Skip to content

Commit

Permalink
Merge branch 'mamedev:master' into master
Browse files Browse the repository at this point in the history
  • Loading branch information
@yeager
yeager authored Jan 6, 2025
2 parents 2c16226 + 71fe25f commit bdaf07d
Show file tree
Hide file tree
Showing 3 changed files with 104 additions and 103 deletions.
2 changes: 1 addition & 1 deletion src/devices/cpu/tms57002/tms57002.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -925,7 +925,7 @@ void tms57002_device::sound_stream_update(sound_stream &stream, std::vector<read
assert(inputs[0].samples() == 1);
assert(outputs[0].samples() == 1);

stream_buffer::sample_t in_scale = 32768.0 * ((st0 & ST0_SIM) ? 256.0 : 1.0);
stream_buffer::sample_t in_scale = 32767.0 * ((st0 & ST0_SIM) ? 256.0 : 1.0);
si[0] = s32(inputs[0].get(0) * in_scale) & 0xffffff;
si[1] = s32(inputs[1].get(0) * in_scale) & 0xffffff;
si[2] = s32(inputs[2].get(0) * in_scale) & 0xffffff;
Expand Down
200 changes: 102 additions & 98 deletions src/devices/sound/zsg2.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,91 +1,93 @@
// license:BSD-3-Clause
// copyright-holders:Olivier Galibert, R. Belmont, hap, superctr
/*
ZOOM ZSG-2 custom wavetable synthesizer
Written by Olivier Galibert
MAME conversion by R. Belmont
Working emulation by The Talentuous Hands Of The Popularious hap
Properly working emulation by superctr
---------------------------------------------------------
Register map:
000-5fe : Channel specific registers (48 channels)
(high) (low)
+000 : xxxxxxxx -------- : Start address (low)
+000 : -------- xxxxxxxx : Unknown register (usually cleared)
+002 : xxxxxxxx -------- : Address page
: -------- xxxxxxxx : Start address (high)
+004 : -------- -------- : Unknown register (usually cleared)
+006 : -----x-- -------- : Unknown bit, always set
+008 : xxxxxxxx xxxxxxxx : Frequency
+00a : xxxxxxxx -------- : DSP ch 3 (right) output gain
: -------- xxxxxxxx : Loop address (low)
+00c : xxxxxxxx xxxxxxxx : End address
+00e : xxxxxxxx -------- : DSP ch 2 (Left) output gain
: -------- xxxxxxxx : Loop address (high)
+010 : xxxxxxxx xxxxxxxx : Initial filter time constant
+012 : xxxxxxxx xxxxxxxx : Current filter time constant
+014 : xxxxxxxx xxxxxxxx : Initial volume
+016 : xxxxxxxx xxxxxxxx : Current volume?
+018 : xxxxxxxx xxxxxxxx : Target filter time constant
+01a : xxxxxxxx -------- : DSP ch 1 (chorus) output gain
: -------- xxxxxxxx : Filter ramping speed
+01c : xxxxxxxx xxxxxxxx : Target volume
+01e : xxxxxxxx -------- : DSP ch 0 (reverb) output gain
: -------- xxxxxxxx : Filter ramping speed
600-604 : Key on flags (each bit corresponds to a channel)
608-60c : Key off flags (each bit corresponds to a channel)
618 : Unknown register (usually 0x5cbc is written)
61a : Unknown register (usually 0x5cbc is written)
620 : Unknown register (usually 0x0128 is written)
628 : Unknown register (usually 0x0066 is written)
630 : Unknown register (usually 0x0001 is written)
638 : ROM readback address low
63a : ROM readback address high
63c : ROM readback word low
63e : ROM readback word high
---------------------------------------------------------
Additional notes on the sample format, reverse-engineered
by Olivier Galibert and David Haywood:
The zoom sample rom is decomposed in 0x40000 bytes pages. Each page
starts by a header and is followed by compressed samples.
The header is a vector of 16 bytes structures composed of 4 32bits
little-endian values representing:
- sample start position in bytes, always a multiple of 4
- sample end position in bytes, minus 4, always...
- loop position in bytes, always....
- flags, probably
It is interesting to note that this header is *not* parsed by the
ZSG. The main program reads the rom through appropriate ZSG
commands, and use the results in subsequent register setups. It's
not even obvious that the ZSG cares about the pages, it may just
see the address space as linear. In the same line, the
interpretation of the flags is obviously dependent on the main
program, not the ZSG, but some of the bits are directly copied to
some of the registers.
The samples are compressed with a 2:1 ratio. Each block of 4-bytes
becomes 4 16-bits samples. Reading the 4 bytes as a *little-endian*
32bits values, the structure is:
42222222 51111111 60000000 ssss3333
's' is a 4-bit scale value. '0000000', '1111111', '2222222' and
'6543333' are signed 7-bits values corresponding to the 4 samples.
To compute the final 16bits value, left-align and shift right by s.
Yes, that simple.
---------------------------------------------------------
ZOOM ZSG-2 custom wavetable synthesizer
Written by Olivier Galibert
MAME conversion by R. Belmont
Working emulation by The Talentuous Hands Of The Popularious hap
Properly working emulation by superctr
---------------------------------------------------------
Register map:
000-5fe : Channel specific registers (48 channels)
(high) (low)
+000 : xxxxxxxx -------- : Start address (low)
+000 : -------- xxxxxxxx : Unknown register (usually cleared)
+002 : xxxxxxxx -------- : Address page
: -------- xxxxxxxx : Start address (high)
+004 : -------- -------- : Unknown register (usually cleared)
+006 : -----x-- -------- : Unknown bit, always set
+008 : xxxxxxxx xxxxxxxx : Frequency
+00a : xxxxxxxx -------- : DSP ch 3 (right) output gain
: -------- xxxxxxxx : Loop address (low)
+00c : xxxxxxxx xxxxxxxx : End address
+00e : xxxxxxxx -------- : DSP ch 2 (Left) output gain
: -------- xxxxxxxx : Loop address (high)
+010 : xxxxxxxx xxxxxxxx : Initial filter time constant
+012 : xxxxxxxx xxxxxxxx : Current filter time constant
+014 : xxxxxxxx xxxxxxxx : Initial volume
+016 : xxxxxxxx xxxxxxxx : Current volume?
+018 : xxxxxxxx xxxxxxxx : Target filter time constant
+01a : xxxxxxxx -------- : DSP ch 1 (chorus) output gain
: -------- xxxxxxxx : Filter ramping speed
+01c : xxxxxxxx xxxxxxxx : Target volume
+01e : xxxxxxxx -------- : DSP ch 0 (reverb) output gain
: -------- xxxxxxxx : Filter ramping speed
600-604 : Key on flags (each bit corresponds to a channel)
608-60c : Key off flags (each bit corresponds to a channel)
618 : Unknown register (usually 0x5cbc is written)
61a : Unknown register (usually 0x5cbc is written)
620 : Unknown register (usually 0x0128 is written)
628 : Unknown register (usually 0x0066 is written)
630 : Unknown register (usually 0x0001 is written)
638 : ROM readback address low
63a : ROM readback address high
63c : ROM readback word low
63e : ROM readback word high
---------------------------------------------------------
Additional notes on the sample format, reverse-engineered
by Olivier Galibert and David Haywood:
The zoom sample rom is decomposed in 0x40000 bytes pages. Each page
starts by a header and is followed by compressed samples.
The header is a vector of 16 bytes structures composed of 4 32bits
little-endian values representing:
- sample start position in bytes, always a multiple of 4
- sample end position in bytes, minus 4, always...
- loop position in bytes, always....
- flags, probably
It is interesting to note that this header is *not* parsed by the
ZSG. The main program reads the rom through appropriate ZSG
commands, and use the results in subsequent register setups. It's
not even obvious that the ZSG cares about the pages, it may just
see the address space as linear. In the same line, the
interpretation of the flags is obviously dependent on the main
program, not the ZSG, but some of the bits are directly copied to
some of the registers.
The samples are compressed with a 2:1 ratio. Each block of 4-bytes
becomes 4 16-bits samples. Reading the 4 bytes as a *little-endian*
32bits values, the structure is:
42222222 51111111 60000000 ssss3333
's' is a 4-bit scale value. '0000000', '1111111', '2222222' and
'6543333' are signed 7-bits values corresponding to the 4 samples.
To compute the final 16bits value, left-align and shift right by s.
Yes, that simple.
---------------------------------------------------------
TODO:
- Filter and ramping behavior might not be perfect.
- clicking / popping noises in gdarius, raystorm: maybe the sample ROMs are bad dumps?
- clicking / popping noises in gdarius, raystorm: maybe the sample ROMs
are bad dumps?
- memory reads out of range sometimes
*/
Expand Down Expand Up @@ -290,7 +292,7 @@ void zsg2_device::sound_stream_update(sound_stream &stream, std::vector<read_str
// loop over all channels
for (auto & elem : m_chan)
{
if(~elem.status & STATUS_ACTIVE)
if (~elem.status & STATUS_ACTIVE)
continue;

elem.step_ptr += elem.step;
Expand All @@ -309,7 +311,7 @@ void zsg2_device::sound_stream_update(sound_stream &stream, std::vector<read_str
}
}

if(elem.cur_pos == elem.start_pos)
if (elem.cur_pos == elem.start_pos)
elem.emphasis_filter_state = EMPHASIS_INITIAL_BIAS;

elem.step_ptr &= 0xffff;
Expand All @@ -320,39 +322,39 @@ void zsg2_device::sound_stream_update(sound_stream &stream, std::vector<read_str
int32_t sample = elem.samples[sample_pos];

// linear interpolation (hardware certainly does something similar)
sample += ((uint16_t)(elem.step_ptr<<2&0xffff) * (int16_t)(elem.samples[sample_pos+1] - sample))>>16;
sample += ((uint16_t)(elem.step_ptr << 2 & 0xffff) * (int16_t)(elem.samples[sample_pos+1] - sample)) >> 16;

// another filter...
elem.output_filter_state += (sample - (elem.output_filter_state>>16)) * elem.output_cutoff;
elem.output_filter_state += (sample - (elem.output_filter_state >> 16)) * elem.output_cutoff;
sample = elem.output_filter_state >> 16;

// To prevent DC bias, we need to slowly discharge the filter when the output filter cutoff is 0
if(!elem.output_cutoff)
if (!elem.output_cutoff)
elem.output_filter_state >>= 1;

sample = (sample * elem.vol)>>16;
sample = (sample * elem.vol) >> 16;

for(int output=0; output<4; output++)
for (int output = 0; output < 4; output++)
{
int output_gain = elem.output_gain[output] & 0x1f; // left / right
int32_t output_sample = sample;

if (elem.output_gain[output] & 0x80) // perhaps ?
output_sample = -output_sample;

mix[output] += (output_sample * m_gain_tab[output_gain&0x1f]) >> 16;
mix[output] += (output_sample * m_gain_tab[output_gain & 0x1f]) >> 16;
}

// Apply ramping every other update
// It's possible key on is handled on the other sample
if(m_sample_count & 1)
if (m_sample_count & 1)
{
elem.vol = ramp(elem.vol, elem.vol_target, elem.vol_delta);
elem.output_cutoff = ramp(elem.output_cutoff, elem.output_cutoff_target, elem.output_cutoff_delta);
}
}

for(int output=0; output<4; output++)
for (int output = 0; output < 4; output++)
outputs[output].put_int_clamp(i, mix[output], 32768);
}
m_sample_count++;
Expand Down Expand Up @@ -491,17 +493,17 @@ uint16_t zsg2_device::chan_r(int ch, int reg)
// calculate this value, for now I'm generating an opproximate inverse.
int16_t zsg2_device::get_ramp(uint8_t val)
{
int16_t frac = val<<12; // sign extend
frac = ((frac>>12) ^ 8) << (val >> 4);
int16_t frac = val << 12; // sign extend
frac = ((frac >> 12) ^ 8) << (val >> 4);
return (frac >> 4);
}

inline uint16_t zsg2_device::ramp(uint16_t current, uint16_t target, int16_t delta)
{
int32_t rampval = current + delta;
if(delta < 0 && rampval < target)
if (delta < 0 && rampval < target)
rampval = target;
else if(delta >= 0 && rampval > target)
else if (delta >= 0 && rampval > target)
rampval = target;

return rampval;
Expand Down Expand Up @@ -572,7 +574,7 @@ void zsg2_device::control_w(int reg, uint16_t data)
break;

default:
if(reg < 0x20)
if (reg < 0x20)
m_reg[reg] = data;
logerror("ZSG2 control %02X = %04X\n", reg, data & 0xffff);
break;
Expand All @@ -596,7 +598,7 @@ uint16_t zsg2_device::control_r(int reg)
return read_memory(m_read_address) >> 16;

default:
if(reg < 0x20)
if (reg < 0x20)
return m_reg[reg];
break;
}
Expand Down Expand Up @@ -639,6 +641,8 @@ uint16_t zsg2_device::read(offs_t offset, uint16_t mem_mask)
return 0;
}

m_stream->update();

if (offset < 0x300)
{
int chan = offset >> 4;
Expand Down
5 changes: 1 addition & 4 deletions src/devices/sound/zsg2.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,12 +58,9 @@ class zsg2_device : public device_t,
int16_t output_cutoff_delta;

int32_t emphasis_filter_state;

int32_t output_filter_state;

// Attenuation for output channels
uint8_t output_gain[4];

uint8_t output_gain[4]; // Attenuation for output channels
int16_t samples[5]; // +1 history
};

Expand Down

0 comments on commit bdaf07d

Please sign in to comment.