Optimisations and experimental HDR display support on mac

This commit completes the previous ones for performance optimisations,
and adds HDR display support on mac

7 files changed, 67 insertions, 24 deletions

diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 42af68c276..9ff483b92b 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -357,7 +357,9 @@ S32 LLImageGL::dataFormatBits(S32 dataformat)
     case GL_BGRA:                                   return 32;      // Used for QuickTime media textures on the Mac
 #endif
     case GL_DEPTH_COMPONENT:                        return 24;
+    case GL_DEPTH_COMPONENT16:                      return 16;
     case GL_DEPTH_COMPONENT24:                      return 24;
+    case GL_DEPTH_COMPONENT32:                      return 32;
     case GL_RGBA16:                                 return 64;
     case GL_R16F:                                   return 16;
     case GL_RG16F:                                  return 32;
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index d0e268f08c..5d26a69d7b 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -530,7 +530,9 @@ void LLTexUnit::setTextureFilteringOption(LLTexUnit::eTextureFilterOptions optio
     {
         if (LLImageGL::sGlobalUseAnisotropic && option == TFO_ANISOTROPIC)
         {
-            glTexParameterf(target, GL_TEXTURE_MAX_ANISOTROPY_EXT, gGLManager.mMaxAnisotropy);
+            //glTexParameterf(target, GL_TEXTURE_MAX_ANISOTROPY_EXT, gGLManager.mMaxAnisotropy);
+            //We plan to add a setting. For now we stick to a low value.
+            glTexParameterf(target, GL_TEXTURE_MAX_ANISOTROPY_EXT, 4.0);
         }
         else
         {
@@ -1592,7 +1594,6 @@ void LLRender::end(std::string comment_)
 
 void LLRender::flush(std::string comment_)
 {
-    LOG_GLERROR("LLRender::flush() begin " + comment_);
     if (mCount > 0)
     {
         LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
@@ -1661,9 +1662,9 @@ void LLRender::flush(std::string comment_)
         }
 
         resetStriders(count);
-    }
 
-    LOG_GLERROR("LLRender::flush() end " + comment_);
+        LOG_GLERROR("LLRender::flush() end " + comment_);
+    }
 }
 
 LLVertexBuffer* LLRender::bufferfromCache(U32 attribute_mask, U32 count)
@@ -1760,9 +1761,11 @@ LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count)
         vb->setColorData(mColorsp.get());
     }
 
-#if LL_DARWIN
-    vb->unmapBuffer();
-#endif
+    if(gGLManager.mIsApple && LLVertexBuffer::getVertexBufferMode() == 0)
+    {
+        vb->unmapBuffer();
+    }
+
     vb->unbind();
 
     return vb;
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index 957c85b606..4fd32620e1 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -368,7 +368,7 @@ bool LLRenderTarget::allocateDepth()
 
     U32 internal_type = LLTexUnit::getInternalType(mUsage);
 
-    LLImageGL::setManualImage(internal_type, 0, GL_DEPTH_COMPONENT24, mResX, mResY, GL_DEPTH_COMPONENT, GL_FLOAT, NULL, false);
+    LLImageGL::setManualImage(internal_type, 0, GL_DEPTH_COMPONENT24, mResX, mResY, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, NULL, false);
     gGL.getTexUnit(0)->setTextureFilteringOption(LLTexUnit::TFO_POINT);
 
     sBytesAllocated += mResX*mResY*4;
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 3bc71a7413..fa1c686e8e 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -1512,6 +1512,8 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("bloomExtractORM");
     mReservedUniforms.push_back("bloomExtractMetal");
     mReservedUniforms.push_back("bloomExtractNonMetal");
+    mReservedUniforms.push_back("bloomClampValue");
+    mReservedUniforms.push_back("mpHDRBoost");
 
     llassert(mReservedUniforms.size() == END_RESERVED_UNIFORMS);
 
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index 8b0bfc6bbe..19416066a5 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -352,6 +352,8 @@ public:
         BLOOM_EXTRACT_ORM,                  //  "bloomExtractORM"
         BLOOM_EXTRACT_METAL,                //  "bloomExtractMetal"
         BLOOM_EXTRACT_NONMETAL,             //  "bloomExtractNonMetal"
+        BLOOM_CLAMP,                        //  "bloomClampValue"
+        MP_HDR_BOOST,                       //  "mpHDRBoost"
 
         END_RESERVED_UNIFORMS
     } eGLSLReservedUniforms;
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 2a0acab05a..040dde7317 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -29,7 +29,6 @@
 #include "llfasttimer.h"
 #include "llsys.h"
 #include "llvertexbuffer.h"
-// #include "llrender.h"
 #include "llglheaders.h"
 #include "llrender.h"
 #include "llvector4a.h"
@@ -272,13 +271,13 @@ static GLuint gen_buffer()
     {
         LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("gen buffer");
         sIndex = pool_size;
-#if !LL_DARWIN
+//#if !LL_DARWIN
         if (!gGLManager.mIsAMD)
         {
             glGenBuffers(pool_size, sNamePool);
         }
         else
-#endif
+//#endif
         { // work around for AMD driver bug
             for (U32 i = 0; i < pool_size; ++i)
             {
@@ -615,6 +614,8 @@ public:
 
 static LLVBOPool* sVBOPool = nullptr;
 
+static U32 sMPVertexBufferMode = 0;
+
 void LLVertexBufferData::drawWithMatrix()
 {
     if (!mVB)
@@ -946,11 +947,13 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
 }
 
 //static
-void LLVertexBuffer::initClass(LLWindow* window)
+void LLVertexBuffer::initClass(LLWindow* window, U32 mode_)
 {
     llassert(sVBOPool == nullptr);
 
-    if (gGLManager.mIsApple)
+    sMPVertexBufferMode = mode_;
+
+    if (mode_ == 0 && gGLManager.mIsApple)
     {
         LL_INFOS() << "VBO Pooling Disabled" << LL_ENDL;
         sVBOPool = new LLAppleVBOPool();
@@ -973,6 +976,12 @@ void LLVertexBuffer::initClass(LLWindow* window)
 }
 
 //static
+U32 LLVertexBuffer::getVertexBufferMode()
+{
+    return sMPVertexBufferMode;
+}
+
+//static
 void LLVertexBuffer::unbind()
 {
     glBindBuffer(GL_ARRAY_BUFFER, 0);
@@ -1291,7 +1300,7 @@ U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 inde
         count = mNumVerts - index;
     }
 
-    if (!gGLManager.mIsApple)
+    if (!gGLManager.mIsApple || sMPVertexBufferMode == 1)
     {
         U32 start = mOffsets[type] + sTypeSize[type] * index;
         U32 end = start + sTypeSize[type] * count-1;
@@ -1328,7 +1337,7 @@ U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count)
         count = mNumIndices-index;
     }
 
-    if (!gGLManager.mIsApple)
+    if (!gGLManager.mIsApple || sMPVertexBufferMode == 1)
     {
         U32 start = sizeof(U16) * index;
         U32 end = start + sizeof(U16) * count-1;
@@ -1365,13 +1374,35 @@ void LLVertexBuffer::flush_vbo(GLenum target, U32 start, U32 end, void* data, U8
 {
     if (gGLManager.mIsApple)
     {
-        // on OS X, flush_vbo doesn't actually write to the GL buffer, so be sure to call
-        // _mapBuffer to tag the buffer for flushing to GL
-        _mapBuffer();
-        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb memcpy");
-        //LOG_GLERROR("LLVertexBuffer::flush_vbo()");
-        // copy into mapped buffer
-        memcpy(dst+start, data, end-start+1);
+        if(sMPVertexBufferMode == 1)
+        {
+            //LL_WARNS() << "flush_vbo mode 1" << LL_ENDL;
+
+            U32 MapBits = GL_MAP_WRITE_BIT;
+            //U32 MapBits = GL_MAP_READ_BIT;
+            U32 buffer_size = end-start+1;
+
+            U8 * mptr = NULL;
+            mptr = (U8*) glMapBufferRange( target, start, end-start+1, MapBits);
+
+            if (mptr)
+            {
+                std::memcpy(mptr, (U8*) data, buffer_size);
+                if(!glUnmapBuffer(target)) LL_WARNS() << "glUnmapBuffer() failed" << LL_ENDL;
+            }
+            else LL_WARNS() << "glMapBufferRange() returned NULL" << LL_ENDL;
+
+        }
+        else
+        {
+            //LL_WARNS() << "flush_vbo mode 0" << LL_ENDL;
+            // on OS X, flush_vbo doesn't actually write to the GL buffer, so be sure to call
+            // _mapBuffer to tag the buffer for flushing to GL
+            _mapBuffer();
+            LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb memcpy");
+            // copy into mapped buffer
+            memcpy(dst+start, data, end-start+1);
+        }
     }
     else
     {
@@ -1428,7 +1459,7 @@ void LLVertexBuffer::_unmapBuffer()
         }
     };
 
-    if (gGLManager.mIsApple)
+    if (gGLManager.mIsApple && sMPVertexBufferMode == 0)
     {
         LOG_GLERROR("LLVertexBuffer::_unmapBuffer() - apple 1");
         if (mMappedData)
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 375ad76fb8..4239477b8d 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -109,7 +109,7 @@ public:
         return *this;
     }
 
-    static void initClass(LLWindow* window);
+    static void initClass(LLWindow* window, U32 mode_ = 0);
     static void cleanupClass();
     static void setupClientArrays(U32 data_mask);
     static void drawArrays(U32 mode, const std::vector<LLVector3>& pos);
@@ -128,6 +128,9 @@ public:
     // flush any pending mapped buffers
     static void flushBuffers();
 
+    static U32 getVertexBufferMode();
+
+
     //WARNING -- when updating these enums you MUST
     // 1 - update LLVertexBuffer::sTypeSize
     // 2 - update LLVertexBuffer::vb_type_name