sceMp3: hand out the stream buffer in halves, like the hardware does

The area after the 0x5c0 workarea is double buffered - a half only becomes
writable again once the decoder has consumed past its end, so decoding a
single frame usually frees nothing at all. We instead reported every byte a
decode had just consumed, which made sceMp3CheckStreamDataNeeded() answer
"yes" after every single frame.

Beats sleeps 50ms whenever that call says the file thread is behind, so it
slept once per decoded frame and delivered audio at 46% of realtime - the
badly stuttering custom soundtracks. It now decodes 3-4 frames per 3360 byte
refill, with the write pointer alternating between the two halves exactly as
audio/mp3/stream records from hardware, and keeps up.

AuGetInfoToAddStreamData/AuNotifyAddStreamData now derive the write position
from how much has been added rather than from how much is still buffered,
since the write pointer walks the halves in turn and doesn't follow the
decoder.

Fixes audio/mp3/notifyadd, moved to tests_good, and the "after decode" case
in audio/mp3/checkneeded.

sceMp3: note that the half-buffer split is only verified at 8192 bytes

Core/HLE/sceMp3.cpp

@@ -60,9 +60,6 @@
 //
 // BUGS
 //
-// Custom music plays but starts stuttering:
-// * Beats
-//
 // Custom music just repeats a small section:
 // * Crazy Taxi
 

Core/HW/SimpleAudioDec.cpp

@@ -577,9 +577,29 @@ int AuCtx::AuStreamBytesNeeded() {
 		// stream instead of what the hardware reports.
 		if ((int64_t)readPos >= (int64_t)endPos)
 			return 0;
-		// Account for the workarea.
-		int offset = AuStreamWorkareaSize();
-		return (int)AuBufSize - AuBufAvailable - offset;
+
+		// The area after the workarea is double buffered: the game may write ahead up to the end
+		// of the half that follows the one the decoder is currently reading from, so a half only
+		// opens up once the decoder has consumed past its end. Decoding a single frame therefore
+		// usually frees nothing at all, which is what the hardware reports (audio/mp3/checkneeded).
+		// Games depend on it: Beats sleeps 50ms every time sceMp3CheckStreamDataNeeded() says it's
+		// behind, so handing back the bytes each decode consumed made it sleep once per frame and
+		// fall to less than half of realtime - badly stuttering custom soundtracks.
+		//
+		// Every case seen so far - the two hardware tests, Beats and Wipeout Pulse - passes the
+		// minimum 8192 byte buffer, so the split being exactly half is unverified for anything
+		// larger. If a game with a bigger buffer ever streams badly, suspect this first: the real
+		// granularity could be a fixed chunk size rather than half of whatever it was given.
+		int half = AuStreamHalfSize();
+		if (half <= 0)
+			return 0;
+		int64_t written = (int64_t)readPos - (int64_t)startPos;
+		int64_t consumed = written - AuBufAvailable;
+		// Floor division - consumed can go negative if a game notifies a negative size.
+		int64_t halvesDone = consumed / half - ((consumed % half < 0) ? 1 : 0);
+		// Note that this is deliberately not clamped to the buffer size. The hardware reports
+		// 6721 bytes to write for an 8192 byte buffer after notifying a size of -1.
+		return (int)std::max((int64_t)0, (halvesDone + 2) * half - written);
 	}
 
 	// TODO: Untested.  Maybe similar to MP3.
@@ -593,12 +613,29 @@ int AuCtx::AuStreamWorkareaSize() {
 	return 0;
 }
 
+// Size of each of the two halves the stream buffer is split into, after the workarea.
+int AuCtx::AuStreamHalfSize() {
+	return ((int)AuBufSize - AuStreamWorkareaSize()) / 2;
+}
+
+// Offset into the stream buffer (past the workarea) that the next added bytes go to. The write
+// position simply walks the two halves in turn and wraps around, it doesn't follow the decoder.
+int AuCtx::AuStreamWriteOffset() {
+	int size = AuStreamHalfSize() * 2;
+	if (size <= 0)
+		return 0;
+	int64_t pos = ((int64_t)readPos - (int64_t)startPos) % size;
+	if (pos < 0)
+		pos += size;
+	return (int)pos;
+}
+
 // check how many bytes we have read from source file
 u32 AuCtx::AuNotifyAddStreamData(int size) {
 	int offset = AuStreamWorkareaSize();
 	// Where AuGetInfoToAddStreamData pointed the game, i.e. where the bytes it just added start.
-	// Data accumulates in the buffer rather than always landing at the beginning.
-	const int writeOffset = AuBufAvailable;
+	// Has to be sampled before readPos moves on below.
+	const int writeOffset = AuStreamWriteOffset();
 
 	if (askedReadSize != 0) {
 		// Old save state, numbers already adjusted.
@@ -634,11 +671,11 @@ u32 AuCtx::AuGetInfoToAddStreamData(u32 bufPtr, u32 sizePtr, u32 srcPosPtr) {
 	int readsize = AuStreamBytesNeeded();
 	int offset = AuStreamWorkareaSize();
 
-	// The game appends to what's already buffered, so point it past that rather than at the
-	// start of the work area - the hardware's pointer walks forward as data is added.
+	// The write position walks forward through the two halves as data is added and wraps around,
+	// so point the game at that rather than at the start of the work area.
 	if (readsize != 0) {
 		if (Memory::IsValidAddress(bufPtr))
-			Memory::WriteUnchecked_U32(AuBuf + offset + AuBufAvailable, bufPtr);
+			Memory::WriteUnchecked_U32(AuBuf + offset + AuStreamWriteOffset(), bufPtr);
 		if (Memory::IsValidAddress(sizePtr))
 			Memory::WriteUnchecked_U32(readsize, sizePtr);
 		if (Memory::IsValidAddress(srcPosPtr))

Core/HW/SimpleAudioDec.h

@@ -64,6 +64,8 @@ public:
 	int AuCheckStreamDataNeeded();
 	int AuStreamBytesNeeded();
 	int AuStreamWorkareaSize();
+	int AuStreamHalfSize();
+	int AuStreamWriteOffset();
 	u32 AuResetPlayPosition();
 	u32 AuResetPlayPositionByFrame(int position);
 	u32 AuGetInfoToAddStreamData(u32 bufPtr, u32 sizePtr, u32 srcPosPtr);

test.py

@@ -108,6 +108,7 @@ tests_good = [
   "audio/mp3/infotoadd",
   "audio/mp3/initresource",
   "audio/mp3/mp3test",
+  "audio/mp3/notifyadd",
   "audio/mp3/release",
   "audio/mp3/reserve",
   "audio/mp3/setloopnum",
@@ -402,7 +403,6 @@ tests_next = [
   "audio/sceaudio/reserve",
   "audio/sascore/setadsr",
   "audio/mp3/init",
-  "audio/mp3/notifyadd",
   "audio/output2/frequency",
   "audio/output2/release",
   "audio/output2/rest",