From 0ff4444118eadf8a16f9b4619074602a2fac6d92 Mon Sep 17 00:00:00 2001
From: ejcoumans <ejcoumans@08e121b0-ae19-0410-a57b-3be3395fd4fd>
Date: Wed, 26 Sep 2007 23:35:47 +0000
Subject: [PATCH] Added libspe2 parallel Cell SPE support contribution by IBM
 Germany 'Extreme Blue' project: Thanks to Benjamin Hoeferlin, Minh Cuong
 Tran,Martina Huellmann,Frederick Roth.

---
 Extras/BulletMultiThreaded/SpuDoubleBuffer.h | 213 ++++++++++---------
 Extras/BulletMultiThreaded/SpuFakeDma.cpp    | 212 +++++++++---------
 Extras/BulletMultiThreaded/SpuFakeDma.h      | 157 ++++++++++----
 3 files changed, 334 insertions(+), 248 deletions(-)

diff --git a/Extras/BulletMultiThreaded/SpuDoubleBuffer.h b/Extras/BulletMultiThreaded/SpuDoubleBuffer.h
index c533a68a3..55740db4b 100644
--- a/Extras/BulletMultiThreaded/SpuDoubleBuffer.h
+++ b/Extras/BulletMultiThreaded/SpuDoubleBuffer.h
@@ -1,106 +1,107 @@
-#ifndef DOUBLE_BUFFER_H
-#define DOUBLE_BUFFER_H
-
-#include "SpuFakeDma.h"
-
-
-///DoubleBuffer
-template<class T, int size>
-class DoubleBuffer
-{
-#ifdef __CELLOS_LV2__
-	T m_buffer0[size] __attribute__ ((aligned (128)));
-	T m_buffer1[size] __attribute__ ((aligned (128)));
-#else
-	T m_buffer0[size];
-	T m_buffer1[size];
-#endif
-	
-	T *m_frontBuffer;
-	T *m_backBuffer;
-
-	unsigned int m_dmaTag;
-	bool m_dmaPending;
-public:
-	bool	isPending() const { return m_dmaPending;}
-	DoubleBuffer();
-
-	void init ();
-
-	// dma get and put commands
-	void backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag);
-	void backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag);
-
-	// gets pointer to a buffer
-	T *getFront();
-	T *getBack();
-
-	// if back buffer dma was started, wait for it to complete
-	// then move back to front and vice versa
-	T *swapBuffers();
-};
-
-template<class T, int size>
-DoubleBuffer<T,size>::DoubleBuffer()
-{
-	init ();
-}
-
-template<class T, int size>
-void DoubleBuffer<T,size>::init()
-{
-	this->m_dmaPending = false;
-	this->m_frontBuffer = &this->m_buffer0[0];
-	this->m_backBuffer = &this->m_buffer1[0];
-}
-
-template<class T, int size>
-void
-DoubleBuffer<T,size>::backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag)
-{
-	m_dmaPending = true;
-	m_dmaTag = tag;
-	cellDmaLargeGet(m_backBuffer, ea, numBytes, tag, 0, 0);
-}
-
-template<class T, int size>
-void
-DoubleBuffer<T,size>::backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag)
-{
-	m_dmaPending = true;
-	m_dmaTag = tag;
-	cellDmaLargePut(m_backBuffer, ea, numBytes, tag, 0, 0);
-}
-
-template<class T, int size>
-T *
-DoubleBuffer<T,size>::getFront()
-{
-	return m_frontBuffer;
-}
-
-template<class T, int size>
-T *
-DoubleBuffer<T,size>::getBack()
-{
-	return m_backBuffer;
-}
-
-template<class T, int size>
-T *
-DoubleBuffer<T,size>::swapBuffers()
-{
-	if (m_dmaPending)
-	{
-		cellDmaWaitTagStatusAll(1<<m_dmaTag);
-		m_dmaPending = false;
-	}
-
-	T *tmp = m_backBuffer;
-	m_backBuffer = m_frontBuffer;
-	m_frontBuffer = tmp;
-
-	return m_frontBuffer;
-}
-
-#endif
+#ifndef DOUBLE_BUFFER_H
+#define DOUBLE_BUFFER_H
+
+#include "SpuFakeDma.h"
+#include <LinearMath/btScalar.h>
+
+
+///DoubleBuffer
+template<class T, int size>
+class DoubleBuffer
+{
+#if defined(__CELLOS_LV2__) || defined(USE_LIBSPE2)
+	ATTRIBUTE_ALIGNED128( T m_buffer0[size] ) ;
+	ATTRIBUTE_ALIGNED128( T m_buffer1[size] ) ;
+#else
+	T m_buffer0[size];
+	T m_buffer1[size];
+#endif
+	
+	T *m_frontBuffer;
+	T *m_backBuffer;
+
+	unsigned int m_dmaTag;
+	bool m_dmaPending;
+public:
+	bool	isPending() const { return m_dmaPending;}
+	DoubleBuffer();
+
+	void init ();
+
+	// dma get and put commands
+	void backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag);
+	void backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag);
+
+	// gets pointer to a buffer
+	T *getFront();
+	T *getBack();
+
+	// if back buffer dma was started, wait for it to complete
+	// then move back to front and vice versa
+	T *swapBuffers();
+};
+
+template<class T, int size>
+DoubleBuffer<T,size>::DoubleBuffer()
+{
+	init ();
+}
+
+template<class T, int size>
+void DoubleBuffer<T,size>::init()
+{
+	this->m_dmaPending = false;
+	this->m_frontBuffer = &this->m_buffer0[0];
+	this->m_backBuffer = &this->m_buffer1[0];
+}
+
+template<class T, int size>
+void
+DoubleBuffer<T,size>::backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag)
+{
+	m_dmaPending = true;
+	m_dmaTag = tag;
+	cellDmaLargeGet(m_backBuffer, ea, numBytes, tag, 0, 0);
+}
+
+template<class T, int size>
+void
+DoubleBuffer<T,size>::backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag)
+{
+	m_dmaPending = true;
+	m_dmaTag = tag;
+	cellDmaLargePut(m_backBuffer, ea, numBytes, tag, 0, 0);
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::getFront()
+{
+	return m_frontBuffer;
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::getBack()
+{
+	return m_backBuffer;
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::swapBuffers()
+{
+	if (m_dmaPending)
+	{
+		cellDmaWaitTagStatusAll(1<<m_dmaTag);
+		m_dmaPending = false;
+	}
+
+	T *tmp = m_backBuffer;
+	m_backBuffer = m_frontBuffer;
+	m_frontBuffer = tmp;
+
+	return m_frontBuffer;
+}
+
+#endif
diff --git a/Extras/BulletMultiThreaded/SpuFakeDma.cpp b/Extras/BulletMultiThreaded/SpuFakeDma.cpp
index 871888426..694617a6d 100644
--- a/Extras/BulletMultiThreaded/SpuFakeDma.cpp
+++ b/Extras/BulletMultiThreaded/SpuFakeDma.cpp
@@ -1,106 +1,106 @@
-
-#include "SpuFakeDma.h"
-#include <LinearMath/btScalar.h> //for btAssert
-//Disabling memcpy sometimes helps debugging DMA
-
-#define USE_MEMCPY 1
-
-
-///this unalignedDma should not be frequently used, only for small data. It handles alignment and performs check on size (<16 bytes)
-int	stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size)
-{
-	btAssert(size<16);
-	ATTRIBUTE_ALIGNED16(char	tmpBuffer[32]);
-
-	char* mainMem = (char*)ea;
-	char* localStore = (char*)ls;
-	uint32_t i;
-	
-
-	///make sure last 4 bits are the same, for cellDmaSmallGet
-	uint32_t last4BitsOffset = ea & 0x0f;
-	char* tmpTarget = tmpBuffer + last4BitsOffset;
-#ifdef WIN32
-
-#ifdef USE_MEMCPY
-	memcpy(tmpTarget,mainMem,size);
-#else
-	for ( i=0;i<size;i++)
-	{
-		tmpTarget[i] = mainMem[i];
-	}
-#endif
-#else
-	cellDmaSmallGet(tmpTarget,ea,size,DMA_TAG(1),0,0);
-	//copy into final destination
-#endif
-
-	cellDmaWaitTagStatusAll(DMA_MASK(1));
-
-	//this is slowish, perhaps memcpy on SPU is smarter?
-	for (i=0;i<size;i++)
-	{
-		localStore[i] = tmpTarget[i];
-	}
-
-	return 0;
-}
-
-
-#ifdef WIN32
-
-int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
-{
-	char* mainMem = (char*)ea;
-	char* localStore = (char*)ls;
-
-#ifdef USE_MEMCPY
-	memcpy(localStore,mainMem,size);
-#else
-	for (uint32_t i=0;i<size;i++)
-	{
-		localStore[i] = mainMem[i];
-	}
-#endif
-	return 0;
-}
-
-int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
-{
-	char* mainMem = (char*)ea;
-	char* localStore = (char*)ls;
-#ifdef USE_MEMCPY
-	memcpy(localStore,mainMem,size);
-#else
-	for (uint32_t i=0;i<size;i++)
-	{
-		localStore[i] = mainMem[i];
-	}	
-#endif //#ifdef USE_MEMCPY
-	return 0;
-}
-
-int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
-{
-	char* mainMem = (char*)ea;
-	const char* localStore = (const char*)ls;
-#ifdef USE_MEMCPY
-	memcpy(mainMem,localStore,size);
-#else
-	for (uint32_t i=0;i<size;i++)
-	{
-		mainMem[i] = localStore[i];
-	}	
-#endif //#ifdef USE_MEMCPY
-
-	return 0;
-}
-
-
-
-void	cellDmaWaitTagStatusAll(int ignore)
-{
-
-}
-
-#endif //WIN32
+
+#include "SpuFakeDma.h"
+#include <LinearMath/btScalar.h> //for btAssert
+//Disabling memcpy sometimes helps debugging DMA
+
+#define USE_MEMCPY 1
+
+
+///this unalignedDma should not be frequently used, only for small data. It handles alignment and performs check on size (<16 bytes)
+int stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size)
+{
+	btAssert(size<16);
+	ATTRIBUTE_ALIGNED16(char	tmpBuffer[32]);
+
+	char* mainMem = (char*)ea;
+	char* localStore = (char*)ls;
+	uint32_t i;
+	
+
+	///make sure last 4 bits are the same, for cellDmaSmallGet
+	uint32_t last4BitsOffset = ea & 0x0f;
+	char* tmpTarget = tmpBuffer + last4BitsOffset;
+#ifdef WIN32
+
+#ifdef USE_MEMCPY
+		memcpy(tmpTarget,mainMem,size);
+#else
+		for ( i=0;i<size;i++)
+		{
+			tmpTarget[i] = mainMem[i];
+		}
+#endif //USE_MEMCPY
+#else
+	cellDmaSmallGet(tmpTarget,ea,size,DMA_TAG(1),0,0);
+	//copy into final destination
+#endif //WIN32
+
+	cellDmaWaitTagStatusAll(DMA_MASK(1));
+
+	//this is slowish, perhaps memcpy on SPU is smarter?
+	for (i=0; btLikely( i<size );i++)
+	{
+		localStore[i] = tmpTarget[i];
+	}
+
+	return 0;
+}
+
+
+#ifdef WIN32
+
+int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	char* mainMem = (char*)ea;
+	char* localStore = (char*)ls;
+
+#ifdef USE_MEMCPY
+	memcpy(localStore,mainMem,size);
+#else
+	for (uint32_t i=0;i<size;i++)
+	{
+		localStore[i] = mainMem[i];
+	}
+#endif
+	return 0;
+}
+
+int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	char* mainMem = (char*)ea;
+	char* localStore = (char*)ls;
+#ifdef USE_MEMCPY
+	memcpy(localStore,mainMem,size);
+#else
+	for (uint32_t i=0;i<size;i++)
+	{
+		localStore[i] = mainMem[i];
+	}	
+#endif //#ifdef USE_MEMCPY
+	return 0;
+}
+
+int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	char* mainMem = (char*)ea;
+	const char* localStore = (const char*)ls;
+#ifdef USE_MEMCPY
+	memcpy(mainMem,localStore,size);
+#else
+	for (uint32_t i=0;i<size;i++)
+	{
+		mainMem[i] = localStore[i];
+	}	
+#endif //#ifdef USE_MEMCPY
+
+	return 0;
+}
+
+
+
+void	cellDmaWaitTagStatusAll(int ignore)
+{
+
+}
+
+#endif //WIN32
diff --git a/Extras/BulletMultiThreaded/SpuFakeDma.h b/Extras/BulletMultiThreaded/SpuFakeDma.h
index f8ee231f5..45bec865b 100644
--- a/Extras/BulletMultiThreaded/SpuFakeDma.h
+++ b/Extras/BulletMultiThreaded/SpuFakeDma.h
@@ -1,36 +1,121 @@
-
-#ifndef FAKE_DMA_H
-#define FAKE_DMA_H
-
-
-#include "PlatformDefinitions.h"
-
-
-
-#ifdef __CELLOS_LV2__
-
-#include <cell/dma.h>
-#include <stdint.h>
-
-#define DMA_TAG(xfer) (xfer + 1)
-#define DMA_MASK(xfer) (1 << DMA_TAG(xfer))
-
-#elif defined (WIN32)
-
-#define DMA_TAG(a) (a)
-#define DMA_MASK(a) (a)
-
-
-/// cellDmaLargeGet Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
-int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
-int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
-/// cellDmaLargePut Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
-int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
-/// cellDmaWaitTagStatusAll Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
-void	cellDmaWaitTagStatusAll(int ignore);
-#endif //WIN32
-
-///stallingUnalignedDmaSmallGet internally uses DMA_TAG(1)
-int	stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size);
-
-#endif //FAKE_DMA_H
+
+#ifndef FAKE_DMA_H
+#define FAKE_DMA_H
+
+
+#include "PlatformDefinitions.h"
+
+
+
+#ifdef __CELLOS_LV2__
+
+#include <cell/dma.h>
+#include <stdint.h>
+
+#define DMA_TAG(xfer) (xfer + 1)
+#define DMA_MASK(xfer) (1 << DMA_TAG(xfer))
+
+
+#else
+#ifdef WIN32
+
+#define DMA_TAG(a) (a)
+#define DMA_MASK(a) (a)
+
+
+		/// cellDmaLargeGet Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+		int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+		int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+		/// cellDmaLargePut Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+		int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+		/// cellDmaWaitTagStatusAll Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+		void	cellDmaWaitTagStatusAll(int ignore);
+
+
+#elif defined(USE_LIBSPE2)
+
+#define DMA_TAG(xfer) (xfer + 1)
+#define DMA_MASK(xfer) (1 << DMA_TAG(xfer))
+		
+#include <spu_mfcio.h>		
+		
+//#define DEBUG_DMA		
+#ifdef DEBUG_DMA
+#define dUASSERT(a,b) if (!(a)) { printf(b);}
+#ifdef USE_ADDR64
+#define uintsize unsigned long long
+#else
+#define uintsize unsigned int
+#endif
+		
+#define cellDmaLargeGet(ls, ea, size, tag, tid, rid) if (  (((uintsize)ls%16) != ((uintsize)ea%16)) || ((((uintsize)ea%16) || ((uintsize)ls%16)) && (( ((uintsize)ls%16) != ((uintsize)size%16) ) || ( ((uintsize)ea%16) != ((uintsize)size%16) ) ) ) || ( ((uintsize)size%16) && ((uintsize)size!=1) && ((uintsize)size!=2) && ((uintsize)size!=4) && ((uintsize)size!=8) ) || (size >= 16384) || !(uintsize)ls || !(uintsize)ea) { \
+															dUASSERT( (((uintsize)ea % 16) == 0) || (size < 16), "XDR Address not aligned: "); \
+															dUASSERT( (((uintsize)ls % 16) == 0) || (size < 16), "LS Address not aligned: "); \
+															dUASSERT( ((((uintsize)ls % size) == 0) && (((uintsize)ea % size) == 0))  || (size > 16), "Not naturally aligned: "); \
+															dUASSERT((size == 1) || (size == 2) || (size == 4) || (size == 8) || ((size % 16) == 0), "size not a multiple of 16byte: "); \
+															dUASSERT(size < 16384, "size too big: "); \
+															dUASSERT( ((uintsize)ea%16)==((uintsize)ls%16), "wrong Quadword alignment of LS and EA: "); \
+	    													dUASSERT(ea != 0, "Nullpointer EA: "); dUASSERT(ls != 0, "Nullpointer LS: ");\
+															printf("GET %s:%d from: 0x%x, to: 0x%x - %d bytes\n", __FILE__, __LINE__, (unsigned int)ea,(unsigned int)ls,(unsigned int)size);\
+															} \
+															mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaGet(ls, ea, size, tag, tid, rid) if (  (((uintsize)ls%16) != ((uintsize)ea%16)) || ((((uintsize)ea%16) || ((uintsize)ls%16)) && (( ((uintsize)ls%16) != ((uintsize)size%16) ) || ( ((uintsize)ea%16) != ((uintsize)size%16) ) ) ) || ( ((uintsize)size%16) && ((uintsize)size!=1) && ((uintsize)size!=2) && ((uintsize)size!=4) && ((uintsize)size!=8) ) || (size >= 16384) || !(uintsize)ls || !(uintsize)ea) { \
+														dUASSERT( (((uintsize)ea % 16) == 0) || (size < 16), "XDR Address not aligned: "); \
+														dUASSERT( (((uintsize)ls % 16) == 0) || (size < 16), "LS Address not aligned: "); \
+														dUASSERT( ((((uintsize)ls % size) == 0) && (((uintsize)ea % size) == 0))  || (size > 16), "Not naturally aligned: "); \
+														dUASSERT((size == 1) || (size == 2) || (size == 4) || (size == 8) || ((size % 16) == 0), "size not a multiple of 16byte: "); \
+    													dUASSERT(size < 16384, "size too big: "); \
+														dUASSERT( ((uintsize)ea%16)==((uintsize)ls%16), "wrong Quadword alignment of LS and EA: "); \
+    													dUASSERT(ea != 0, "Nullpointer EA: "); dUASSERT(ls != 0, "Nullpointer LS: ");\
+    													printf("GET %s:%d from: 0x%x, to: 0x%x - %d bytes\n", __FILE__, __LINE__, (unsigned int)ea,(unsigned int)ls,(unsigned int)size);\
+														} \
+														mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaLargePut(ls, ea, size, tag, tid, rid) if (  (((uintsize)ls%16) != ((uintsize)ea%16)) || ((((uintsize)ea%16) || ((uintsize)ls%16)) && (( ((uintsize)ls%16) != ((uintsize)size%16) ) || ( ((uintsize)ea%16) != ((uintsize)size%16) ) ) ) || ( ((uintsize)size%16) && ((uintsize)size!=1) && ((uintsize)size!=2) && ((uintsize)size!=4) && ((uintsize)size!=8) ) || (size >= 16384) || !(uintsize)ls || !(uintsize)ea) { \
+															dUASSERT( (((uintsize)ea % 16) == 0) || (size < 16), "XDR Address not aligned: "); \
+															dUASSERT( (((uintsize)ls % 16) == 0) || (size < 16), "LS Address not aligned: "); \
+															dUASSERT( ((((uintsize)ls % size) == 0) && (((uintsize)ea % size) == 0))  || (size > 16), "Not naturally aligned: "); \
+															dUASSERT((size == 1) || (size == 2) || (size == 4) || (size == 8) || ((size % 16) == 0), "size not a multiple of 16byte: "); \
+        													dUASSERT(size < 16384, "size too big: "); \
+															dUASSERT( ((uintsize)ea%16)==((uintsize)ls%16), "wrong Quadword alignment of LS and EA: "); \
+        													dUASSERT(ea != 0, "Nullpointer EA: "); dUASSERT(ls != 0, "Nullpointer LS: ");\
+    														printf("PUT %s:%d from: 0x%x, to: 0x%x - %d bytes\n", __FILE__, __LINE__, (unsigned int)ls,(unsigned int)ea,(unsigned int)size); \
+															} \
+															mfc_put(ls, ea, size, tag, tid, rid)
+#define cellDmaSmallGet(ls, ea, size, tag, tid, rid) if (  (((uintsize)ls%16) != ((uintsize)ea%16)) || ((((uintsize)ea%16) || ((uintsize)ls%16)) && (( ((uintsize)ls%16) != ((uintsize)size%16) ) || ( ((uintsize)ea%16) != ((uintsize)size%16) ) ) ) || ( ((uintsize)size%16) && ((uintsize)size!=1) && ((uintsize)size!=2) && ((uintsize)size!=4) && ((uintsize)size!=8) ) || (size >= 16384) || !(uintsize)ls || !(uintsize)ea) { \
+																dUASSERT( (((uintsize)ea % 16) == 0) || (size < 16), "XDR Address not aligned: "); \
+																dUASSERT( (((uintsize)ls % 16) == 0) || (size < 16), "LS Address not aligned: "); \
+																dUASSERT( ((((uintsize)ls % size) == 0) && (((uintsize)ea % size) == 0))  || (size > 16), "Not naturally aligned: "); \
+    															dUASSERT((size == 1) || (size == 2) || (size == 4) || (size == 8) || ((size % 16) == 0), "size not a multiple of 16byte: "); \
+    															dUASSERT(size < 16384, "size too big: "); \
+    															dUASSERT( ((uintsize)ea%16)==((uintsize)ls%16), "wrong Quadword alignment of LS and EA: "); \
+    	    													dUASSERT(ea != 0, "Nullpointer EA: "); dUASSERT(ls != 0, "Nullpointer LS: ");\
+    															printf("GET %s:%d from: 0x%x, to: 0x%x - %d bytes\n", __FILE__, __LINE__, (unsigned int)ea,(unsigned int)ls,(unsigned int)size);\
+																} \
+																mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaWaitTagStatusAll(ignore) mfc_write_tag_mask(ignore) ; mfc_read_tag_status_all()
+
+#else
+#define cellDmaLargeGet(ls, ea, size, tag, tid, rid) mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaGet(ls, ea, size, tag, tid, rid) mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaLargePut(ls, ea, size, tag, tid, rid) mfc_put(ls, ea, size, tag, tid, rid)
+#define cellDmaSmallGet(ls, ea, size, tag, tid, rid) mfc_get(ls, ea, size, tag, tid, rid)
+#define cellDmaWaitTagStatusAll(ignore) mfc_write_tag_mask(ignore) ; mfc_read_tag_status_all()
+#endif // DEBUG_DMA
+
+		
+		
+		
+		
+		
+		
+		
+		
+		
+#endif // WIN32
+
+#endif //__CELLOS_LV2__
+
+///stallingUnalignedDmaSmallGet internally uses DMA_TAG(1)
+int	stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size);
+
+#endif //FAKE_DMA_H