From 0de71f27cbcd1a889b4458f6b3adedb0c640f7fd Mon Sep 17 00:00:00 2001
From: Henrik Rydgard <hrydgard@gmail.com>
Date: Sat, 4 Jan 2014 01:57:22 +0100
Subject: [PATCH] Optimize the VAGADPCM decoder.

---
 Core/HLE/sceSas.cpp  |  8 ++++----
 Core/HW/SasAudio.cpp | 35 ++++++++++++++++++++++-------------
 Core/HW/SasAudio.h   | 16 +++++-----------
 3 files changed, 31 insertions(+), 28 deletions(-)

diff --git a/Core/HLE/sceSas.cpp b/Core/HLE/sceSas.cpp
index 3bc169e0a8..481f70a4fd 100644
--- a/Core/HLE/sceSas.cpp
+++ b/Core/HLE/sceSas.cpp
@@ -394,15 +394,15 @@ u32 sceSasSetSimpleADSR(u32 core, int voiceNum, u32 ADSREnv1, u32 ADSREnv2) {
 }
 
 u32 sceSasGetEnvelopeHeight(u32 core, int voiceNum) {
-	DEBUG_LOG(SCESAS, "sceSasGetEnvelopeHeight(%08x, %i)", core, voiceNum);
-	
 	if (voiceNum >= PSP_SAS_VOICES_MAX || voiceNum < 0)	{
-		WARN_LOG(SCESAS, "%s: invalid voicenum %d", __FUNCTION__, voiceNum);
+		ERROR_LOG(SCESAS, "%s: invalid voicenum %d", __FUNCTION__, voiceNum);
 		return ERROR_SAS_INVALID_VOICE;
 	}
 
 	SasVoice &v = sas->voices[voiceNum];
-	return v.envelope.GetHeight();
+	int height = v.envelope.GetHeight();
+	DEBUG_LOG(SCESAS, "%i = sceSasGetEnvelopeHeight(%08x, %i)", height, core, voiceNum);
+	return height;
 }
 
 u32 sceSasRevType(u32 core, int type) {
diff --git a/Core/HW/SasAudio.cpp b/Core/HW/SasAudio.cpp
index db12e1dc3f..1a577db1f8 100644
--- a/Core/HW/SasAudio.cpp
+++ b/Core/HW/SasAudio.cpp
@@ -61,7 +61,8 @@ void VagDecoder::Start(u32 data, int vagSize, bool loopEnabled) {
 	s_2 = 0;
 }
 
-void VagDecoder::DecodeBlock(u8 *&readp) {
+void VagDecoder::DecodeBlock(u8 *&read_pointer) {
+	u8 *readp = read_pointer;
 	int predict_nr = *readp++;
 	int shift_factor = predict_nr & 0xf;
 	predict_nr >>= 4;
@@ -79,24 +80,33 @@ void VagDecoder::DecodeBlock(u8 *&readp) {
 			loopAtNextBlock_ = true;
 		}
 	}
+
+	// Keep state in locals to avoid bouncing to memory.
+	int s1 = s_1;
+	int s2 = s_2;
+
+	int coef1 = f[predict_nr][0];
+	int coef2 = f[predict_nr][1];
+
 	for (int i = 0; i < 28; i += 2) {
-		int d = *readp++;
-		int s = (short)((d & 0xf) << 12);
-		DecodeSample(i, s >> shift_factor, predict_nr);
-		s = (short)((d & 0xf0) << 8);
-		DecodeSample(i + 1, s >> shift_factor, predict_nr);
+		u8 d = *readp++;
+		int sample1 = (short)((d & 0xf) << 12) >> shift_factor;
+		int sample2 = (short)((d & 0xf0) << 8) >> shift_factor;
+		s2 = (int)(sample1 + ((s1 * coef1 + s2 * coef2) >> 6));
+		s1 = (int)(sample2 + ((s2 * coef1 + s1 * coef2) >> 6));
+		samples[i] = s2;
+		samples[i + 1] = s1;
 	}
+
+	s_1 = s1;
+	s_2 = s2;
 	curSample = 0;
 	curBlock_++;
 	if (curBlock_ == numBlocks_) {
 		end_ = true;
 	}
-}
 
-inline void VagDecoder::DecodeSample(int i, int sample, int predict_nr) {
-	samples[i] = (int) (sample + ((s_1 * f[predict_nr][0] + s_2 * f[predict_nr][1]) >> 6));
-	s_2 = s_1;
-	s_1 = samples[i];
+	read_pointer = readp;
 }
 
 void VagDecoder::GetSamples(s16 *outSamples, int numSamples) {
@@ -137,8 +147,7 @@ void VagDecoder::GetSamples(s16 *outSamples, int numSamples) {
 	}
 }
 
-void VagDecoder::DoState(PointerWrap &p)
-{
+void VagDecoder::DoState(PointerWrap &p) {
 	auto s = p.Section("VagDecoder", 1);
 	if (!s)
 		return;
diff --git a/Core/HW/SasAudio.h b/Core/HW/SasAudio.h
index 35739aef6b..789223dca6 100644
--- a/Core/HW/SasAudio.h
+++ b/Core/HW/SasAudio.h
@@ -62,8 +62,7 @@ enum {
 	PSP_SAS_EFFECT_TYPE_PIPE = 8,
 };
 
-struct WaveformEffect
-{
+struct WaveformEffect {
 	int type;
 	int delay;
 	int feedback;
@@ -98,7 +97,6 @@ public:
 	void DoState(PointerWrap &p);
 
 private:
-	void DecodeSample(int i, int sample, int predict_nr);
 	int samples[28];
 	int curSample;
 
@@ -132,9 +130,7 @@ private:
 	BufferQueue *sampleQueue;
 };
 
-// Max height: 0x40000000 I think
-class ADSREnvelope
-{
+class ADSREnvelope {
 public:
 	ADSREnvelope();
 	void SetSimpleEnvelope(u32 ADSREnv1, u32 ADSREnv2);
@@ -184,14 +180,13 @@ private:
 
 	ADSRState state_;
 	int steps_;
-	s64 height_;  // s64 to avoid having to care about overflow when calculatimg. TODO: this should be fine as s32
+	s64 height_;  // s64 to avoid having to care about overflow when calculating. TODO: this should be fine as s32
 };
 
 // A SAS voice.
 // TODO: Look into pre-decoding the VAG samples on SetVoice instead of decoding them on the fly.
 // It's not very likely that games encode VAG dynamically.
-struct SasVoice
-{
+struct SasVoice {
 	SasVoice()
 		: playing(false),
 			paused(false),
@@ -262,8 +257,7 @@ struct SasVoice
 	SasAtrac3 atrac3;
 };
 
-class SasInstance
-{
+class SasInstance {
 public:
 	SasInstance();
 	~SasInstance();