// license:BSD-3-Clause // copyright-holders:Nicola Salmoria /*************************************************************************** sn76496.c by Nicola Salmoria with contributions by others Routines to emulate the: Texas Instruments SN76489, SN76489A, SN76494/SN76496 ( Also known as, or at least compatible with, the TMS9919 and SN94624.) and the Sega 'PSG' used on the Master System, Game Gear, and Megadrive/Genesis This chip is known as the Programmable Sound Generator, or PSG, and is a 4 channel sound generator, with three squarewave channels and a noise/arbitrary duty cycle channel. Noise emulation for all verified chips should be accurate: ** SN76489 uses a 15-bit shift register with taps on bits D and E, output on E, XOR function. It uses a 15-bit ring buffer for periodic noise/arbitrary duty cycle. Its output is inverted. ** SN94624 is the same as SN76489 but lacks the /8 divider on its clock input. ** SN76489A uses a 15-bit shift register with taps on bits D and E, output on F, XOR function. It uses a 15-bit ring buffer for periodic noise/arbitrary duty cycle. Its output is not inverted. ** SN76494 is the same as SN76489A but lacks the /8 divider on its clock input. ** SN76496 is identical in operation to the SN76489A, but the audio input on pin 9 is documented. All the TI-made PSG chips have an audio input line which is mixed with the 4 channels of output. (It is undocumented and may not function properly on the sn76489, 76489a and 76494; the sn76489a input is mentioned in datasheets for the tms5200) All the TI-made PSG chips act as if the frequency was set to 0x400 if 0 is written to the frequency register. ** Sega Master System III/MD/Genesis PSG uses a 16-bit shift register with taps on bits C and F, output on F It uses a 16-bit ring buffer for periodic noise/arbitrary duty cycle. (whether it uses an XOR or XNOR needs to be verified, assumed XOR) (whether output is inverted or not needs to be verified, assumed to be inverted) ** Sega Game Gear PSG is identical to the SMS3/MD/Genesis one except it has an extra register for mapping which channels go to which speaker. The register, connected to a z80 port, means: for bits 7 6 5 4 3 2 1 0 L3 L2 L1 L0 R3 R2 R1 R0 Noise is an XOR function, and audio output is negated before being output. All the Sega-made PSG chips act as if the frequency was set to 0 if 0 is written to the frequency register. ** NCR8496 (as used on the Tandy 1000TX) is similar to the SN76489 but with a different noise LFSR pattern: taps on bits A and E, output on E, XNOR function It uses a 15-bit ring buffer for periodic noise/arbitrary duty cycle. Its output is inverted. ** PSSJ-3 (as used on the later Tandy 1000 series computers) is the same as the NCR8496 with the exception that its output is not inverted. 28/03/2005 : Sebastien Chevalier Update th SN76496Write func, according to SN76489 doc found on SMSPower. - On write with 0x80 set to 0, when LastRegister is other then TONE, the function is similar than update with 0x80 set to 1 23/04/2007 : Lord Nightmare Major update, implement all three different noise generation algorithms and a set_variant call to discern among them. 28/04/2009 : Lord Nightmare Add READY line readback; cleaned up struct a bit. Cleaned up comments. Add more TODOs. Fixed some unsaved savestate related stuff. 04/11/2009 : Lord Nightmare Changed the way that the invert works (it now selects between XOR and XNOR for the taps), and added R->OldNoise to simulate the extra 0 that is always output before the noise LFSR contents are after an LFSR reset. This fixes SN76489/A to match chips. Added SN94624. 14/11/2009 : Lord Nightmare Removed STEP mess, vastly simplifying the code. Made output bipolar rather than always above the 0 line, but disabled that code due to pending issues. 16/11/2009 : Lord Nightmare Fix screeching in regulus: When summing together four equal channels, the size of the max amplitude per channel should be 1/4 of the max range, not 1/3. Added NCR8496. 18/11/2009 : Lord Nightmare Modify Init functions to support negating the audio output. The gamegear psg does this. Change gamegear and sega psgs to use XOR rather than XNOR based on testing. Got rid of R->OldNoise and fixed taps accordingly. Added stereo support for game gear. 15/01/2010 : Lord Nightmare Fix an issue with SN76489 and SN76489A having the wrong periodic noise periods. Note that properly emulating the noise cycle bit timing accurately may require extensive rewriting. 24/01/2010: Lord Nightmare Implement periodic noise as forcing one of the XNOR or XOR taps to 1 or 0 respectively. Thanks to PlgDavid for providing samples which helped immensely here. Added true clock divider emulation, so sn94624 and sn76494 run 8x faster than the others, as in real life. 15/02/2010: Lord Nightmare & Michael Zapf (additional testing by PlgDavid) Fix noise period when set to mirror channel 3 and channel 3 period is set to 0 (tested on hardware for noise, wave needs tests) - MZ Fix phase of noise on sn94624 and sn76489; all chips use a standard XOR, the only inversion is the output itself - LN, Plgdavid Thanks to PlgDavid and Michael Zapf for providing samples which helped immensely here. 23/02/2011: Lord Nightmare & Enik Made it so the Sega PSG chips have a frequency of 0 if 0 is written to the frequency register, while the others have 0x400 as before. Should fix a bug or two on sega games, particularly Vigilante on Sega Master System. Verified on SMS hardware. 27/06/2012: Michael Zapf Converted to modern device, legacy devices were gradually removed afterwards. 16/09/2015: Lord Nightmare Fix PSG chips to have volume reg inited on reset to 0x0 based on tests by ValleyBell. Made Sega PSG chips start up with register 0x3 selected (volume for channel 2) based on hardware tests by Nemesis. 03/09/2018: Lord Nightmare, Qbix, ValleyBell, NewRisingSun * renamed the NCR8496 to its correct name, based on chip pictures on VGMPF * fixed NCR8496's noise LFSR behavior so it is only reset if the mode bit in register 6 is changed. * NCR8496's LFSR feedback function is an XNOR, which is now supported. * add PSSJ-3 support for the later Tandy 1000 series computers. * NCR8496's output is inverted, PSSJ-3's output is not. 10/12/2019: Michael Zapf * READY line handling by own emu_timer, not depending on sound_stream_update additional modifications by tildearrow for furnace TODO: * Implement the TMS9919 - any difference to sn94624? * Implement the T6W28; has registers in a weird order, needs writes to be 'sanitized' first. Also is stereo, similar to game gear. * Factor out common code so that the SAA1099 can share some code. * verify NCR8496/PSSJ-3 behavior on write to mirrored registers; unlike the other variants, the NCR-derived variants are implied to ignore writes to regs 1,3,5,6,7 if 0x80 is not set. This needs to be verified on real hardware. ***************************************************************************/ #include "sn76496.h" #include #define MAX_OUTPUT 0x7fff #define NOISE_START 0x8000 //#define NOISE_START 0x0f35 sn76496_base_device::sn76496_base_device( int feedbackmask, int noisetap1, int noisetap2, bool negate, int clockdivider, bool ncr, bool sega, uint32_t clock) : m_feedback_mask(feedbackmask) , m_whitenoise_tap1(noisetap1) , m_whitenoise_tap2(noisetap2) , m_negate(negate) , m_clock_divider(clockdivider) , m_ncr_style_psg(ncr) , m_sega_style_psg(sega) { } sn76496_device::sn76496_device(uint32_t clock) : sn76496_base_device(0x8000, 0x01, 0x08, false, 1, false, false, clock) { } void sn76496_base_device::device_start() { int i; double out; int gain; for (i = 0; i < 4; i++) m_volume[i] = 0; m_last_register = m_sega_style_psg?3:0; // Sega VDP PSG defaults to selected period reg for 2nd channel for (i = 0; i < 8; i+=2) { m_register[i] = 0; m_register[i + 1] = 0x0; // volume = 0x0 (max volume) on reset; this needs testing on chips other than SN76489A and Sega VDP PSG } for (i = 0; i < 4; i++) { m_output[i] = 0; m_period[i] = 0; m_count[i] = 0; } m_RNG = NOISE_START; m_output[3] = m_RNG & 1; m_current_clock = m_clock_divider-1; // set gain gain = 0; gain &= 0xff; // increase max output basing on gain (0.2 dB per step) out = MAX_OUTPUT / 4; // four channels, each gets 1/4 of the total range while (gain-- > 0) out *= 1.023292992; // = (10 ^ (0.2/20)) // build volume table (2dB per step) for (i = 0; i < 15; i++) { // limit volume to avoid clipping if (out > MAX_OUTPUT / 4) m_vol_table[i] = MAX_OUTPUT / 4; else m_vol_table[i] = out; out /= 1.258925412; /* = 10 ^ (2/20) = 2dB */ } m_vol_table[15] = 0; m_ready_state = true; } void sn76496_base_device::write(u8 data) { int n, r, c; if (data & 0x80) { r = (data & 0x70) >> 4; m_last_register = r; if (((m_ncr_style_psg) && (r == 6)) && ((data&0x04) != (m_register[6]&0x04))) m_RNG = NOISE_START; // NCR-style PSG resets the LFSR only on a mode write which actually changes the state of bit 2 of register 6 m_register[r] = (m_register[r] & 0x3f0) | (data & 0x0f); } else { r = m_last_register; //if ((m_ncr_style_psg) && ((r & 1) || (r == 6))) return; // NCR-style PSG ignores writes to regs 1, 3, 5, 6 and 7 with bit 7 clear; this behavior is not verified on hardware yet, uncomment it once verified. } c = r >> 1; switch (r) { case 0: // tone 0: frequency case 2: // tone 1: frequency case 4: // tone 2: frequency if ((data & 0x80) == 0) m_register[r] = (m_register[r] & 0x0f) | ((data & 0x3f) << 4); if ((m_register[r] != 0) || (!m_sega_style_psg)) m_period[c] = m_register[r]; else m_period[c] = 0x400; if (r == 4) { // update noise shift frequency if ((m_register[6] & 0x03) == 0x03) m_period[3] = m_period[2]<<1; } break; case 1: // tone 0: volume case 3: // tone 1: volume case 5: // tone 2: volume case 7: // noise: volume m_volume[c] = m_vol_table[data & 0x0f]; if ((data & 0x80) == 0) m_register[r] = (m_register[r] & 0x3f0) | (data & 0x0f); break; case 6: // noise: frequency, mode { if ((data & 0x80) == 0) printf("sn76496_base_device: write to reg 6 with bit 7 clear; data was %03x, new write is %02x! report this to LN!\n", m_register[6], data); if ((data & 0x80) == 0) m_register[r] = (m_register[r] & 0x3f0) | (data & 0x0f); n = m_register[6]; // N/512,N/1024,N/2048,Tone #3 output m_period[3] = ((n&3) == 3)? (m_period[2]<<1) : (1 << (5+(n&3))); if (!(m_ncr_style_psg)) m_RNG = NOISE_START; } break; } //m_ready_state = false; } inline bool sn76496_base_device::in_noise_mode() { return ((m_register[6] & 4)!=0); } void sn76496_base_device::sound_stream_update(short* outputs, int outLen) { int i; int16_t out; int16_t out2 = 0; for (int sampindex = 0; sampindex < outLen; sampindex++) { // clock chip once if (m_current_clock > 0) // not ready for new divided clock { m_current_clock--; } else // ready for new divided clock, make a new sample { m_current_clock = m_clock_divider-1; // handle channels 0,1,2 for (i = 0; i < 3; i++) { m_count[i]--; if (m_count[i] <= 0) { m_output[i] ^= 1; m_count[i] = m_period[i]; } } // handle channel 3 m_count[3]--; if (m_count[3] <= 0) { // if noisemode is 1, both taps are enabled // if noisemode is 0, the lower tap, whitenoisetap2, is held at 0 // The != was a bit-XOR (^) before if (((m_RNG & m_whitenoise_tap1)!=0) != (((int32_t)(m_RNG & m_whitenoise_tap2)!=(m_ncr_style_psg?m_whitenoise_tap2:0)) && in_noise_mode())) { m_RNG >>= 1; m_RNG |= m_feedback_mask; } else { m_RNG >>= 1; } m_output[3] = m_RNG & 1; m_count[3] = m_period[3]; } } out= ((m_output[0]!=0)? m_volume[0]:0) +((m_output[1]!=0)? m_volume[1]:0) +((m_output[2]!=0)? m_volume[2]:0) +((m_output[3]!=0)? m_volume[3]:0); if (m_negate) { out = -out; out2 = -out2; } outputs[sampindex]=out; } }