Added in altivec code, but it does not work, so it is disabled for now

Libraries/MAC/MAC.xcodeproj/project.pbxproj

@@ -349,27 +349,10 @@
 				8CA0B48909CAD9E300B4B731 /* UnBitArray.h */,
 				8CA0B48A09CAD9E300B4B731 /* UnBitArrayBase.h */,
 				8CA0B48B09CAD9E300B4B731 /* WAVInputSource.h */,
-				8CA0B39709CAD77400B4B731 /* Assembly */,
 			);
 			name = "Library Headers";
 			sourceTree = "<group>";
 		};
-		8CA0B39709CAD77400B4B731 /* Assembly */ = {
-			isa = PBXGroup;
-			children = (
-				8CA0B39809CAD77400B4B731 /* .deps */,
-			);
-			name = Assembly;
-			path = src/MACLib/Assembly;
-			sourceTree = "<group>";
-		};
-		8CA0B39809CAD77400B4B731 /* .deps */ = {
-			isa = PBXGroup;
-			children = (
-			);
-			path = .deps;
-			sourceTree = "<group>";
-		};
 /* End PBXGroup section */
 
 /* Begin PBXHeadersBuildPhase section */
@@ -566,10 +549,13 @@
 				FRAMEWORK_VERSION = A;
 				GCC_GENERATE_DEBUGGING_SYMBOLS = NO;
 				GCC_MODEL_TUNING = G5;
+				GCC_OPTIMIZATION_LEVEL = 3;
 				GCC_PRECOMPILE_PREFIX_HEADER = YES;
 				GCC_PREFIX_HEADER = MAC_Prefix.pch;
 				INFOPLIST_FILE = Info.plist;
 				INSTALL_PATH = "@executable_path/../Frameworks";
+				OTHER_LDFLAGS = "-Wl,-read_only_relocs,warning";
+				PER_ARCH_CFLAGS_ppc = "-maltivec";
 				PRODUCT_NAME = MAC;
 				USER_HEADER_SEARCH_PATHS = "mac-src";
 				WRAPPER_EXTENSION = framework;
@@ -591,6 +577,7 @@
 			buildSettings = {
 				GCC_WARN_ABOUT_RETURN_TYPE = YES;
 				GCC_WARN_UNUSED_VARIABLE = YES;
+				PER_ARCH_CFLAGS_ppc = "-maltivec";
 				PREBINDING = NO;
 				SDKROOT = /Developer/SDKs/MacOSX10.4u.sdk;
 			};

Libraries/MAC/mac-src/src/MACLib/NNFilter.cpp

@@ -3,6 +3,10 @@
 #include "NNFilter.h"
 #include "Assembly/Assembly.h"
 
+#ifdef __ppc__
+#include <altivec.h>
+#endif
+
 CNNFilter::CNNFilter(int nOrder, int nShift, int nVersion)
 {
     if ((nOrder <= 0) || ((nOrder % 16) != 0)) throw(1);
@@ -11,6 +15,7 @@ CNNFilter::CNNFilter(int nOrder, int nShift, int nVersion)
     m_nVersion = nVersion;
     
     m_bMMXAvailable = GetMMXAvailable();
+    m_AltiVecAvailable = IsAltiVecAvailable();
 	
     m_rbInput.Create(NN_WINDOW_ELEMENTS, m_nOrder);
     m_rbDeltaM.Create(NN_WINDOW_ELEMENTS, m_nOrder);
@@ -41,7 +46,14 @@ int CNNFilter::Compress(int nInput)
 
     // figure a dot product
     int nDotProduct;
-    if (m_bMMXAvailable)
+	if(m_AltiVecAvailable)
+	{
+		nDotProduct = CalculateDotProductAltiVec(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
+//		printf("Dot product altivec: %i\n", nDotProduct);
+		nDotProduct = CalculateDotProductNoMMX(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
+//		printf("Dot product: %i\n", nDotProduct);
+	}
+    else if (m_bMMXAvailable)
         nDotProduct = CalculateDotProduct(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
     else
         nDotProduct = CalculateDotProductNoMMX(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
@@ -50,7 +62,10 @@ int CNNFilter::Compress(int nInput)
     int nOutput = nInput - ((nDotProduct + (1 << (m_nShift - 1))) >> m_nShift);
 
     // adapt
-    if (m_bMMXAvailable)
+	if(m_AltiVecAvailable)
+//		AdaptAltiVec(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nOutput, m_nOrder);
+		AdaptNoMMX(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nOutput, m_nOrder);
+    else if (m_bMMXAvailable)
         Adapt(&m_paryM[0], &m_rbDeltaM[-m_nOrder], -nOutput, m_nOrder);
     else
         AdaptNoMMX(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nOutput, m_nOrder);
@@ -84,13 +99,23 @@ int CNNFilter::Decompress(int nInput)
     // figure a dot product
     int nDotProduct;
 
-    if (m_bMMXAvailable)
+	if(m_AltiVecAvailable)
+	{
+//		nDotProduct = CalculateDotProductAltiVec(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
+//		printf("Dot product altivec: %i\n", nDotProduct);
+		nDotProduct = CalculateDotProductNoMMX(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
+//		printf("Dot product: %i\n", nDotProduct);
+	}
+    else if (m_bMMXAvailable)
         nDotProduct = CalculateDotProduct(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
     else
         nDotProduct = CalculateDotProductNoMMX(&m_rbInput[-m_nOrder], &m_paryM[0], m_nOrder);
     
     // adapt
-    if (m_bMMXAvailable)
+	if(m_AltiVecAvailable)
+		AdaptAltiVec(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nInput, m_nOrder);
+//		AdaptNoMMX(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nInput, m_nOrder);
+	else if (m_bMMXAvailable)
         Adapt(&m_paryM[0], &m_rbDeltaM[-m_nOrder], -nInput, m_nOrder);
     else
         AdaptNoMMX(&m_paryM[0], &m_rbDeltaM[-m_nOrder], nInput, m_nOrder);
@@ -134,6 +159,159 @@ int CNNFilter::Decompress(int nInput)
     return nOutput;
 }
 
+#ifdef __ppc__
+void CNNFilter::AdaptAltiVec(short * pM, short * pAdapt, int nDirection, int nOrder)
+{
+	vector signed short LSQ2, LSQ4, v1, v2;
+	vector unsigned char mask2;
+	
+	nOrder >>= 4;
+	
+	//mask1 = vec_lvsl(0,pM);
+	mask2 = vec_lvsl(0,pAdapt);
+	//align = vec_lvsr(0,pM);
+	//zero = (vector unsigned char)(0);
+	//(vector signed char) one = (vector signed char)(-1);
+	//mask3 = vec_perm((vector unsigned char)(0),(vector unsigned char)(-1),align);
+	
+	//LSQ3 = vec_ld(0,pM);
+	LSQ4 = vec_ld(0,pAdapt);
+	
+	if (nDirection < 0) 
+	{
+		while (nOrder--)
+		{
+			
+			v1 = vec_ld(0,pM);
+			LSQ2 = vec_ld(16,pAdapt);
+			v2 = vec_perm(LSQ4,LSQ2,mask2);
+			v1 = vec_add(v1,v2);
+			vec_st(v1,0,pM);
+			
+			/*
+				v1 = vec_perm(v1,v1,align);
+			 LSQ3 = vec_sel(LSQ3,v1,(vector unsigned short)mask3);
+			 vec_st(LSQ3,0,pM);
+			 LSQ4 = vec_sel(v1,LSQ,(vector unsigned short)mask3);
+			 vec_st(LSQ4,16,pM);
+			 */
+			
+			v1 = vec_ld(16,pM);
+			LSQ4 = vec_ld(32,pAdapt);
+			v2 = vec_perm(LSQ2,LSQ4,mask2);
+			v1 = vec_add(v1,v2);
+			vec_st(v1,16,pM);
+			
+			/*
+				v1 = vec_perm(v1,v1,align);
+			 LSQ = vec_sel(LSQ,v1,(vector unsigned short)mask3);
+			 vec_st(LSQ,16,pM);
+			 LSQ2 = vec_sel(v1,LSQ3,(vector unsigned short)mask3);
+			 vec_st(LSQ2,32,pM);
+			 */
+			
+			//memcpy(pM,buffer,32);
+			pM = pM + 16;
+			pAdapt = pAdapt + 16;
+		}
+	}
+	else if (nDirection > 0)
+	{
+		while (nOrder--)
+		{
+			
+			v1 = vec_ld(0,pM);
+			LSQ2 = vec_ld(16,pAdapt);
+			v2 = vec_perm(LSQ4,LSQ2,mask2);
+			v1 = vec_sub(v1,v2);
+			vec_st(v1,0,pM);
+			
+			/*
+				v1 = vec_perm(v1,v1,align);
+			 LSQ3 = vec_sel(LSQ3,v1,(vector unsigned short)mask3);
+			 vec_st(LSQ3,0,pM);
+			 LSQ4 = vec_sel(v1,LSQ,(vector unsigned short)mask3);
+			 vec_st(LSQ4,16,pM);
+			 */
+			
+			v1 = vec_ld(16,pM);
+			LSQ4 = vec_ld(32,pAdapt);
+			v2 = vec_perm(LSQ2,LSQ4,mask2);
+			v1 = vec_sub(v1,v2);
+			vec_st(v1,16,pM);
+			
+			/*
+				v1 = vec_perm(v1,v1,align);
+			 LSQ = vec_sel(LSQ,v1,(vector unsigned short)mask3);
+			 vec_st(LSQ,16,pM);
+			 LSQ2 = vec_sel(v1,LSQ3,(vector unsigned short)mask3);
+			 vec_st(LSQ2,32,pM);
+			 */
+			
+			//memcpy(pM,buffer,32);
+			pM = pM + 16;
+			pAdapt = pAdapt + 16;
+			
+			
+		}
+	}
+}
+
+int CNNFilter::CalculateDotProductAltiVec(short * pA, short * pB, int nOrder)
+{
+	vector signed short LSQ, LSQ3, v1, v2;
+	vector unsigned char mask1;
+	
+	vector signed int vzero = (vector signed int)(0);
+	vector signed int sum = (vector signed int)(0);
+//	sum = vec_xor(sum,sum);
+	
+	//int nDotProduct;
+	int p[4];
+	nOrder >>= 4;
+	
+	mask1 = vec_lvsl(0,pA);
+	//mask2 = vec_lvsl(0,pB);
+	
+	
+	LSQ3 = vec_ld(0, pA);
+	//LSQ4 = vec_ld(0, pB);
+	
+	while (nOrder--)
+	{
+		
+		LSQ = vec_ld(16,pA);
+		v1 = vec_perm(LSQ3,LSQ,mask1);
+		v2 = vec_ld(0,pB);
+		sum = vec_msum(v1,v2,sum);
+		
+		LSQ3 = vec_ld(32,pA);
+		v1 = vec_perm(LSQ,LSQ3,mask1);
+		v2 = vec_ld(16,pB);
+		sum = vec_msum(v1,v2,sum);
+		
+		pA = pA + 16;
+		pB = pB + 16;
+		
+	}
+	
+	sum = vec_sums(sum,vzero);
+	vec_st(sum,0,p); 
+	
+	return p[3];
+}
+
+#else
+void CNNFilter::AdaptAltiVec(short * pM, short * pAdapt, int nDirection, int nOrder)
+{
+	AdaptNoMMX(pM, pAdapt, nDirection, nOrder);
+}
+int CNNFilter::CalculateDotProductAltiVec(short * pA, short * pB, int nOrder)
+{
+	return CalculateDotProductNoMMX(pA, pB, nOrder);
+}
+#endif
+
 void CNNFilter::AdaptNoMMX(short * pM, short * pAdapt, int nDirection, int nOrder)
 {
     nOrder >>= 4;

Libraries/MAC/mac-src/src/MACLib/NNFilter.h

@@ -22,6 +22,7 @@ private:
     int m_nShift;
     int m_nVersion;
     BOOL m_bMMXAvailable;
+	int m_AltiVecAvailable;
     int m_nRunningAverage;
 
     CRollBuffer<short> m_rbInput;
@@ -36,6 +37,9 @@ private:
 
     inline int CalculateDotProductNoMMX(short * pA, short * pB, int nOrder);
     inline void AdaptNoMMX(short * pM, short * pAdapt, int nDirection, int nOrder);
+	int CalculateDotProductAltiVec(short * pA, short * pB, int nOrder);
+	void AdaptAltiVec(short * pM, short * pAdapt, int nDirection, int nOrder);
+
 };
 
 #endif // #ifndef APE_NNFILTER_H

Libraries/MAC/mac-src/src/Shared/GlobalFunctions.cpp

@@ -2,6 +2,7 @@
 #include "GlobalFunctions.h"
 #include "IO.h"
 #include "CharacterHelper.h"
+#include <sys/sysctl.h>
 
 /*
 #ifndef __GNUC_IA32__
@@ -38,6 +39,19 @@ extern "C" BOOL GetMMXAvailable(void)
 #endif // #ifndef __GNUC_IA32__
 */
 
+int IsAltiVecAvailable( void )
+{
+	
+	int selectors[2] = { CTL_HW, HW_VECTORUNIT };
+	int hasVectorUnit = 0;
+	size_t length = sizeof(hasVectorUnit);
+	int error = sysctl(selectors, 2, &hasVectorUnit, &length, NULL, 0);
+	printf("ALTIVEC: %i\n", hasVectorUnit);
+	if( 0 == error ) return hasVectorUnit != 0;
+	
+	return 0;
+}
+
 int ReadSafe(CIO * pIO, void * pBuffer, int nBytes)
 {
     unsigned int nBytesRead = 0;

Libraries/MAC/mac-src/src/Shared/GlobalFunctions.h

@@ -6,6 +6,8 @@ Definitions
 *************************************************************************************/
 class CIO;
 
+int IsAltiVecAvailable( void );
+
 /*************************************************************************************
 Read / Write from an IO source and return failure if the number of bytes specified
 isn't read or written