More accurate emulation of the APU’s analog characteristics

Core/apu.c

@@ -23,8 +23,15 @@ static void refresh_channel(GB_gameboy_t *gb, unsigned index, unsigned cycles_of
 
 static void update_sample(GB_gameboy_t *gb, unsigned index, int8_t value, unsigned cycles_offset)
 {
+    if (!gb->apu.is_active[index]) {
+        value = gb->apu.samples[index];
+    }
+    else {
         gb->apu.samples[index] = value;
-    if (gb->apu.is_active[index] && gb->apu_output.sample_rate) {
+        gb->apu_output.dac_discharge[index] = 1.0;
+    }
+    
+    if (gb->apu_output.sample_rate) {
         unsigned left_volume = 0;
         if (gb->io_registers[GB_IO_NR51] & (1 << index)) {
             left_volume = (gb->io_registers[GB_IO_NR50] & 7) + 1;
@@ -41,19 +48,30 @@ static void update_sample(GB_gameboy_t *gb, unsigned index, int8_t value, unsign
     }
 }
 
-static void render(GB_gameboy_t *gb)
+static void render(GB_gameboy_t *gb, bool no_downsampling, GB_sample_t *dest)
 {
     GB_sample_t output = {0,0};
     for (unsigned i = GB_N_CHANNELS; i--;) {
-        if (likely(gb->apu_output.last_update[i] == 0)) {
+        double multiplier = CH_STEP;
-            output.left += gb->apu_output.current_sample[i].left * CH_STEP;
+        if (!gb->apu.is_active[i]) {
-            output.right += gb->apu_output.current_sample[i].right * CH_STEP;
+            gb->apu_output.dac_discharge[i] -= ((double) DAC_DECAY_SPEED) / gb->apu_output.sample_rate;
+            if (gb->apu_output.dac_discharge[i] < 0) {
+                multiplier = 0;
+            }
+            else {
+                multiplier *= pow(0.05, 1 - gb->apu_output.dac_discharge[i]) * (gb->apu_output.dac_discharge[i]);
+            }
+        }
+
+        if (likely(gb->apu_output.last_update[i] == 0 || no_downsampling)) {
+            output.left += gb->apu_output.current_sample[i].left * multiplier;
+            output.right += gb->apu_output.current_sample[i].right * multiplier;
         }
         else {
             refresh_channel(gb, i, 0);
-            output.left += (signed long) gb->apu_output.summed_samples[i].left * CH_STEP
+            output.left += (signed long) gb->apu_output.summed_samples[i].left * multiplier
                             / gb->apu_output.cycles_since_render;
-            output.right += (signed long) gb->apu_output.summed_samples[i].right * CH_STEP
+            output.right += (signed long) gb->apu_output.summed_samples[i].right * multiplier
                             / gb->apu_output.cycles_since_render;
             gb->apu_output.summed_samples[i] = (GB_sample_t){0, 0};
         }
@@ -61,8 +79,6 @@ static void render(GB_gameboy_t *gb)
     }
     gb->apu_output.cycles_since_render = 0;
 
-
-
     GB_sample_t filtered_output = gb->apu_output.highpass_mode?
         (GB_sample_t) {output.left - gb->apu_output.highpass_diff.left,
                        output.right - gb->apu_output.highpass_diff.right} :
@@ -104,6 +120,10 @@ static void render(GB_gameboy_t *gb)
         }
             
     }
+    if (dest) {
+        *dest = filtered_output;
+        return;
+    }
 
     while (gb->apu_output.copy_in_progress);
     while (!__sync_bool_compare_and_swap(&gb->apu_output.lock, false, true));
@@ -369,7 +389,7 @@ void GB_apu_run(GB_gameboy_t *gb)
         
         if (gb->apu_output.sample_cycles > cycles_per_sample) {
             gb->apu_output.sample_cycles -= cycles_per_sample;
-            render(gb);
+            render(gb, false, NULL);
         }
     }
 }
@@ -386,15 +406,13 @@ void GB_apu_copy_buffer(GB_gameboy_t *gb, GB_sample_t *dest, size_t count)
 
     if (count > gb->apu_output.buffer_position) {
         // GB_log(gb, "Audio underflow: %d\n", count - gb->apu_output.buffer_position);
-        GB_sample_t output = {-gb->apu_output.highpass_diff.left, -gb->apu_output.highpass_diff.right};
+        GB_sample_t output;
-        for (unsigned i = GB_N_CHANNELS; i--;) {
+        render(gb, true, &output);
-            output.left += gb->apu_output.current_sample[i].left * CH_STEP;
-            output.right += gb->apu_output.current_sample[i].right * CH_STEP;
-        }
         
         for (unsigned i = 0; i < count - gb->apu_output.buffer_position; i++) {
             dest[gb->apu_output.buffer_position + i] = output;
         }
+        
         count = gb->apu_output.buffer_position;
     }
     memcpy(dest, gb->apu_output.buffer, count * sizeof(*gb->apu_output.buffer));

Core/apu.h

@@ -6,6 +6,9 @@
 
 
 #ifdef GB_INTERNAL
+/* Speed = 1 / Length (in seconds) */
+#define DAC_DECAY_SPEED 500.0
+
 /* Divides nicely and never overflows with 4 channels and 8 (1-8) volume levels */
 #ifdef WIIU
 /* Todo: Remove this hack once https://github.com/libretro/RetroArch/issues/6252 is fixed*/
@@ -16,6 +19,8 @@
 #define CH_STEP (MAX_CH_AMP/0xF/8)
 #endif
 
+
+
 /* APU ticks are 2MHz, triggered by an internal APU clock. */
 
 typedef struct
@@ -129,6 +134,7 @@ typedef struct {
     unsigned last_update[GB_N_CHANNELS];
     GB_sample_t current_sample[GB_N_CHANNELS];
     GB_sample_t summed_samples[GB_N_CHANNELS];
+    double dac_discharge[GB_N_CHANNELS];
     
     GB_highpass_mode_t highpass_mode;
     double highpass_rate;

Core/memory.c

@@ -149,10 +149,12 @@ static uint8_t read_high_memory(GB_gameboy_t *gb, uint16_t addr)
 
             case GB_IO_PCM_12:
                 if (!gb->is_cgb) return 0xFF;
-                return (gb->apu.samples[GB_SQUARE_2] << 4) | gb->apu.samples[GB_SQUARE_1];
+                return (gb->apu.is_active[GB_SQUARE_2] ? (gb->apu.samples[GB_SQUARE_2] << 4) : 0) |
+                        (gb->apu.is_active[GB_SQUARE_1] ? (gb->apu.samples[GB_SQUARE_1]) : 0);
             case GB_IO_PCM_34:
                 if (!gb->is_cgb) return 0xFF;
-                return (gb->apu.samples[GB_NOISE] << 4) | gb->apu.samples[GB_WAVE];
+                return (gb->apu.is_active[GB_NOISE] ? (gb->apu.samples[GB_NOISE] << 4) : 0) |
+                       (gb->apu.is_active[GB_WAVE] ? (gb->apu.samples[GB_WAVE]) : 0);
             case GB_IO_JOYP:
             case GB_IO_TMA:
             case GB_IO_LCDC: