From 346614b0da4dc74fe482624f2c0e903feda0bea1 Mon Sep 17 00:00:00 2001
From: Michael Yu <ItzWarty@gmail.com>
Date: Mon, 19 May 2014 02:07:45 -0700
Subject: [PATCH 1/4] gitignore now ignores files generated by visual studio
 profiler, solution configured for profiling.

---
 .gitignore | 4 ++++
 rpcs3.sln  | 3 +++
 2 files changed, 7 insertions(+)

diff --git a/.gitignore b/.gitignore
index 2e4c753253..46313b81af 100644
--- a/.gitignore
+++ b/.gitignore
@@ -45,6 +45,10 @@
 /bin/*.exp
 rpcs3/git-version.h
 
+# Visual Studio Profiler Files
+*.vspx
+*.psess
+
 # Copyrighted files
 /bin/data/
 /bin/dev_flash/data/font
diff --git a/rpcs3.sln b/rpcs3.sln
index 7b0d30d2e0..d06a3dabd7 100644
--- a/rpcs3.sln
+++ b/rpcs3.sln
@@ -457,4 +457,7 @@ Global
 		{23E1C437-A951-5943-8639-A17F3CF2E606} = {5812E712-6213-4372-B095-9EB9BAA1F2DF}
 		{74827EBD-93DC-5110-BA95-3F2AB029B6B0} = {5812E712-6213-4372-B095-9EB9BAA1F2DF}
 	EndGlobalSection
+	GlobalSection(Performance) = preSolution
+		HasPerformanceSessions = true
+	EndGlobalSection
 EndGlobal

From 2834697f1f8dbad36aa7a932a5a2b9e60ecd3470 Mon Sep 17 00:00:00 2001
From: Michael Yu <ItzWarty@gmail.com>
Date: Mon, 19 May 2014 02:09:54 -0700
Subject: [PATCH 2/4] MEMORY: Copy from real's destination parameter should be
 immutable

---
 rpcs3/Emu/Memory/Memory.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/rpcs3/Emu/Memory/Memory.h b/rpcs3/Emu/Memory/Memory.h
index d1dcef50e2..61d7c872ff 100644
--- a/rpcs3/Emu/Memory/Memory.h
+++ b/rpcs3/Emu/Memory/Memory.h
@@ -333,11 +333,11 @@ public:
 		return true;
 	}
 
-	bool CopyFromReal(u32 to, void* real, u32 count) // (4K pages) copy from real to virtual memory
+	bool CopyFromReal(u32 to, const void* real, u32 count) // (4K pages) copy from real to virtual memory
 	{
 		if (!count) return true;
 
-		u8* from = (u8*)real;
+		const u8* from = (const u8*)real;
 
 		if (u32 frag = to & 4095)
 		{

From 9bdb12e3da1f7f3f0607c6ff2673a54f1ed5dc72 Mon Sep 17 00:00:00 2001
From: Michael Yu <ItzWarty@gmail.com>
Date: Mon, 19 May 2014 02:10:33 -0700
Subject: [PATCH 3/4] MEMORY: Added AppendRawBytes, which also advances the
 memory pointer unlike Memory.CopyFromReal

---
 rpcs3/Emu/Memory/Memory.h | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/rpcs3/Emu/Memory/Memory.h b/rpcs3/Emu/Memory/Memory.h
index 61d7c872ff..e5d56a111c 100644
--- a/rpcs3/Emu/Memory/Memory.h
+++ b/rpcs3/Emu/Memory/Memory.h
@@ -795,6 +795,12 @@ public:
 		return this->m_addr;
 	}
 
+	u32 AppendRawBytes(const u8 * bytes, size_t count) {
+		Memory.CopyFromReal(this->m_addr, bytes, count);
+		this->m_addr += count;
+		return this->m_addr;
+	}
+
 	u32 Skip(const u32 offset) { return this->m_addr += offset; }
 
 	operator be_t<T>*()			{ return GetPtr(); }

From 3aeb0b0f9552cbae1856f396189f1dff7d5ca8c6 Mon Sep 17 00:00:00 2001
From: Michael Yu <ItzWarty@gmail.com>
Date: Mon, 19 May 2014 02:14:07 -0700
Subject: [PATCH 4/4] cellPngDecDecodeData handles CELL_PNGDEC_ARGB case much
 faster.

Profiling done with two samples on Solar v2.1 from rpcs3 init to first frame.
Before optimization, profiler found rpcs3 in cellPngDecDecodeData 15.3% of the time.
Post-optimization, profiler finds rpcs3 in cellPngDecDecodeData 0.33% of the time for ~50x improvement.
---
 rpcs3/Emu/SysCalls/Modules/cellPngDec.cpp | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/rpcs3/Emu/SysCalls/Modules/cellPngDec.cpp b/rpcs3/Emu/SysCalls/Modules/cellPngDec.cpp
index d169321c51..31886ab113 100644
--- a/rpcs3/Emu/SysCalls/Modules/cellPngDec.cpp
+++ b/rpcs3/Emu/SysCalls/Modules/cellPngDec.cpp
@@ -201,7 +201,7 @@ int cellPngDecDecodeData(u32 mainHandle, u32 subHandle, mem8_ptr_t data, const m
 
 	case CELL_PNGDEC_ARGB:
 	{
-		const char nComponents = 4;
+		const int nComponents = 4;
 		image_size *= nComponents;
 		if (bytesPerLine > width * nComponents || flip) //check if we need padding
 		{
@@ -225,13 +225,19 @@ int cellPngDecDecodeData(u32 mainHandle, u32 subHandle, mem8_ptr_t data, const m
 		}
 		else
 		{
-			for (uint i = 0; i < image_size; i += nComponents)
+			uint* dest = (uint*)new char[image_size];
+			uint* source_current = (uint*)&(image.get()[0]);
+			uint* dest_current = dest;
+			for (uint i = 0; i < image_size / nComponents; i++) 
 			{
-				data += image.get()[i + 3];
-				data += image.get()[i + 0];
-				data += image.get()[i + 1];
-				data += image.get()[i + 2];
+				uint val = *source_current;
+				*dest_current = (val >> 24) | (val << 8); // set alpha (A8) as leftmost byte
+				source_current++;
+				dest_current++;
 			}
+			// NOTE: AppendRawBytes has diff side-effect vs Memory.CopyFromReal
+			data.AppendRawBytes((u8*)dest, image_size); 
+			delete[] dest;
 		}
 	}
 	break;