sceAtrac: Charge setting data and decoding what the ME takes

Setting data decodes and throws away the frames before the first sample,
so on hardware it costs a decoder setup plus that decode, with the caller
waiting: ~900us for mono Atrac3 and ~3.5ms for stereo Atrac3+, whatever the
buffer size (pspautotests threads/scheduling/callcosts). It was charged
100us.

Decoding a frame now costs what the same frame costs through
sceAudiocodecDecode, through the shared ME queue, instead of a flat
2300us. That's about the same for stereo Atrac3+ and less for Atrac3
(685us mono, ~1100us stereo). The first Atrac3+ frames after setup still
come out ~500us short.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

Core/HLE/AtracBase.h

@@ -222,6 +222,8 @@ public:
 	virtual int ResetPlayPosition(int sample, int bytesWrittenFirstBuf, int bytesWrittenSecondBuf, bool *delay) = 0;
 	virtual int GetBufferInfoForResetting(AtracResetBufferInfo *bufferInfo, int sample, bool *delay) = 0;  // NOTE: Not const! This can cause SkipFrames!
 	virtual int SetData(const Track &track, u32 buffer, u32 readSize, u32 bufferSize, u32 fileSize, int outputChannels, bool isAA3) = 0;
+	// How many frames the last SetData decoded and threw away, to get to the first sample.
+	int SkippedFramesOnSetData() const { return setDataSkippedFrames_; }
 
 	virtual int GetSecondBufferInfo(u32 *fileOffset, u32 *desiredSize) const = 0;
 	virtual int SetSecondBuffer(u32 secondBuffer, u32 secondBufferSize) = 0;
@@ -242,6 +244,7 @@ public:
 
 protected:
 	u16 outputChannels_ = 2;
+	int setDataSkippedFrames_ = 0;
 
 	// TODO: Save the internal state of this, now technically possible.
 	AudioDecoder *decoder_ = nullptr;

Core/HLE/AtracCtx2.cpp

@@ -1019,8 +1019,9 @@ int Atrac2::SetData(const Track &track, u32 bufferAddr, u32 readSize, u32 buffer
 		info.fileDataEnd, info.decodePos, info.numSkipFrames, info.numChan
 	);
 
-	int skipCount = 0;  // TODO: use for delay
+	int skipCount = 0;
 	retval = SkipFrames(&skipCount);
+	setDataSkippedFrames_ = skipCount;
 
 	// Seen in Mui Mui house. Things go very wrong after this..
 	if (retval == SCE_ERROR_ATRAC_API_FAIL) {

Core/HLE/sceAtrac.cpp

@@ -32,8 +32,11 @@
 
 #include "Core/HLE/sceKernel.h"
 #include "Core/HLE/sceUtility.h"
+#include "Core/HLE/sceVideocodec.h"
 #include "Core/HLE/sceKernelMemory.h"
+#include "Core/HLE/scePower.h"
 #include "Core/HLE/sceAtrac.h"
+#include "Core/HLE/sceAudiocodec.h"
 #include "Core/HLE/AtracCtx.h"
 #include "Core/HLE/AtracCtx2.h"
 #include "Core/System.h"
@@ -81,7 +84,21 @@
 
 // TODO: We should add checks that the utility module is loaded.
 
-static const int atracDecodeDelay = 2300;
+// The Media Engine does the decoding while the caller waits. Setting data decodes the frames before
+// the first sample (thrown away), which on hardware makes it cost a decoder setup plus a frame
+// decode: ~900us for mono Atrac3, ~3.5ms for stereo Atrac3+ (pspautotests threads/scheduling/callcosts).
+static int AtracFrameUs(const AtracBase *atrac) {
+	return AudioCodecDecodeUs(atrac->CodecType(), atrac->Channels(), atrac->BytesPerFrame());
+}
+
+static int AtracSetDataDelay(const AtracBase *atrac) {
+	const int us = AudioCodecInitUs(atrac->CodecType(), atrac->Channels() == 1) + atrac->SkippedFramesOnSetData() * AtracFrameUs(atrac);
+	return MEScheduleJob(PowerScaleFromDefaultClock(us));
+}
+
+static int AtracDecodeDelay(const AtracBase *atrac) {
+	return MEScheduleJob(PowerScaleFromDefaultClock(AtracFrameUs(atrac)));
+}
 
 static bool atracInited = true;
 static AtracBase *atracContexts[PSP_MAX_ATRAC_IDS];
@@ -94,7 +111,7 @@ static int g_atracBSS = 0;
 static bool g_muteFlag[PSP_MAX_ATRAC_IDS]{};  // Not saved, just for debugging.
 
 // On a PSP, the Media Engine does the decoding, and the samples land in the output buffer when
-// sceAtracDecodeData returns, about atracDecodeDelay later. A game can still be playing out of that
+// sceAtracDecodeData returns, a frame decode later. A game can still be playing out of that
 // memory in the meantime: Fired Up decodes into a buffer that overlaps the first 16 samples of the
 // one it has just handed to sceAudio, and relies on the mixer having read them first. So the
 // samples are written just before the thread wakes rather than when the call is made.
@@ -423,6 +440,7 @@ static u32 sceAtracDecodeData(int atracID, u32 outAddr, u32 numSamplesAddr, u32
 	}
 
 	if (ret == 0 || ret == SCE_ERROR_ATRAC_API_FAIL) {
+		const int delay = AtracDecodeDelay(atrac);
 		const u32 written = std::min((u32)(atrac->GetOutputChannels() * 2 * numSamplesWritten), (u32)previous.size());
 		if (outPtr && written != 0) {
 			AtracPendingOutput pending{ ++g_pendingOutputId, outAddr };
@@ -430,10 +448,10 @@ static u32 sceAtracDecodeData(int atracID, u32 outAddr, u32 numSamplesAddr, u32
 			memcpy(outPtr, previous.data(), previous.size());
 			g_pendingOutput.push_back(std::move(pending));
 			// Just ahead of the thread waking up.
-			CoreTiming::ScheduleEvent(usToCycles(atracDecodeDelay) - 1, g_atracOutputEvent, g_pendingOutputId);
+			CoreTiming::ScheduleEvent(usToCycles(delay) - 1, g_atracOutputEvent, g_pendingOutputId);
 		}
 		// Decoded or at least attempted to decode data, delay thread
-		return hleDelayResult(hleNoLog(ret), "atrac decode data", atracDecodeDelay);
+		return hleDelayResult(hleNoLog(ret), "atrac decode data", delay);
 	}
 
 	return hleNoLog(ret);
@@ -753,7 +771,7 @@ static u32 sceAtracSetHalfwayBuffer(int atracID, u32 buffer, u32 readSize, u32 b
 	}
 
 	// not sure the real delay time
-	return hleDelayResult(hleLogDebug(Log::Atrac, ret), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, ret), "atrac set data", AtracSetDataDelay(atrac));
 }
 
 static u32 sceAtracSetSecondBuffer(int atracID, u32 secondBuffer, u32 secondBufferSize) {
@@ -788,7 +806,7 @@ static u32 sceAtracSetData(int atracID, u32 buffer, u32 bufferSize) {
 		return hleLogError(Log::Atrac, ret);
 	}
 
-	return hleDelayResult(hleLogDebug(Log::Atrac, ret), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, ret), "atrac set data", AtracSetDataDelay(atrac));
 }
 
 static int sceAtracSetDataAndGetID(u32 buffer, int bufferSize) {
@@ -819,7 +837,7 @@ static int sceAtracSetDataAndGetID(u32 buffer, int bufferSize) {
 		return hleLogError(Log::Atrac, ret);
 	}
 
-	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 static int sceAtracSetHalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 bufferSize) {
@@ -845,7 +863,7 @@ static int sceAtracSetHalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 buffer
 		return hleLogError(Log::Atrac, ret);
 	}
 
-	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 static u32 sceAtracStartEntry() {
@@ -959,7 +977,7 @@ static int sceAtracSetMOutHalfwayBuffer(int atracID, u32 buffer, u32 readSize, u
 		// Must not delay.
 		return hleLogError(Log::Atrac, ret);
 	}
-	return hleDelayResult(hleLogDebugOrError(Log::Atrac, ret), "atrac set data mono", 100);
+	return hleDelayResult(hleLogDebugOrError(Log::Atrac, ret), "atrac set data mono", AtracSetDataDelay(atrac));
 }
 
 // Note: This doesn't seem to be part of any available libatrac3plus library.
@@ -985,7 +1003,7 @@ static u32 sceAtracSetMOutData(int atracID, u32 buffer, u32 bufferSize) {
 		return hleLogError(Log::Atrac, ret);
 	}
 	// It's OK if this fails, at least with NO_MONO...
-	return hleDelayResult(hleLogDebugOrError(Log::Atrac, ret), "atrac set data mono", 100);
+	return hleDelayResult(hleLogDebugOrError(Log::Atrac, ret), "atrac set data mono", AtracSetDataDelay(atrac));
 }
 
 // Note: This doesn't seem to be part of any available libatrac3plus library.
@@ -1012,7 +1030,7 @@ static int sceAtracSetMOutDataAndGetID(u32 buffer, u32 bufferSize) {
 		UnregisterAndDeleteAtrac(atracID);
 		return hleLogError(Log::Atrac, ret);
 	}
-	return hleDelayResult(hleLogDebugOrError(Log::Atrac, atracID), "atrac set data", 100);
+	return hleDelayResult(hleLogDebugOrError(Log::Atrac, atracID), "atrac set data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 static int sceAtracSetMOutHalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 bufferSize) {
@@ -1041,7 +1059,7 @@ static int sceAtracSetMOutHalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 bu
 		UnregisterAndDeleteAtrac(atracID);
 		return hleLogError(Log::Atrac, ret);
 	}
-	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 static int sceAtracSetAA3DataAndGetID(u32 buffer, u32 bufferSize, u32 fileSize, u32 metadataSizeAddr) {
@@ -1063,7 +1081,7 @@ static int sceAtracSetAA3DataAndGetID(u32 buffer, u32 bufferSize, u32 fileSize,
 		return hleLogError(Log::Atrac, ret);
 	}
 
-	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set aa3 data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set aa3 data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 static int sceAtracSetAA3HalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 bufferSize, u32 fileSize) {
@@ -1089,7 +1107,7 @@ static int sceAtracSetAA3HalfwayBufferAndGetID(u32 buffer, u32 readSize, u32 buf
 		return hleLogError(Log::Atrac, ret);
 	}
 
-	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", 100);
+	return hleDelayResult(hleLogDebug(Log::Atrac, atracID), "atrac set data", AtracSetDataDelay(atracContexts[atracID]));
 }
 
 // TODO: Should see if these are stored contiguously in memory somewhere, or if there really are
@@ -1155,7 +1173,7 @@ static int sceAtracLowLevelDecode(int atracID, u32 sourceAddr, u32 sourceBytesCo
 	}
 
 	NotifyMemInfo(MemBlockFlags::WRITE, samplesAddr, bytesWritten, "AtracLowLevelDecode");
-	return hleDelayResult(hleLogDebug(Log::Atrac, retval), "low level atrac decode data", atracDecodeDelay);
+	return hleDelayResult(hleLogDebug(Log::Atrac, retval), "low level atrac decode data", AtracDecodeDelay(atrac));
 }
 
 // These three are the external interface used by sceSas' AT3 integration.

Core/HLE/sceAudiocodec.cpp

@@ -239,15 +239,19 @@ static int MECall(int result, int us) {
 	return hleDelayResult(result, "audiocodec", MEScheduleJob(PowerScaleFromDefaultClock(us)));
 }
 
-static int InitUs(int codec, const SceAudiocodecCodec *ctx) {
+int AudioCodecInitUs(int codec, bool monoAt3Plus) {
 	switch (codec) {
-	case PSP_CODEC_AT3PLUS: return ((ctx->fmt.at3.formatByte1 >> 2) & 7) == 1 ? 524 : 646;
+	case PSP_CODEC_AT3PLUS: return monoAt3Plus ? 524 : 646;
 	case PSP_CODEC_AT3: return 210;
 	case PSP_CODEC_MP3: return 517;
 	default: return 230;
 	}
 }
 
+static int InitUs(int codec, const SceAudiocodecCodec *ctx) {
+	return AudioCodecInitUs(codec, codec == PSP_CODEC_AT3PLUS && ((ctx->fmt.at3.formatByte1 >> 2) & 7) == 1);
+}
+
 // libmp4.prx puts the sample rate here. On hardware 22050 and 44100 are accepted, 0 and 12345
 // aren't; that the rest of the standard AAC rates are accepted is an assumption. 0 if not valid.
 static int AacSampleRateFromContext(const SceAudiocodecCodec *ctx) {
@@ -395,6 +399,11 @@ static int EstimateDecodeUs(int codec, int channels, int frameBytes, const SceAu
 	}
 }
 
+int AudioCodecDecodeUs(int codec, int channels, int frameBytes) {
+	_dbg_assert_(codec == PSP_CODEC_AT3PLUS || codec == PSP_CODEC_AT3);
+	return EstimateDecodeUs(codec, channels, frameBytes, nullptr);
+}
+
 static int sceAudiocodecInit(u32 ctxPtr, int codec) {
 	return __AudioCodecInitCommon(ctxPtr, codec, false);
 }

Core/HLE/sceAudiocodec.h

@@ -128,3 +128,8 @@ class AudioDecoder;
 extern std::map<u32, AudioDecoder *> g_audioDecoderContexts;
 
 bool IsAtrac3StreamJointStereo(int codecType, int bytesPerFrame, int channels);
+
+// ME time at the default clock, for sceAtrac too (it drives the same decoder): setting up a decoder,
+// and decoding one frame. AudioCodecDecodeUs takes Atrac3 and Atrac3+ only.
+int AudioCodecInitUs(int codec, bool monoAt3Plus);
+int AudioCodecDecodeUs(int codec, int channels, int frameBytes);