SH-2038 Fix for some compiler errors from the apple tweaks -- also add a fence API (disabled for now).

5 files changed, 182 insertions, 10 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 02c975a17a..e07ff0015c 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -128,9 +128,21 @@ PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB = NULL;
 PFNGLGETBUFFERPOINTERVARBPROC		glGetBufferPointervARB = NULL;
 
 // GL_ARB_map_buffer_range
-PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
-PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
-
+PFNGLMAPBUFFERRANGEPROC			glMapBufferRange = NULL;
+PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange = NULL;
+
+// GL_ARB_sync
+PFNGLFENCESYNCPROC				glFenceSync = NULL;
+PFNGLISSYNCPROC					glIsSync = NULL;
+PFNGLDELETESYNCPROC				glDeleteSync = NULL;
+PFNGLCLIENTWAITSYNCPROC			glClientWaitSync = NULL;
+PFNGLWAITSYNCPROC				glWaitSync = NULL;
+PFNGLGETINTEGER64VPROC			glGetInteger64v = NULL;
+PFNGLGETSYNCIVPROC				glGetSynciv = NULL;
+
+// GL_APPLE_flush_buffer_range
+PFNGLBUFFERPARAMETERIAPPLEPROC	glBufferParameteriAPPLE = NULL;
+PFNGLFLUSHMAPPEDBUFFERRANGEAPPLEPROC glFlushMappedBufferRangeAPPLE = NULL;
 
 // vertex object prototypes
 PFNGLNEWOBJECTBUFFERATIPROC			glNewObjectBufferATI = NULL;
@@ -334,7 +346,7 @@ LLGLManager::LLGLManager() :
 	mHasFramebufferObject(FALSE),
 	mMaxSamples(0),
 	mHasBlendFuncSeparate(FALSE),
-
+	mHasSync(FALSE),
 	mHasVertexBufferObject(FALSE),
 	mHasMapBufferRange(FALSE),
 	mHasFlushBufferRange(FALSE),
@@ -775,6 +787,7 @@ void LLGLManager::initExtensions()
 	mHasOcclusionQuery = ExtensionExists("GL_ARB_occlusion_query", gGLHExts.mSysExts);
 	mHasOcclusionQuery2 = ExtensionExists("GL_ARB_occlusion_query2", gGLHExts.mSysExts);
 	mHasVertexBufferObject = ExtensionExists("GL_ARB_vertex_buffer_object", gGLHExts.mSysExts);
+	mHasSync = ExtensionExists("GL_ARB_sync", gGLHExts.mSysExts);
 	mHasMapBufferRange = ExtensionExists("GL_ARB_map_buffer_range", gGLHExts.mSysExts);
 	mHasFlushBufferRange = ExtensionExists("GL_APPLE_flush_buffer_range", gGLHExts.mSysExts);
 	mHasDepthClamp = ExtensionExists("GL_ARB_depth_clamp", gGLHExts.mSysExts) || ExtensionExists("GL_NV_depth_clamp", gGLHExts.mSysExts);
@@ -971,6 +984,16 @@ void LLGLManager::initExtensions()
 			mHasVertexBufferObject = FALSE;
 		}
 	}
+	if (mHasSync)
+	{
+		glFenceSync = (PFNGLFENCESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glFenceSync");

+		glIsSync = (PFNGLISSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glIsSync");

+		glDeleteSync = (PFNGLDELETESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteSync");

+		glClientWaitSync = (PFNGLCLIENTWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glClientWaitSync");

+		glWaitSync = (PFNGLWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glWaitSync");

+		glGetInteger64v = (PFNGLGETINTEGER64VPROC) GLH_EXT_GET_PROC_ADDRESS("glGetInteger64v");

+		glGetSynciv = (PFNGLGETSYNCIVPROC) GLH_EXT_GET_PROC_ADDRESS("glGetSynciv");
+	}
 	if (mHasMapBufferRange)
 	{
 		glMapBufferRange = (PFNGLMAPBUFFERRANGEPROC) GLH_EXT_GET_PROC_ADDRESS("glMapBufferRange");
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 3e98c04321..d736133f3f 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -88,6 +88,7 @@ public:
 		
 	// ARB Extensions
 	BOOL mHasVertexBufferObject;
+	BOOL mHasSync;
 	BOOL mHasMapBufferRange;
 	BOOL mHasFlushBufferRange;
 	BOOL mHasPBuffer;
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index f35f329f00..851a75629e 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -68,6 +68,19 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_sync
+extern PFNGLFENCESYNCPROC				glFenceSync;
+extern PFNGLISSYNCPROC					glIsSync;
+extern PFNGLDELETESYNCPROC				glDeleteSync;
+extern PFNGLCLIENTWAITSYNCPROC			glClientWaitSync;
+extern PFNGLWAITSYNCPROC				glWaitSync;
+extern PFNGLGETINTEGER64VPROC			glGetInteger64v;
+extern PFNGLGETSYNCIVPROC				glGetSynciv;
+
+// GL_APPLE_flush_buffer_range
+extern PFNGLBUFFERPARAMETERIAPPLEPROC	glBufferParameteriAPPLE;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEAPPLEPROC glFlushMappedBufferRangeAPPLE;
+
 // GL_ARB_map_buffer_range
 extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
 extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
@@ -310,6 +323,19 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_sync
+extern PFNGLFENCESYNCPROC				glFenceSync;
+extern PFNGLISSYNCPROC					glIsSync;
+extern PFNGLDELETESYNCPROC				glDeleteSync;
+extern PFNGLCLIENTWAITSYNCPROC			glClientWaitSync;
+extern PFNGLWAITSYNCPROC				glWaitSync;
+extern PFNGLGETINTEGER64VPROC			glGetInteger64v;
+extern PFNGLGETSYNCIVPROC				glGetSynciv;
+
+// GL_APPLE_flush_buffer_range
+extern PFNGLBUFFERPARAMETERIAPPLEPROC	glBufferParameteriAPPLE;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEAPPLEPROC glFlushMappedBufferRangeAPPLE;
+
 // GL_ARB_map_buffer_range
 extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
 extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
@@ -519,6 +545,19 @@ extern PFNGLUNMAPBUFFERARBPROC		glUnmapBufferARB;
 extern PFNGLGETBUFFERPARAMETERIVARBPROC	glGetBufferParameterivARB;
 extern PFNGLGETBUFFERPOINTERVARBPROC	glGetBufferPointervARB;
 
+// GL_ARB_sync
+extern PFNGLFENCESYNCPROC				glFenceSync;
+extern PFNGLISSYNCPROC					glIsSync;
+extern PFNGLDELETESYNCPROC				glDeleteSync;
+extern PFNGLCLIENTWAITSYNCPROC			glClientWaitSync;
+extern PFNGLWAITSYNCPROC				glWaitSync;
+extern PFNGLGETINTEGER64VPROC			glGetInteger64v;
+extern PFNGLGETSYNCIVPROC				glGetSynciv;
+
+// GL_APPLE_flush_buffer_range
+extern PFNGLBUFFERPARAMETERIAPPLEPROC	glBufferParameteriAPPLE;
+extern PFNGLFLUSHMAPPEDBUFFERRANGEAPPLEPROC glFlushMappedBufferRangeAPPLE;
+
 // GL_ARB_map_buffer_range
 extern PFNGLMAPBUFFERRANGEPROC			glMapBufferRange;
 extern PFNGLFLUSHMAPPEDBUFFERRANGEPROC	glFlushMappedBufferRange;
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index db9c8f83f8..2a297bcc45 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -65,6 +65,50 @@ S32	LLVertexBuffer::sWeight4Loc = -1;
 std::vector<U32> LLVertexBuffer::sDeleteList;
 
 
+const U32 FENCE_WAIT_TIME_NANOSECONDS = 10000;  //1 ms
+
+class LLGLSyncFence : public LLGLFence
+{
+public:
+	GLsync mSync;
+	
+	LLGLSyncFence()
+	{
+		mSync = 0;
+	}
+
+	~LLGLSyncFence()
+	{
+		if (mSync)
+		{
+			glDeleteSync(mSync);
+		}
+	}
+
+	void placeFence()
+	{
+		if (mSync)
+		{
+			glDeleteSync(mSync);
+		}
+		mSync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+	}
+
+	void wait()
+	{
+		if (mSync)
+		{
+			while (glClientWaitSync(mSync, 0, FENCE_WAIT_TIME_NANOSECONDS) == GL_TIMEOUT_EXPIRED)
+			{ //track the number of times we've waited here
+				static S32 waits = 0;
+				waits++;
+			}
+		}
+	}
+
+
+};
+
 S32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] =
 {
 	sizeof(LLVector4), // TYPE_VERTEX,
@@ -309,6 +353,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi
 	glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, 
 		idx);
 	stop_glerror();
+	placeFence();
 }
 
 void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
@@ -340,6 +385,7 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 	glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT,
 		((U16*) getIndicesPointer()) + indices_offset);
 	stop_glerror();
+	placeFence();
 }
 
 void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
@@ -365,6 +411,7 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
 	stop_glerror();
 	glDrawArrays(sGLMode[mode], first, count);
 	stop_glerror();
+	placeFence();
 }
 
 //static
@@ -444,9 +491,11 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) :
 	mFilthy(FALSE),
 	mEmpty(TRUE),
 	mResized(FALSE),
-	mDynamicSize(FALSE)
+	mDynamicSize(FALSE),
+	mFence(NULL)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_CONSTRUCTOR);
+	mFence = NULL;
 	if (!sEnableVBOs)
 	{
 		mUsage = 0 ; 
@@ -527,9 +576,40 @@ LLVertexBuffer::~LLVertexBuffer()
 	destroyGLIndices();
 	sCount--;
 
+	if (mFence)
+	{
+		delete mFence;
+	}
+	
+	mFence = NULL;
+
 	llassert_always(!mMappedData && !mMappedIndexData) ;
 };
 
+void LLVertexBuffer::placeFence() const
+{
+	/*if (!mFence && useVBOs())
+	{
+		if (gGLManager.mHasSync)
+		{
+			mFence = new LLGLSyncFence();
+		}
+	}
+
+	if (mFence)
+	{
+		mFence->placeFence();
+	}*/
+}
+
+void LLVertexBuffer::waitFence() const
+{
+	/*if (mFence)
+	{
+		mFence->wait();
+	}*/
+}
+
 //----------------------------------------------------------------------------
 
 void LLVertexBuffer::genBuffer()
@@ -967,7 +1047,6 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 		
 	if (useVBOs())
 	{
-
 		if (sDisableVBOMapping || gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange)
 		{
 			if (count == -1)
@@ -1019,6 +1098,7 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 			else
 			{
 				U8* src = NULL;
+				waitFence();
 				if (gGLManager.mHasMapBufferRange)
 				{
 					if (map_range)
@@ -1026,13 +1106,20 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 #ifdef GL_ARB_map_buffer_range
 						S32 offset = mOffsets[type] + sTypeSize[type]*index;
 						S32 length = (sTypeSize[type]*count+0xF) & ~0xF;
-						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT | GL_MAP_INVALIDATE_RANGE_BIT);
+						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, 
+							GL_MAP_WRITE_BIT | 
+							GL_MAP_FLUSH_EXPLICIT_BIT | 
+							GL_MAP_INVALIDATE_RANGE_BIT | 
+							GL_MAP_UNSYNCHRONIZED_BIT);
 #endif
 					}
 					else
 					{
 #ifdef GL_ARB_map_buffer_range
-						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT);
+						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, 
+							GL_MAP_WRITE_BIT | 
+							GL_MAP_FLUSH_EXPLICIT_BIT |
+							GL_MAP_UNSYNCHRONIZED_BIT);
 #endif
 					}
 				}
@@ -1178,6 +1265,7 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 			else
 			{
 				U8* src = NULL;
+				waitFence();
 				if (gGLManager.mHasMapBufferRange)
 				{
 					if (map_range)
@@ -1185,13 +1273,20 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 #ifdef GL_ARB_map_buffer_range
 						S32 offset = sizeof(U16)*index;
 						S32 length = sizeof(U16)*count;
-						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT | GL_MAP_INVALIDATE_RANGE_BIT);
+						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, 
+							GL_MAP_WRITE_BIT | 
+							GL_MAP_FLUSH_EXPLICIT_BIT | 
+							GL_MAP_INVALIDATE_RANGE_BIT |
+							GL_MAP_UNSYNCHRONIZED_BIT);
 #endif
 					}
 					else
 					{
 #ifdef GL_ARB_map_buffer_range
-						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT);
+						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, 
+							GL_MAP_WRITE_BIT | 
+							GL_MAP_FLUSH_EXPLICIT_BIT |
+							GL_MAP_UNSYNCHRONIZED_BIT);
 #endif
 					}
 				}
@@ -1378,7 +1473,9 @@ void LLVertexBuffer::unmapBuffer(S32 type)
 						}
 						else if (gGLManager.mHasFlushBufferRange)
 						{
+#ifdef GL_APPLE_flush_buffer_range
 							glFlushMappedBufferRangeAPPLE(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length);
+#endif
 						}
 						stop_glerror();
 					}
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index aa5df305a6..cc5d11e1c2 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -70,6 +70,12 @@ protected:
 	}
 };
 
+class LLGLFence
+{
+public:
+	virtual void placeFence() = 0;
+	virtual void wait() = 0;
+};
 
 //============================================================================
 // base class 
@@ -270,6 +276,12 @@ protected:
 	std::vector<MappedRegion> mMappedVertexRegions;
 	std::vector<MappedRegion> mMappedIndexRegions;
 
+	mutable LLGLFence* mFence;
+
+	void placeFence() const;
+	void waitFence() const;
+
+
 public:
 	static S32 sCount;
 	static S32 sGLCount;