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
lea.l 40(%sp), %sp | cleanup stack
rts | buh-bye
.dsend:
.size dsp_downsample,.dsend-dsp_downsample
/****************************************************************************
* int dsp_upsample(int count, struct dsp_data *dsp,
* in32_t *src[], int32_t *dst[])
*/
.section .text
.global dsp_upsample
dsp_upsample:
lea.l -40(%sp), %sp | save non-clobberables
movem.l %d2-%d7/%a2-%a5, (%sp) |
movem.l 44(%sp), %d2/%a0-%a2 | %d2 = count
| %a0 = data
| %a1 = src
| %a2 = dst
movem.l 4(%a0), %d3-%d4 | %d3 = ch = channels
| %d4 = delta = data->resample_data.delta
swap %d4 | swap delta to high word to use
| carries to increment position
.uschannel_loop:
move.l 12(%a0), %d5 | %d5 = phase = data->resample_data.phase
move.l -4(%a1, %d3.l*4), %a3 | %a3 = s = src[ch-1]
lea.l 12(%a0, %d3.l*4), %a4 | %a4 = &data->resample_data.last_sample[ch-1]
lea.l (%a3, %d2.l*4), %a5 | %a5 = src_end = &src[count]
move.l (%a4), %d0 | %d0 = last = data->resample_data.last_sample[ch-1]
move.l -(%a5), (%a4) | data->resample_data.last_sample[ch-1] = s[count-1]
move.l -4(%a2, %d3.l*4), %a4 | %a4 = d = dst[ch-1]
swap %d5 | swap phase to high word to use
| carries to increment position
move.l %d5, %d6 | %d6 = pos = phase >> 16
clr.w %d5 |
eor.l %d5, %d6 | pos == 0?
beq.b .usstart_0 | no? transistion from down
cmp.l %d2, %d6 | past end of samples?
bge.b .usloop_skip | yes? skip loop
lea.l -4(%a3, %d6.l*4), %a3 | %a3 = s = &s[pos-1] (previous)
move.l (%a3)+, %d0 | %d0 = *s++
.word 0x51fa | tpf.w - trap next instruction
.usloop_1:
move.l %d6, %d0 | move previous sample to %d0
.usstart_0:
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:
lsr.l #1, %d5 | make phase into frac
mac.l %d1, %d5, %acc0 | %acc0 = diff * frac
lsl.l #1, %d5 | restore frac to phase
movclr.l %acc0, %d7 | %d7 = product
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?
cmp.l %a5, %a3 | src <= src_end?
ble.b .usloop_1 | yes? continue resampling
.usloop_skip:
subq.l #1, %d3 | ch > 0?
bgt.b .uschannel_loop | yes? process next channel
swap %d5 | wrap phase to start of next frame
move.l %d5, 12(%a0) | ...and save in data->resample_data.phase
move.l %a4, %d0 | return d - d[0]
sub.l (%a2), %d0 |
movem.l (%sp), %d2-%d7/%a2-%a5 | restore non-clobberables
asr.l #2, %d0 | convert bytes->samples
lea.l 40(%sp), %sp | cleanup stack
rts | buh-bye
.usend:
.size dsp_upsample,.usend-dsp_upsample
/* These routines might benefit from burst transfers but we'll keep them
* small for now since they're rather light weight
*/
/****************************************************************************
* void channels_process_sound_chan_mono(int count, int32_t *buf[])
*
* Mix left and right channels 50/50 into a center channel.
*/
.section .text
.global channels_process_sound_chan_mono
channels_process_sound_chan_mono:
movem.l 4(%sp), %d0/%a0 | %d0 = count, %a0 = buf
lea.l -12(%sp), %sp | save registers
move.l %macsr, %d1 |
movem.l %d1-%d3, (%sp) |
move.l #0xb0, %macsr | put emac in rounding fractional mode
movem.l (%a0), %a0-%a1 | get channel pointers
move.l #0x40000000, %d3 | %d3 = 0.5
1:
move.l (%a0), %d1 | L = R = l/2 + r/2
mac.l %d1, %d3, (%a1), %d2, %acc0 |
mac.l %d2, %d3, %acc0 |
movclr.l %acc0, %d1 |
move.l %d1, (%a0)+ | output to original buffer
move.l %d1, (%a1)+ |
subq.l #1, %d0 |
bgt.s 1b |
movem.l (%sp), %d1-%d3 | restore registers
move.l %d1, %macsr |
lea.l 12(%sp), %sp | cleanup
rts
.cpmono_end:
.size channels_process_sound_chan_mono, .cpmono_end-channels_process_sound_chan_mono
/****************************************************************************
* void channels_process_sound_chan_custom(int count, int32_t *buf[])
*
* Apply stereo width (narrowing/expanding) effect.
*/
.section .text
.global channels_process_sound_chan_custom
channels_process_sound_chan_custom:
movem.l 4(%sp), %d0/%a0 | %d0 = count, %a0 = buf
lea.l -16(%sp), %sp | save registers
move.l %macsr, %d1 |
movem.l %d1-%d4, (%sp) |
move.l #0xb0, %macsr | put emac in rounding fractional mode
movem.l (%a0), %a0-%a1 | get channel pointers
move.l dsp_sw_gain, %d3 | load straight (mid) gain
move.l dsp_sw_cross, %d4 | load cross (side) gain
1:
move.l (%a0), %d1 |
mac.l %d1, %d3, (%a1), %d2, %acc0 | L = l*gain + r*cross
mac.l %d1, %d4 , %acc1 | R = r*gain + l*cross
mac.l %d2, %d4 , %acc0 |
mac.l %d2, %d3 , %acc1 |
movclr.l %acc0, %d1 |
movclr.l %acc1, %d2 |
move.l %d1, (%a0)+ |
move.l %d2, (%a1)+ |
subq.l #1, %d0 |
bgt.s 1b |
movem.l (%sp), %d1-%d4 | restore registers
move.l %d1, %macsr |
lea.l 16(%sp), %sp | cleanup
rts
.cpcustom_end:
.size channels_process_sound_chan_custom, .cpcustom_end-channels_process_sound_chan_custom
/****************************************************************************
* void channels_process_sound_chan_karaoke(int count, int32_t *buf[])
*
* Separate channels into side channels.
*/
.section .text
.global channels_process_sound_chan_karaoke
channels_process_sound_chan_karaoke:
movem.l 4(%sp), %d0/%a0 | %d0 = count, %a0 = buf
lea.l -16(%sp), %sp | save registers
move.l %macsr, %d1 |
movem.l %d1-%d4, (%sp) |
move.l #0xb0, %macsr | put emac in rounding fractional mode
movem.l (%a0), %a0-%a1 | get channel pointers
move.l #0x40000000, %d4 | %d3 = 0.5
1:
move.l (%a0), %d1 |
msac.l %d1, %d4, (%a1), %d2, %acc0 | R = r/2 - l/2
mac.l %d2, %d4 , %acc0 |
movclr.l %acc0, %d1 |
move.l %d1, (%a1)+ |
neg.l %d1 | L = -R = -(r/2 - l/2) = l/2 - r/2
move.l %d1, (%a0)+ |
subq.l #1, %d0 |
bgt.s 1b |
movem.l (%sp), %d1-%d4 | restore registers
move.l %d1, %macsr |
lea.l 16(%sp), %sp | cleanup
rts
.cpkaraoke_end:
.size channels_process_sound_chan_karaoke, .cpkaraoke_end-channels_process_sound_chan_karaoke
/****************************************************************************
* void sample_output_stereo(int count, struct dsp_data *data,
* int32_t *src[], int16_t *dst)
*
* Framework based on the ubiquitous Rockbox line transfer logic for
* Coldfire CPUs.
*
* Does emac clamping and scaling (which proved faster than the usual
* checks and branches - even single test clamping) and writes using
* line burst transfers. Also better than writing a single L-R pair per
* loop but a good deal more code.
*
* Attemping bursting during reads is rather futile since the source and
* destination alignments rarely agree and too much complication will
* slow us up. The parallel loads seem to do a bit better at least until
* a pcm buffer can always give line aligned chunk and then aligning the
* dest can then imply the source is aligned if the source buffers are.
* For now longword alignment is assumed of both the source and dest.
*
*/
.section .text
.global sample_output_stereo
sample_output_stereo:
lea.l -44(%sp), %sp | save registers
move.l %macsr, %d1 | do it now as at many lines will
movem.l %d1-%d7/%a2-%a5, (%sp) | be the far more common condition
move.l #0x80, %macsr | put emac unit in signed int mode
movem.l 48(%sp), %a0-%a2/%a4 |
lea.l (%a4, %a0.l*4), %a0 | %a0 = end address
move.l (%a1), %d1 | %a1 = multiplier: (1 << (16 - scale))
sub.l #16, %d1 |
neg.l %d1 |
moveq.l #1, %d0 |
asl.l %d1, %d0 |
move.l %d0, %a1 |
movem.l (%a2), %a2-%a3 | get L/R channel pointers
moveq.l #28, %d0 | %d0 = second line bound
add.l %a4, %d0 |
and.l #0xfffffff0, %d0 |
cmp.l %a0, %d0 | at least a full line?
bhi.w .sos_longloop_1_start | no? jump to trailing longword
sub.l #16, %d0 | %d1 = first line bound
cmp.l %a4, %d0 | any leading longwords?
bls.b .sos_lineloop_start | no? jump to line loop
.sos_longloop_0:
move.l (%a2)+, %d1 | read longword from L and R
mac.l %d1, %a1, (%a3)+, %d2, %acc0 | shift L to high word
mac.l %d2, %a1, %acc1 | shift R to high word
movclr.l %acc0, %d1 | get possibly saturated results
movclr.l %acc1, %d2 |
swap %d2 | move R to low word
move.w %d2, %d1 | interleave MS 16 bits of each
move.l %d1, (%a4)+ | ...and write both
cmp.l %a4, %d0 |
bhi.b .sos_longloop_0 |
.sos_lineloop_start:
lea.l -12(%a0), %a5 | %a5 = at or just before last line bound
.sos_lineloop:
move.l (%a3)+, %d4 | get next 4 R samples and scale
mac.l %d4, %a1, (%a3)+, %d5, %acc0 | with saturation
mac.l %d5, %a1, (%a3)+, %d6, %acc1 |
mac.l %d6, %a1, (%a3)+, %d7, %acc2 |
mac.l %d7, %a1, (%a2)+, %d0, %acc3 |
lea.l 16(%a4), %a4 | increment dest here, mitigate stalls
movclr.l %acc0, %d4 | obtain R results
movclr.l %acc1, %d5 |
movclr.l %acc2, %d6 |
movclr.l %acc3, %d7 |
mac.l %d0, %a1, (%a2)+, %d1, %acc0 | get next 4 L samples and scale
mac.l %d1, %a1, (%a2)+, %d2, %acc1 | with saturation
mac.l %d2, %a1, (%a2)+, %d3, %acc2 |
mac.l %d3, %a1 , %acc3 |
swap %d4 | a) interleave most significant...
swap %d5 |
swap %d6 |
swap %d7 |
movclr.l %acc0, %d0 | obtain L results
movclr.l %acc1, %d1 |
movclr.l %acc2, %d2 |
movclr.l %acc3, %d3 |
move.w %d4, %d0 | a) ... 16 bits of L and R
move.w %d5, %d1 |
move.w %d6, %d2 |
move.w %d7, %d3 |
movem.l %d0-%d3, -16(%a4) | write four stereo samples
cmp.l %a4, %a5 |
bhi.b .sos_lineloop |
.sos_longloop_1_start:
cmp.l %a4, %a0 | any longwords left?
bls.b .sos_done | no? finished.
.sos_longloop_1:
move.l (%a2)+, %d1 | handle trailing longwords
mac.l %d1, %a1, (%a3)+, %d2, %acc0 | the same way as leading ones
mac.l %d2, %a1, %acc1 |
movclr.l %acc0, %d1 |
movclr.l %acc1, %d2 |
swap %d2 |
move.w %d2, %d1 |
move.l %d1, (%a4)+ |
cmp.l %a4, %a0 |
bhi.b .sos_longloop_1 |
.sos_done:
movem.l (%sp), %d1-%d7/%a2-%a5 | restore registers
move.l %d1, %macsr |
lea.l 44(%sp), %sp | cleanup
rts |
.sos_end:
.size sample_output_stereo, .sos_end-sample_output_stereo
/****************************************************************************
* void sample_output_mono(int count, struct dsp_data *data,
* int32_t *src[], int16_t *dst)
*
* Same treatment as sample_output_stereo but for one channel.
*/
.section .text
.global sample_output_mono
sample_output_mono:
lea.l -28(%sp), %sp | save registers
move.l %macsr, %d1 | do it now as at many lines will
movem.l %d1-%d5/%a2-%a3, (%sp) | be the far more common condition
move.l #0x80, %macsr | put emac unit in signed int mode
movem.l 32(%sp), %a0-%a3 |
lea.l (%a3, %a0.l*4), %a0 | %a0 = end address
move.l (%a1), %d1 | %d5 = multiplier: (1 << (16 - scale))
sub.l #16, %d1 |
neg.l %d1 |
moveq.l #1, %d5 |
asl.l %d1, %d5 |
movem.l (%a2), %a2 | get source channel pointer
moveq.l #28, %d0 | %d0 = second line bound
add.l %a3, %d0 |
and.l #0xfffffff0, %d0 |
cmp.l %a0, %d0 | at least a full line?
bhi.w .som_longloop_1_start | no? jump to trailing longword
sub.l #16, %d0 | %d1 = first line bound
cmp.l %a3, %d0 | any leading longwords?
bls.b .som_lineloop_start | no? jump to line loop
.som_longloop_0:
move.l (%a2)+, %d1 | read longword from L and R
mac.l %d1, %d5, %acc0 | shift L to high word
movclr.l %acc0, %d1 | get possibly saturated results
move.l %d1, %d2 |
swap %d2 | move R to low word
move.w %d2, %d1 | duplicate single channel into
move.l %d1, (%a3)+ | L and R
cmp.l %a3, %d0 |
bhi.b .som_longloop_0 |
.som_lineloop_start:
lea.l -12(%a0), %a1 | %a1 = at or just before last line bound
.som_lineloop:
move.l (%a2)+, %d0 | get next 4 L samples and scale
mac.l %d0, %d5, (%a2)+, %d1, %acc0 | with saturation
mac.l %d1, %d5, (%a2)+, %d2, %acc1 |
mac.l %d2, %d5, (%a2)+, %d3, %acc2 |
mac.l %d3, %d5 , %acc3 |
lea.l 16(%a3), %a3 | increment dest here, mitigate stalls
movclr.l %acc0, %d0 | obtain results
movclr.l %acc1, %d1 |
movclr.l %acc2, %d2 |
movclr.l %acc3, %d3 |
move.l %d0, %d4 | duplicate single channel
swap %d4 | into L and R
move.w %d4, %d0 |
move.l %d1, %d4 |
swap %d4 |
move.w %d4, %d1 |
move.l %d2, %d4 |
swap %d4 |
move.w %d4, %d2 |
move.l %d3, %d4 |
swap %d4 |
move.w %d4, %d3 |
movem.l %d0-%d3, -16(%a3) | write four stereo samples
cmp.l %a3, %a1 |
bhi.b .som_lineloop |
.som_longloop_1_start:
cmp.l %a3, %a0 | any longwords left?
bls.b .som_done | no? finished.
.som_longloop_1:
move.l (%a2)+, %d1 | handle trailing longwords
mac.l %d1, %d5, %acc0 | the same way as leading ones
movclr.l %acc0, %d1 |
move.l %d1, %d2 |
swap %d2 |
move.w %d2, %d1 |
move.l %d1, (%a3)+ |
cmp.l %a3, %a0 |
bhi.b .som_longloop_1 |
.som_done:
movem.l (%sp), %d1-%d5/%a2-%a3 | restore registers
move.l %d1, %macsr |
lea.l 28(%sp), %sp | cleanup
rts |
.som_end:
.size sample_output_mono, .som_end-sample_output_mono
Reference in a new issue View git blame Copy permalink