@@ -48,8 +48,6 @@ CircleBufferInit(&audio->left, GBA_AUDIO_SAMPLES * sizeof(int32_t));
 	CircleBufferInit(&audio->right, GBA_AUDIO_SAMPLES * sizeof(int32_t));
 	CircleBufferInit(&audio->chA.fifo, GBA_AUDIO_FIFO_SIZE);
 	CircleBufferInit(&audio->chB.fifo, GBA_AUDIO_FIFO_SIZE);
-
-	pthread_mutex_init(&audio->bufferMutex, 0);
 }
 
 void GBAAudioDeinit(struct GBAAudio* audio) {
@@ -57,9 +55,6 @@ CircleBufferDeinit(&audio->left);
 	CircleBufferDeinit(&audio->right);
 	CircleBufferDeinit(&audio->chA.fifo);
 	CircleBufferDeinit(&audio->chB.fifo);
-
-	pthread_mutex_lock(&audio->bufferMutex);
-	pthread_mutex_destroy(&audio->bufferMutex);
 }
 
 int32_t GBAAudioProcessEvents(struct GBAAudio* audio, int32_t cycles) {
@@ -397,6 +392,27 @@ }
 	CircleBufferRead8(&channel->fifo, &channel->sample);
 }
 
+unsigned GBAAudioCopy(struct GBAAudio* audio, void* left, void* right, unsigned nSamples) {
+	GBASyncLockAudio(audio->p->sync);
+	unsigned read = 0;
+	if (left) {
+		unsigned readL = CircleBufferRead(&audio->left, left, nSamples * sizeof(int32_t)) >> 2;
+		if (readL < nSamples) {
+			memset((int32_t*) left + readL, 0, nSamples - readL);
+		}
+		read = readL;
+	}
+	if (right) {
+		unsigned readR = CircleBufferRead(&audio->right, right, nSamples * sizeof(int32_t)) >> 2;
+		if (readR < nSamples) {
+			memset((int32_t*) right + readR, 0, nSamples - readR);
+		}
+		read = read >= readR ? read : readR;
+	}
+	GBASyncConsumeAudio(audio->p->sync);
+	return read;
+}
+
 static int32_t _updateSquareChannel(struct GBAAudioSquareControl* control, int duty) {
 	control->hi = !control->hi;
 	int period = 16 * (2048 - control->frequency);
@@ -579,14 +595,9 @@ if (audio->chBRight) {
 		sampleRight += (audio->chB.sample << 2) >> !audio->volumeChB;
 	}
 
-	pthread_mutex_lock(&audio->bufferMutex);
-	while (CircleBufferSize(&audio->left) + (GBA_AUDIO_SAMPLES * 2 / 5) >= audio->left.capacity) {
-		if (!audio->p->sync->audioWait) {
-			break;
-		}
-		GBASyncProduceAudio(audio->p->sync, &audio->bufferMutex);
-	}
-	CircleBufferWrite32(&audio->left, sampleLeft);
-	CircleBufferWrite32(&audio->right, sampleRight);
-	pthread_mutex_unlock(&audio->bufferMutex);
+	GBASyncLockAudio(audio->p->sync);
+	CircleBufferWrite32(&audio->left, sampleLeft << 5);
+	CircleBufferWrite32(&audio->right, sampleRight << 5);
+	unsigned produced = CircleBufferSize(&audio->left);
+	GBASyncProduceAudio(audio->p->sync, produced >= GBA_AUDIO_SAMPLES * 3);
 }

@@ -3,7 +3,6 @@ #define GBA_AUDIO_H
 
 #include "circle-buffer.h"
 
-#include <pthread.h>
 #include <stdint.h>
 
 struct GBADMA;
@@ -203,8 +202,6 @@ int32_t nextCh4;
 	int32_t nextSample;
 
 	int32_t sampleInterval;
-
-	pthread_mutex_t bufferMutex;
 };
 
 void GBAAudioInit(struct GBAAudio* audio);
@@ -230,5 +227,7 @@
 void GBAAudioWriteWaveRAM(struct GBAAudio* audio, int address, uint32_t value);
 void GBAAudioWriteFIFO(struct GBAAudio* audio, int address, uint32_t value);
 void GBAAudioSampleFIFO(struct GBAAudio* audio, int fifoId);
+
+unsigned GBAAudioCopy(struct GBAAudio* audio, void* left, void* right, unsigned nSamples);
 
 #endif

@@ -135,6 +135,7 @@
 	pthread_mutex_init(&threadContext->sync.videoFrameMutex, 0);
 	pthread_cond_init(&threadContext->sync.videoFrameAvailableCond, 0);
 	pthread_cond_init(&threadContext->sync.videoFrameRequiredCond, 0);
+	pthread_mutex_init(&threadContext->sync.audioBufferMutex, 0);
 	pthread_cond_init(&threadContext->sync.audioRequiredCond, 0);
 
 	pthread_mutex_lock(&threadContext->stateMutex);
@@ -168,6 +169,7 @@ pthread_cond_destroy(&threadContext->sync.videoFrameRequiredCond);
 
 	pthread_cond_broadcast(&threadContext->sync.audioRequiredCond);
 	pthread_cond_destroy(&threadContext->sync.audioRequiredCond);
+	pthread_mutex_destroy(&threadContext->sync.audioBufferMutex);
 }
 
 void GBAThreadPause(struct GBAThread* threadContext) {
@@ -277,12 +279,18 @@ int GBASyncDrawingFrame(struct GBASync* sync) {
 	return sync->videoFrameSkip <= 0;
 }
 
-void GBASyncProduceAudio(struct GBASync* sync, pthread_mutex_t* mutex) {
-	if (&sync->audioWait) {
-		pthread_cond_wait(&sync->audioRequiredCond, mutex);
+void GBASyncProduceAudio(struct GBASync* sync, int wait) {
+	if (sync->audioWait && wait) {
+		pthread_cond_wait(&sync->audioRequiredCond, &sync->audioBufferMutex);
 	}
+	pthread_mutex_unlock(&sync->audioBufferMutex);
+}
+
+void GBASyncLockAudio(struct GBASync* sync) {
+	pthread_mutex_lock(&sync->audioBufferMutex);
 }
 
 void GBASyncConsumeAudio(struct GBASync* sync) {
 	pthread_cond_broadcast(&sync->audioRequiredCond);
+	pthread_mutex_unlock(&sync->audioBufferMutex);
 }

@@ -49,6 +49,7 @@ pthread_cond_t videoFrameRequiredCond;
 
 		int audioWait;
 		pthread_cond_t audioRequiredCond;
+		pthread_mutex_t audioBufferMutex;
 	} sync;
 };
 
@@ -65,7 +66,8 @@ int GBASyncWaitFrameStart(struct GBASync* sync, int frameskip);
 void GBASyncWaitFrameEnd(struct GBASync* sync);
 int GBASyncDrawingFrame(struct GBASync* sync);
 
-void GBASyncProduceAudio(struct GBASync* sync, pthread_mutex_t* mutex);
+void GBASyncProduceAudio(struct GBASync* sync, int wait);
+void GBASyncLockAudio(struct GBASync* sync);
 void GBASyncConsumeAudio(struct GBASync* sync);
 
 #endif

@@ -3,6 +3,13 @@
 #include "gba.h"
 #include "gba-thread.h"
 
+#define BUFFER_SIZE (GBA_AUDIO_SAMPLES >> 2)
+
+struct StereoSample {
+	Sint16 left;
+	Sint16 right;
+};
+
 static void _GBASDLAudioCallback(void* context, Uint8* data, int len);
 
 int GBASDLInitAudio(struct GBASDLAudio* context) {
@@ -34,42 +41,52 @@ SDL_CloseAudio();
 	SDL_QuitSubSystem(SDL_INIT_AUDIO);
 }
 
-static void _pulldownResample(struct GBASDLAudio* context) {
-	int32_t value;
-	if (CircleBufferRead32(&context->audio->left, &value)) {
-		context->currentSample.left = value << 5;
-	} else {
-		context->currentSample.left = 0;
-	}
-	if (CircleBufferRead32(&context->audio->right, &value)) {
-		context->currentSample.right = value << 5;
-	} else {
-		context->currentSample.right = 0;
+static void _pulldownResample(struct GBASDLAudio* context, struct StereoSample* output, ssize_t samples) {
+	int32_t left[BUFFER_SIZE];
+	int32_t right[BUFFER_SIZE];
+
+	// toRead is in GBA samples
+	int toRead = samples / context->ratio;
+	while (samples > 0) {
+		int currentRead = BUFFER_SIZE >> 2;
+		if (currentRead > toRead) {
+			currentRead = toRead;
+		}
+		unsigned read = GBAAudioCopy(context->audio, left, right, currentRead);
+		toRead -= read;
+		unsigned i;
+		for (i = 0; i < read; ++i) {
+			context->drift += context->ratio;
+			while (context->drift >= 0) {
+				output->left = left[i];
+				output->right = right[i];
+				++output;
+				--samples;
+#ifndef NDEBUG
+				if (samples < 0) {
+					abort();
+				}
+#endif
+				context->drift -= 1.f;
+			}
+		}
+		if (read < BUFFER_SIZE >> 2) {
+			memset(output, 0, toRead);
+			return;
+		}
 	}
 }
 
 static void _GBASDLAudioCallback(void* context, Uint8* data, int len) {
 	struct GBASDLAudio* audioContext = context;
-	int i;
 	if (!context || !audioContext->audio) {
-		for (i = 0; i < len; ++i) {
-			data[i] = 0;
-		}
+		memset(data, 0, len);
 		return;
 	}
+	audioContext->ratio = audioContext->obtainedSpec.freq / (float) audioContext->audio->sampleRate;
 	struct StereoSample* ssamples = (struct StereoSample*) data;
 	len /= 2 * audioContext->obtainedSpec.channels;
 	if (audioContext->obtainedSpec.channels == 2) {
-		pthread_mutex_lock(&audioContext->audio->bufferMutex);
-		for (i = 0; i < len; ++i) {
-			audioContext->drift += audioContext->audio->sampleRate / (float) audioContext->obtainedSpec.freq;
-			while (audioContext->drift >= 0) {
-				_pulldownResample(audioContext);
-				audioContext->drift -= 1.f;
-			}
-			ssamples[i] = audioContext->currentSample;
-		}
-		GBASyncConsumeAudio(audioContext->audio->p->sync);
-		pthread_mutex_unlock(&audioContext->audio->bufferMutex);
+		_pulldownResample(audioContext, ssamples, len);
 	}
 }

@@ -3,17 +3,12 @@ #define SDL_AUDIO_H
 
 #include <SDL.h>
 
-struct StereoSample {
-	Sint16 left;
-	Sint16 right;
-};
-
 struct GBASDLAudio {
 	SDL_AudioSpec desiredSpec;
 	SDL_AudioSpec obtainedSpec;
 	float drift;
+	float ratio;
 	struct GBAAudio* audio;
-	struct StereoSample currentSample;
 };
 
 int GBASDLInitAudio(struct GBASDLAudio* context);

@@ -87,3 +87,29 @@ }
 	buffer->size -= sizeof(int32_t);
 	return 1;
 }
+
+int CircleBufferRead(struct CircleBuffer* buffer, void* output, size_t length) {
+	int8_t* data = buffer->readPtr;
+	if (buffer->size == 0) {
+		return 0;
+	}
+	if (length > buffer->size) {
+		length = buffer->size;
+	}
+	size_t remaining = buffer->capacity - ((int8_t*) data - (int8_t*) buffer->data);
+	if (length <= remaining) {
+		memcpy(output, data, length);
+		if (length == remaining) {
+			buffer->readPtr = buffer->data;
+		} else {
+			buffer->readPtr = (int8_t*) data + length;
+		}
+	} else {
+		memcpy(output, data, remaining);
+		memcpy((int8_t*) output + remaining, buffer->data, length - remaining);
+		buffer->readPtr = (int8_t*) buffer->data + length - remaining;
+	}
+
+	buffer->size -= length;
+	return length;
+}

@@ -2,6 +2,7 @@ #ifndef CIRCLE_BUFFER_H
 #define CIRCLE_BUFFER_H
 
 #include <stdint.h>
+#include <string.h>
 
 struct CircleBuffer {
 	void* data;
@@ -18,5 +19,6 @@ int CircleBufferWrite8(struct CircleBuffer* buffer, int8_t value);
 int CircleBufferWrite32(struct CircleBuffer* buffer, int32_t value);
 int CircleBufferRead8(struct CircleBuffer* buffer, int8_t* value);
 int CircleBufferRead32(struct CircleBuffer* buffer, int32_t* value);
+int CircleBufferRead(struct CircleBuffer* buffer, void* output, size_t length);
 
 #endif