void Spc_Dsp::run()

in doom_py/src/vizdoom/game-music-emu/gme/Spc_Dsp.cpp [186:626]


void Spc_Dsp::run( int clock_count )
{
	int new_phase = m.phase + clock_count;
	int count = new_phase >> 5;
	m.phase = new_phase & 31;
	if ( !count )
		return;
	
	uint8_t* const ram = m.ram;
	uint8_t const* const dir = &ram [REG(dir) * 0x100];
	int const slow_gaussian = (REG(pmon) >> 1) | REG(non);
	int const noise_rate = REG(flg) & 0x1F;
	
	// Global volume
	int mvoll = (int8_t) REG(mvoll);
	int mvolr = (int8_t) REG(mvolr);
	if ( mvoll * mvolr < m.surround_threshold )
		mvoll = -mvoll; // eliminate surround
	
	do
	{
		// KON/KOFF reading
		if ( (m.every_other_sample ^= 1) != 0 )
		{
			m.new_kon &= ~m.kon;
			m.kon    = m.new_kon;
			m.t_koff = REG(koff); 
		}
		
		run_counter( 1 );
		run_counter( 2 );
		run_counter( 3 );
		
		// Noise
		if ( !READ_COUNTER( noise_rate ) )
		{
			int feedback = (m.noise << 13) ^ (m.noise << 14);
			m.noise = (feedback & 0x4000) ^ (m.noise >> 1);
		}
		
		// Voices
		int pmon_input = 0;
		int main_out_l = 0;
		int main_out_r = 0;
		int echo_out_l = 0;
		int echo_out_r = 0;
		voice_t* v = m.voices;
		uint8_t* v_regs = m.regs;
		int vbit = 1;
		do
		{
			#define SAMPLE_PTR(i) GET_LE16A( &dir [VREG(v_regs,srcn) * 4 + i * 2] )
			
			int brr_header = ram [v->brr_addr];
			int kon_delay = v->kon_delay;
			
			// Pitch
			int pitch = GET_LE16A( &VREG(v_regs,pitchl) ) & 0x3FFF;
			if ( REG(pmon) & vbit )
				pitch += ((pmon_input >> 5) * pitch) >> 10;
			
			// KON phases
			if ( --kon_delay >= 0 )
			{
				v->kon_delay = kon_delay;
				
				// Get ready to start BRR decoding on next sample
				if ( kon_delay == 4 )
				{
					v->brr_addr   = SAMPLE_PTR( 0 );
					v->brr_offset = 1;
					v->buf_pos    = v->buf;
					brr_header    = 0; // header is ignored on this sample
				}
				
				// Envelope is never run during KON
				v->env        = 0;
				v->hidden_env = 0;
				
				// Disable BRR decoding until last three samples
				v->interp_pos = (kon_delay & 3 ? 0x4000 : 0);
				
				// Pitch is never added during KON
				pitch = 0;
			}
			
			int env = v->env;
			
			// Gaussian interpolation
			{
				int output = 0;
				VREG(v_regs,envx) = (uint8_t) (env >> 4);
				if ( env )
				{
					// Make pointers into gaussian based on fractional position between samples
					int offset = (unsigned) v->interp_pos >> 3 & 0x1FE;
					short const* fwd = interleved_gauss       + offset;
					short const* rev = interleved_gauss + 510 - offset; // mirror left half of gaussian
					
					int const* in = &v->buf_pos [(unsigned) v->interp_pos >> 12];
					
					if ( !(slow_gaussian & vbit) ) // 99%
					{
						// Faster approximation when exact sample value isn't necessary for pitch mod
						output = (fwd [0] * in [0] +
						          fwd [1] * in [1] +
						          rev [1] * in [2] +
						          rev [0] * in [3]) >> 11;
						output = (output * env) >> 11;
					}
					else
					{
						output = (int16_t) (m.noise * 2);
						if ( !(REG(non) & vbit) )
						{
							output  = (fwd [0] * in [0]) >> 11;
							output += (fwd [1] * in [1]) >> 11;
							output += (rev [1] * in [2]) >> 11;
							output = (int16_t) output;
							output += (rev [0] * in [3]) >> 11;
							
							CLAMP16( output );
							output &= ~1;
						}
						output = (output * env) >> 11 & ~1;
					}
					
					// Output
					int l = output * v->volume [0];
					int r = output * v->volume [1];
					
					main_out_l += l;
					main_out_r += r;
					
					if ( REG(eon) & vbit )
					{
						echo_out_l += l;
						echo_out_r += r;
					}
				}
				
				pmon_input = output;
				VREG(v_regs,outx) = (uint8_t) (output >> 8);
			}
			
			// Soft reset or end of sample
			if ( REG(flg) & 0x80 || (brr_header & 3) == 1 )
			{
				v->env_mode = env_release;
				env         = 0;
			}
			
			if ( m.every_other_sample )
			{
				// KOFF
				if ( m.t_koff & vbit )
					v->env_mode = env_release;
				
				// KON
				if ( m.kon & vbit )
				{
					v->kon_delay = 5;
					v->env_mode  = env_attack;
					REG(endx) &= ~vbit;
				}
			}
			
			// Envelope
			if ( !v->kon_delay )
			{
				if ( v->env_mode == env_release ) // 97%
				{
					env -= 0x8;
					v->env = env;
					if ( env <= 0 )
					{
						v->env = 0;
						goto skip_brr; // no BRR decoding for you!
					}
				}
				else // 3%
				{
					int rate;
					int const adsr0 = VREG(v_regs,adsr0);
					int env_data = VREG(v_regs,adsr1);
					if ( adsr0 >= 0x80 ) // 97% ADSR
					{
						if ( v->env_mode > env_decay ) // 89%
						{
							env--;
							env -= env >> 8;
							rate = env_data & 0x1F;
							
							// optimized handling
							v->hidden_env = env;
							if ( READ_COUNTER( rate ) )
								goto exit_env;
							v->env = env;
							goto exit_env;
						}
						else if ( v->env_mode == env_decay )
						{
							env--;
							env -= env >> 8;
							rate = (adsr0 >> 3 & 0x0E) + 0x10;
						}
						else // env_attack
						{
							rate = (adsr0 & 0x0F) * 2 + 1;
							env += rate < 31 ? 0x20 : 0x400;
						}
					}
					else // GAIN
					{
						int mode;
						env_data = VREG(v_regs,gain);
						mode = env_data >> 5;
						if ( mode < 4 ) // direct
						{
							env = env_data * 0x10;
							rate = 31;
						}
						else
						{
							rate = env_data & 0x1F;
							if ( mode == 4 ) // 4: linear decrease
							{
								env -= 0x20;
							}
							else if ( mode < 6 ) // 5: exponential decrease
							{
								env--;
								env -= env >> 8;
							}
							else // 6,7: linear increase
							{
								env += 0x20;
								if ( mode > 6 && (unsigned) v->hidden_env >= 0x600 )
									env += 0x8 - 0x20; // 7: two-slope linear increase
							}
						}
					}
					
					// Sustain level
					if ( (env >> 8) == (env_data >> 5) && v->env_mode == env_decay )
						v->env_mode = env_sustain;
					
					v->hidden_env = env;
					
					// unsigned cast because linear decrease going negative also triggers this
					if ( (unsigned) env > 0x7FF )
					{
						env = (env < 0 ? 0 : 0x7FF);
						if ( v->env_mode == env_attack )
							v->env_mode = env_decay;
					}
					
					if ( !READ_COUNTER( rate ) )
						v->env = env; // nothing else is controlled by the counter
				}
			}
		exit_env:
			
			{
				// Apply pitch
				int old_pos = v->interp_pos;
				int interp_pos = (old_pos & 0x3FFF) + pitch;
				if ( interp_pos > 0x7FFF )
					interp_pos = 0x7FFF;
				v->interp_pos = interp_pos;
				
				// BRR decode if necessary
				if ( old_pos >= 0x4000 )
				{
					// Arrange the four input nybbles in 0xABCD order for easy decoding
					int nybbles = ram [(v->brr_addr + v->brr_offset) & 0xFFFF] * 0x100 +
							ram [(v->brr_addr + v->brr_offset + 1) & 0xFFFF];
					
					// Advance read position
					int const brr_block_size = 9;
					int brr_offset = v->brr_offset;
					if ( (brr_offset += 2) >= brr_block_size )
					{
						// Next BRR block
						int brr_addr = (v->brr_addr + brr_block_size) & 0xFFFF;
						assert( brr_offset == brr_block_size );
						if ( brr_header & 1 )
						{
							brr_addr = SAMPLE_PTR( 1 );
							if ( !v->kon_delay )
								REG(endx) |= vbit;
						}
						v->brr_addr = brr_addr;
						brr_offset  = 1;
					}
					v->brr_offset = brr_offset;
					
					// Decode
					
					// 0: >>1  1: <<0  2: <<1 ... 12: <<11  13-15: >>4 <<11
					static unsigned char const shifts [16 * 2] = {
						13,12,12,12,12,12,12,12,12,12,12, 12, 12, 16, 16, 16,
						 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 11, 11, 11
					};
					int const scale = brr_header >> 4;
					int const right_shift = shifts [scale];
					int const left_shift  = shifts [scale + 16];
					
					// Write to next four samples in circular buffer
					int* pos = v->buf_pos;
					int* end;
					
					// Decode four samples
					for ( end = pos + 4; pos < end; pos++, nybbles <<= 4 )
					{
						// Extract upper nybble and scale appropriately
						int s = ((int16_t) nybbles >> right_shift) << left_shift;
						
						// Apply IIR filter (8 is the most commonly used)
						int const filter = brr_header & 0x0C;
						int const p1 = pos [brr_buf_size - 1];
						int const p2 = pos [brr_buf_size - 2] >> 1;
						if ( filter >= 8 )
						{
							s += p1;
							s -= p2;
							if ( filter == 8 ) // s += p1 * 0.953125 - p2 * 0.46875
							{
								s += p2 >> 4;
								s += (p1 * -3) >> 6;
							}
							else // s += p1 * 0.8984375 - p2 * 0.40625
							{
								s += (p1 * -13) >> 7;
								s += (p2 * 3) >> 4;
							}
						}
						else if ( filter ) // s += p1 * 0.46875
						{
							s += p1 >> 1;
							s += (-p1) >> 5;
						}
						
						// Adjust and write sample
						CLAMP16( s );
						s = (int16_t) (s * 2);
						pos [brr_buf_size] = pos [0] = s; // second copy simplifies wrap-around
					}
					
					if ( pos >= &v->buf [brr_buf_size] )
						pos = v->buf;
					v->buf_pos = pos;
				}
			}
skip_brr:
			// Next voice
			vbit <<= 1;
			v_regs += 0x10;
			v++;
		}
		while ( vbit < 0x100 );
		
		// Echo position
		int echo_offset = m.echo_offset;
		uint8_t* const echo_ptr = &ram [(REG(esa) * 0x100 + echo_offset) & 0xFFFF];
		if ( !echo_offset )
			m.echo_length = (REG(edl) & 0x0F) * 0x800;
		echo_offset += 4;
		if ( echo_offset >= m.echo_length )
			echo_offset = 0;
		m.echo_offset = echo_offset;
		
		// FIR
		int echo_in_l = GET_LE16SA( echo_ptr + 0 );
		int echo_in_r = GET_LE16SA( echo_ptr + 2 );
		
		int (*echo_hist_pos) [2] = m.echo_hist_pos;
		if ( ++echo_hist_pos >= &m.echo_hist [echo_hist_size] )
			echo_hist_pos = m.echo_hist;
		m.echo_hist_pos = echo_hist_pos;
		
		echo_hist_pos [0] [0] = echo_hist_pos [8] [0] = echo_in_l;
		echo_hist_pos [0] [1] = echo_hist_pos [8] [1] = echo_in_r;
		
		#define CALC_FIR_( i, in )  ((in) * (int8_t) REG(fir + i * 0x10))
		echo_in_l = CALC_FIR_( 7, echo_in_l );
		echo_in_r = CALC_FIR_( 7, echo_in_r );
		
		#define CALC_FIR( i, ch )   CALC_FIR_( i, echo_hist_pos [i + 1] [ch] )
		#define DO_FIR( i )\
			echo_in_l += CALC_FIR( i, 0 );\
			echo_in_r += CALC_FIR( i, 1 );
		DO_FIR( 0 );
		DO_FIR( 1 );
		DO_FIR( 2 );
		#if defined (__MWERKS__) && __MWERKS__ < 0x3200
			__eieio(); // keeps compiler from stupidly "caching" things in memory
		#endif
		DO_FIR( 3 );
		DO_FIR( 4 );
		DO_FIR( 5 );
		DO_FIR( 6 );
		
		// Echo out
		if ( !(REG(flg) & 0x20) )
		{
			int l = (echo_out_l >> 7) + ((echo_in_l * (int8_t) REG(efb)) >> 14);
			int r = (echo_out_r >> 7) + ((echo_in_r * (int8_t) REG(efb)) >> 14);
			
			// just to help pass more validation tests
			#if SPC_MORE_ACCURACY
				l &= ~1;
				r &= ~1;
			#endif
			
			CLAMP16( l );
			CLAMP16( r );
			
			SET_LE16A( echo_ptr + 0, l );
			SET_LE16A( echo_ptr + 2, r );
		}
		
		// Sound out
		int l = (main_out_l * mvoll + echo_in_l * (int8_t) REG(evoll)) >> 14;
		int r = (main_out_r * mvolr + echo_in_r * (int8_t) REG(evolr)) >> 14;
		
		CLAMP16( l );
		CLAMP16( r );
		
		if ( (REG(flg) & 0x40) )
		{
			l = 0;
			r = 0;
		}
		
		sample_t* out = m.out;
		WRITE_SAMPLES( l, r, out );
		m.out = out;
	}
	while ( --count );
}