Merge branch 'main' into marchcat/slua-editor-merge

# Conflicts:
#	.github/workflows/build.yaml
#	indra/newview/llmeshrepository.cpp
#	indra/newview/llmeshrepository.h
#	indra/newview/llpreviewscript.h

17 files changed, 191 insertions, 62 deletions

diff --git a/indra/llrender/llcubemaparray.cpp b/indra/llrender/llcubemaparray.cpp
index 4f5e13765a..998b57217d 100644
--- a/indra/llrender/llcubemaparray.cpp
+++ b/indra/llrender/llcubemaparray.cpp
@@ -105,16 +105,48 @@ LLCubeMapArray::LLCubeMapArray()
 
 }
 
+LLCubeMapArray::LLCubeMapArray(LLCubeMapArray& lhs, U32 width, U32 count) : mTextureStage(0)
+{
+    mWidth = width;
+    mCount = count;
+
+    // Allocate a new cubemap array with the same criteria as the incoming cubemap array
+    allocate(mWidth, lhs.mImage->getComponents(), count, lhs.mImage->getUseMipMaps(), lhs.mHDR);
+
+    // Copy each cubemap from the incoming array to the new array
+    U32 min_count = std::min(count, lhs.mCount);
+    for (U32 i = 0; i < min_count * 6; ++i)
+    {
+        U32 src_resolution = lhs.mWidth;
+        U32 dst_resolution = mWidth;
+        {
+            GLint components = GL_RGB;
+            if (mImage->getComponents() == 4)
+                components = GL_RGBA;
+            GLint format = GL_RGB;
+
+            // Handle different resolutions by scaling the image
+            LLPointer<LLImageRaw> src_image = new LLImageRaw(lhs.mWidth, lhs.mWidth, lhs.mImage->getComponents());
+            glGetTexImage(GL_TEXTURE_CUBE_MAP_ARRAY, 0, components, GL_UNSIGNED_BYTE, src_image->getData());
+
+            LLPointer<LLImageRaw> scaled_image = src_image->scaled(mWidth, mWidth);
+            glTexSubImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, 0, 0, 0, i, mWidth, mWidth, 1, components, GL_UNSIGNED_BYTE, scaled_image->getData());
+        }
+    }
+}
+
 LLCubeMapArray::~LLCubeMapArray()
 {
 }
 
-void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool use_mips)
+void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool use_mips, bool hdr)
 {
     U32 texname = 0;
     mWidth = resolution;
     mCount = count;
 
+    mHDR = hdr;
+
     LLImageGL::generateTextures(1, &texname);
 
     mImage = new LLImageGL(resolution, resolution, components, use_mips);
@@ -127,7 +159,11 @@ void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool us
     bind(0);
     free_cur_tex_image();
 
-    U32 format = components == 4 ? GL_RGBA16F : GL_RGB16F;
+    U32 format = components == 4 ? GL_RGBA16F : GL_R11F_G11F_B10F;
+    if (!hdr)
+    {
+        format = components == 4 ? GL_RGBA8 : GL_RGB8;
+    }
     U32 mip = 0;
     U32 mip_resolution = resolution;
     while (mip_resolution >= 1)
diff --git a/indra/llrender/llcubemaparray.h b/indra/llrender/llcubemaparray.h
index 675aaaf07c..6b4288cb23 100644
--- a/indra/llrender/llcubemaparray.h
+++ b/indra/llrender/llcubemaparray.h
@@ -36,6 +36,7 @@ class LLCubeMapArray : public LLRefCount
 {
 public:
     LLCubeMapArray();
+    LLCubeMapArray(LLCubeMapArray& lhs, U32 width, U32 count);
 
     static GLenum sTargets[6];
 
@@ -52,7 +53,7 @@ public:
     // components - number of components per pixel
     // count - number of cube maps in the array
     // use_mips - if true, mipmaps will be allocated for this cube map array and anisotropic filtering will be used
-    void allocate(U32 res, U32 components, U32 count, bool use_mips = true);
+    void allocate(U32 res, U32 components, U32 count, bool use_mips = true, bool hdr = true);
     void bind(S32 stage);
     void unbind();
 
@@ -73,4 +74,5 @@ protected:
     U32 mWidth = 0;
     U32 mCount = 0;
     S32 mTextureStage;
+    bool mHDR;
 };
diff --git a/indra/llrender/llfontbitmapcache.cpp b/indra/llrender/llfontbitmapcache.cpp
index ee9cfd0719..6a3af1e608 100644
--- a/indra/llrender/llfontbitmapcache.cpp
+++ b/indra/llrender/llfontbitmapcache.cpp
@@ -107,7 +107,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp
             mBitmapHeight = image_height;
 
             S32 num_components = getNumComponents(bitmap_type);
-            mImageRawVec[bitmap_idx].push_back(new LLImageRaw(mBitmapWidth, mBitmapHeight, num_components));
+            mImageRawVec[bitmap_idx].emplace_back(new LLImageRaw(mBitmapWidth, mBitmapHeight, num_components));
             bitmap_num = static_cast<U32>(mImageRawVec[bitmap_idx].size()) - 1;
 
             LLImageRaw* image_raw = getImageRaw(bitmap_type, bitmap_num);
@@ -117,7 +117,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp
             }
 
             // Make corresponding GL image.
-            mImageGLVec[bitmap_idx].push_back(new LLImageGL(image_raw, false, false));
+            mImageGLVec[bitmap_idx].emplace_back(new LLImageGL(image_raw, false, false));
             LLImageGL* image_gl = getImageGL(bitmap_type, bitmap_num);
 
             // Start at beginning of the new image.
@@ -141,6 +141,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp
     bitmap_num = getNumBitmaps(bitmap_type) - 1;
 
     mCurrentOffsetX[bitmap_idx] += width + 1;
+    mGeneration++;
 
     return true;
 }
@@ -168,6 +169,7 @@ void LLFontBitmapCache::reset()
 
     mBitmapWidth = 0;
     mBitmapHeight = 0;
+    mGeneration++;
 }
 
 //static
diff --git a/indra/llrender/llfontbitmapcache.h b/indra/llrender/llfontbitmapcache.h
index f2dfd87877..0ae4e6bed0 100644
--- a/indra/llrender/llfontbitmapcache.h
+++ b/indra/llrender/llfontbitmapcache.h
@@ -63,6 +63,7 @@ public:
     U32 getNumBitmaps(EFontGlyphType bitmapType) const { return (bitmapType < EFontGlyphType::Count) ? static_cast<U32>(mImageRawVec[static_cast<U32>(bitmapType)].size()) : 0U; }
     S32 getBitmapWidth() const { return mBitmapWidth; }
     S32 getBitmapHeight() const { return mBitmapHeight; }
+    S32 getCacheGeneration() const { return mGeneration; }
 
 protected:
     static U32 getNumComponents(EFontGlyphType bitmap_type);
@@ -74,6 +75,7 @@ private:
     S32 mCurrentOffsetY[static_cast<U32>(EFontGlyphType::Count)] = { 1 };
     S32 mMaxCharWidth = 0;
     S32 mMaxCharHeight = 0;
+    S32 mGeneration = 0;
     std::vector<LLPointer<LLImageRaw>> mImageRawVec[static_cast<U32>(EFontGlyphType::Count)];
     std::vector<LLPointer<LLImageGL>> mImageGLVec[static_cast<U32>(EFontGlyphType::Count)];
 };
diff --git a/indra/llrender/llfontfreetype.cpp b/indra/llrender/llfontfreetype.cpp
index 6128e03fa7..5f04664267 100644
--- a/indra/llrender/llfontfreetype.cpp
+++ b/indra/llrender/llfontfreetype.cpp
@@ -146,7 +146,6 @@ LLFontFreetype::LLFontFreetype()
     mIsFallback(false),
     mFTFace(NULL),
     mRenderGlyphCount(0),
-    mAddGlyphCount(0),
     mStyle(0),
     mPointSize(0)
 {
@@ -552,7 +551,7 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l
         return NULL;
 
     llassert(!mIsFallback);
-    fontp->renderGlyph(requested_glyph_type, glyph_index);
+    fontp->renderGlyph(requested_glyph_type, glyph_index, wch);
 
     EFontGlyphType bitmap_glyph_type = EFontGlyphType::Unspecified;
     switch (fontp->mFTFace->glyph->bitmap.pixel_mode)
@@ -574,7 +573,6 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l
     S32 pos_x, pos_y;
     U32 bitmap_num;
     mFontBitmapCachep->nextOpenPos(width, pos_x, pos_y, bitmap_glyph_type, bitmap_num);
-    mAddGlyphCount++;
 
     LLFontGlyphInfo* gi = new LLFontGlyphInfo(glyph_index, requested_glyph_type);
     gi->mXBitmapOffset = pos_x;
@@ -656,7 +654,14 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l
 
     LLImageGL *image_gl = mFontBitmapCachep->getImageGL(bitmap_glyph_type, bitmap_num);
     LLImageRaw *image_raw = mFontBitmapCachep->getImageRaw(bitmap_glyph_type, bitmap_num);
-    image_gl->setSubImage(image_raw, 0, 0, image_gl->getWidth(), image_gl->getHeight());
+    if (image_gl && image_raw)
+    {
+        image_gl->setSubImage(image_raw, 0, 0, image_gl->getWidth(), image_gl->getHeight());
+    }
+    else
+    {
+        llassert(false); //images were just inserted by nextOpenPos, they shouldn't be missing
+    }
 
     return gi;
 }
@@ -697,7 +702,7 @@ void LLFontFreetype::insertGlyphInfo(llwchar wch, LLFontGlyphInfo* gi) const
     }
 }
 
-void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) const
+void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index, llwchar wch) const
 {
     if (mFTFace == NULL)
         return;
@@ -712,11 +717,28 @@ void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) co
     FT_Error error = FT_Load_Glyph(mFTFace, glyph_index, load_flags);
     if (FT_Err_Ok != error)
     {
+        if (error == FT_Err_Out_Of_Memory)
+        {
+            LLError::LLUserWarningMsg::showOutOfMemory();
+            LL_ERRS() << "Out of memory loading glyph for character " << llformat("U+%xu", U32(wch)) << LL_ENDL;
+        }
+
         std::string message = llformat(
-            "Error %d (%s) loading glyph %u: bitmap_type=%u, load_flags=%d",
-            error, FT_Error_String(error), glyph_index, bitmap_type, load_flags);
+            "Error %d (%s) loading wchar %u glyph %u/%u: bitmap_type=%u, load_flags=%d",
+            error, FT_Error_String(error), wch, glyph_index, mFTFace->num_glyphs, bitmap_type, load_flags);
         LL_WARNS_ONCE() << message << LL_ENDL;
         error = FT_Load_Glyph(mFTFace, glyph_index, load_flags ^ FT_LOAD_COLOR);
+        if (FT_Err_Invalid_Outline == error
+            || FT_Err_Invalid_Composite == error
+            || (FT_Err_Ok != error && LLStringOps::isEmoji(wch)))
+        {
+            // value~0 always corresponds to the 'missing glyph'
+            error = FT_Load_Glyph(mFTFace, 0, FT_LOAD_FORCE_AUTOHINT);
+            if (FT_Err_Ok != error)
+            {
+                LL_ERRS() << "Loading fallback for char '" << (U32)wch << "', glyph " << glyph_index << " failed with error : " << (S32)error << LL_ENDL;
+            }
+        }
         llassert_always_msg(FT_Err_Ok == error, message.c_str());
     }
 
@@ -823,7 +845,12 @@ bool LLFontFreetype::setSubImageBGRA(U32 x, U32 y, U32 bitmap_num, U16 width, U1
 {
     LLImageRaw* image_raw = mFontBitmapCachep->getImageRaw(EFontGlyphType::Color, bitmap_num);
     llassert(!mIsFallback);
-    llassert(image_raw && (image_raw->getComponents() == 4));
+    if (!image_raw)
+    {
+        llassert(false);
+        return false;
+    }
+    llassert(image_raw->getComponents() == 4);
 
     // NOTE: inspired by LLImageRaw::setSubImage()
     U32* image_data = (U32*)image_raw->getData();
@@ -851,10 +878,17 @@ bool LLFontFreetype::setSubImageBGRA(U32 x, U32 y, U32 bitmap_num, U16 width, U1
 void LLFontFreetype::setSubImageLuminanceAlpha(U32 x, U32 y, U32 bitmap_num, U32 width, U32 height, U8 *data, S32 stride) const
 {
     LLImageRaw *image_raw = mFontBitmapCachep->getImageRaw(EFontGlyphType::Grayscale, bitmap_num);
-    LLImageDataLock lock(image_raw);
 
     llassert(!mIsFallback);
-    llassert(image_raw && (image_raw->getComponents() == 2));
+    if (!image_raw)
+    {
+        llassert(false);
+        return;
+    }
+
+    LLImageDataLock lock(image_raw);
+
+    llassert(image_raw->getComponents() == 2);
 
     U8 *target = image_raw->getData();
     llassert(target);
diff --git a/indra/llrender/llfontfreetype.h b/indra/llrender/llfontfreetype.h
index eba89f5def..783bf4a4b3 100644
--- a/indra/llrender/llfontfreetype.h
+++ b/indra/llrender/llfontfreetype.h
@@ -156,7 +156,7 @@ private:
     bool hasGlyph(llwchar wch) const;       // Has a glyph for this character
     LLFontGlyphInfo* addGlyph(llwchar wch, EFontGlyphType glyph_type) const;        // Add a new character to the font if necessary
     LLFontGlyphInfo* addGlyphFromFont(const LLFontFreetype *fontp, llwchar wch, U32 glyph_index, EFontGlyphType bitmap_type) const; // Add a glyph from this font to the other (returns the glyph_index, 0 if not found)
-    void renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) const;
+    void renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index, llwchar wch) const;
     void insertGlyphInfo(llwchar wch, LLFontGlyphInfo* gi) const;
 
     std::string mName;
@@ -187,7 +187,6 @@ private:
     mutable LLFontBitmapCache* mFontBitmapCachep;
 
     mutable S32 mRenderGlyphCount;
-    mutable S32 mAddGlyphCount;
 };
 
 #endif // LL_FONTFREETYPE_H
diff --git a/indra/llrender/llfontgl.cpp b/indra/llrender/llfontgl.cpp
index 4c9a062246..16eec1fdd2 100644
--- a/indra/llrender/llfontgl.cpp
+++ b/indra/llrender/llfontgl.cpp
@@ -58,6 +58,7 @@ F32 LLFontGL::sVertDPI = 96.f;
 F32 LLFontGL::sHorizDPI = 96.f;
 F32 LLFontGL::sScaleX = 1.f;
 F32 LLFontGL::sScaleY = 1.f;
+S32 LLFontGL::sResolutionGeneration = 0;
 bool LLFontGL::sDisplayFont = true ;
 std::string LLFontGL::sAppDir;
 
@@ -109,6 +110,12 @@ S32 LLFontGL::getNumFaces(const std::string& filename)
     return mFontFreetype->getNumFaces(filename);
 }
 
+S32 LLFontGL::getCacheGeneration() const
+{
+    const LLFontBitmapCache* font_bitmap_cache = mFontFreetype->getFontBitmapCache();
+    return font_bitmap_cache->getCacheGeneration();
+}
+
 S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, const LLRect& rect, const LLColor4 &color, HAlign halign, VAlign valign, U8 style,
     ShadowType shadow, S32 max_chars, F32* right_x, bool use_ellipses, bool use_color) const
 {
@@ -249,6 +256,10 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
 
     const LLFontBitmapCache* font_bitmap_cache = mFontFreetype->getFontBitmapCache();
 
+    // This looks wrong, value is dynamic.
+    // LLFontBitmapCache::nextOpenPos can alter these values when
+    // new characters get added to cache, which affects whole string.
+    // Todo: Perhaps value should update after symbols were added?
     F32 inv_width = 1.f / font_bitmap_cache->getBitmapWidth();
     F32 inv_height = 1.f / font_bitmap_cache->getBitmapHeight();
 
@@ -270,6 +281,10 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
 
     const LLFontGlyphInfo* next_glyph = NULL;
 
+    // string can have more than one glyph per char (ex: bold or shadow),
+    // make sure that GLYPH_BATCH_SIZE won't end up with half a symbol.
+    // See drawGlyph.
+    // Ex: with shadows it's 6 glyps per char. 30 fits exactly 5 chars.
     static constexpr S32 GLYPH_BATCH_SIZE = 30;
     static thread_local LLVector4a vertices[GLYPH_BATCH_SIZE * 6];
     static thread_local LLVector2 uvs[GLYPH_BATCH_SIZE * 6];
@@ -282,6 +297,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
 
     std::pair<EFontGlyphType, S32> bitmap_entry = std::make_pair(EFontGlyphType::Grayscale, -1);
     S32 glyph_count = 0;
+    llwchar last_char = wstr[begin_offset];
     for (i = begin_offset; i < begin_offset + length; i++)
     {
         llwchar wch = wstr[i];
@@ -299,7 +315,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
         }
         // Per-glyph bitmap texture.
         std::pair<EFontGlyphType, S32> next_bitmap_entry = fgi->mBitmapEntry;
-        if (next_bitmap_entry != bitmap_entry)
+        if (next_bitmap_entry != bitmap_entry || last_char != wch)
         {
             // Actually draw the queued glyphs before switching their texture;
             // otherwise the queued glyphs will be taken from wrong textures.
@@ -316,6 +332,11 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
             bitmap_entry = next_bitmap_entry;
             LLImageGL* font_image = font_bitmap_cache->getImageGL(bitmap_entry.first, bitmap_entry.second);
             gGL.getTexUnit(0)->bind(font_image);
+
+            // For some reason it's not enough to compare by bitmap_entry.
+            // Issue hits emojis, japenese and chinese glyphs, only on first run.
+            // Todo: figure it out, there might be a bug with raw image data.
+            last_char = wch;
         }
 
         if ((start_x + scaled_max_pixels) < (cur_x + fgi->mXBearing + fgi->mWidth))
diff --git a/indra/llrender/llfontgl.h b/indra/llrender/llfontgl.h
index 4bb6c55c65..1c8e036f58 100644
--- a/indra/llrender/llfontgl.h
+++ b/indra/llrender/llfontgl.h
@@ -90,6 +90,7 @@ public:
     bool loadFace(const std::string& filename, F32 point_size, const F32 vert_dpi, const F32 horz_dpi, bool is_fallback, S32 face_n);
 
     S32 getNumFaces(const std::string& filename);
+    S32 getCacheGeneration() const;
 
     S32 render(const LLWString &text, S32 begin_offset,
                 const LLRect& rect,
@@ -224,6 +225,7 @@ public:
     static F32 sHorizDPI;
     static F32 sScaleX;
     static F32 sScaleY;
+    static S32 sResolutionGeneration;
     static bool sDisplayFont ;
     static std::string sAppDir;         // For loading fonts
 
diff --git a/indra/llrender/llfontvertexbuffer.cpp b/indra/llrender/llfontvertexbuffer.cpp
index 5bd1ca5eed..a223509d30 100644
--- a/indra/llrender/llfontvertexbuffer.cpp
+++ b/indra/llrender/llfontvertexbuffer.cpp
@@ -146,7 +146,9 @@ S32 LLFontVertexBuffer::render(
              || mLastScaleY != LLFontGL::sScaleY
              || mLastVertDPI != LLFontGL::sVertDPI
              || mLastHorizDPI != LLFontGL::sHorizDPI
-             || mLastOrigin != LLFontGL::sCurOrigin)
+             || mLastOrigin != LLFontGL::sCurOrigin
+             || mLastResGeneration != LLFontGL::sResolutionGeneration
+             || mLastFontCacheGen != fontp->getCacheGeneration())
     {
         genBuffers(fontp, text, begin_offset, x, y, color, halign, valign,
             style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color);
@@ -178,6 +180,9 @@ void LLFontVertexBuffer::genBuffers(
 {
     // todo: add a debug build assert if this triggers too often for to long?
     mBufferList.clear();
+    // Save before rendreing, it can change mid-render,
+    // so will need to rerender previous characters
+    mLastFontCacheGen = fontp->getCacheGeneration();
 
     gGL.beginList(&mBufferList);
     mChars = fontp->render(text, begin_offset, x, y, color, halign, valign,
@@ -201,6 +206,7 @@ void LLFontVertexBuffer::genBuffers(
     mLastVertDPI = LLFontGL::sVertDPI;
     mLastHorizDPI = LLFontGL::sHorizDPI;
     mLastOrigin = LLFontGL::sCurOrigin;
+    mLastResGeneration = LLFontGL::sResolutionGeneration;
 
     if (right_x)
     {
diff --git a/indra/llrender/llfontvertexbuffer.h b/indra/llrender/llfontvertexbuffer.h
index af195dfff9..a9e1e2337c 100644
--- a/indra/llrender/llfontvertexbuffer.h
+++ b/indra/llrender/llfontvertexbuffer.h
@@ -117,8 +117,13 @@ private:
     F32 mLastScaleY = 1.f;
     F32 mLastVertDPI = 0.f;
     F32 mLastHorizDPI = 0.f;
+    S32 mLastResGeneration = 0;
     LLCoordGL mLastOrigin;
 
+    // Adding new characters to bitmap cache can alter value from getBitmapWidth();
+    // which alters whole string. So rerender when new characters were added to cache.
+    S32 mLastFontCacheGen = 0;
+
     static bool sEnableBufferCollection;
 };
 
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 1bc5e79835..0be27b8975 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -1228,28 +1228,9 @@ bool LLGLManager::initGL()
     }
 #endif
 
-#if LL_WINDOWS
-    if (mVRAM < 256)
-    {
-        // Something likely went wrong using the above extensions
-        // try WMI first and fall back to old method (from dxdiag) if all else fails
-        // Function will check all GPUs WMI knows of and will pick up the one with most
-        // memory. We need to check all GPUs because system can switch active GPU to
-        // weaker one, to preserve power when not under load.
-        U32 mem = LLDXHardware::getMBVideoMemoryViaWMI();
-        if (mem != 0)
-        {
-            mVRAM = mem;
-            LL_WARNS("RenderInit") << "VRAM Detected (WMI):" << mVRAM<< LL_ENDL;
-        }
-    }
-#endif
-
     if (mVRAM < 256 && old_vram > 0)
     {
         // fall back to old method
-        // Note: on Windows value will be from LLDXHardware.
-        // Either received via dxdiag or via WMI by id from dxdiag.
         mVRAM = old_vram;
     }
 
@@ -2458,12 +2439,15 @@ void LLGLState::checkStates(GLboolean writeAlpha)
         return;
     }
 
-    GLint src;
-    GLint dst;
-    glGetIntegerv(GL_BLEND_SRC, &src);
-    glGetIntegerv(GL_BLEND_DST, &dst);
-    llassert_always(src == GL_SRC_ALPHA);
-    llassert_always(dst == GL_ONE_MINUS_SRC_ALPHA);
+    GLint srcRGB, dstRGB, srcAlpha, dstAlpha;
+    glGetIntegerv(GL_BLEND_SRC_RGB, &srcRGB);
+    glGetIntegerv(GL_BLEND_DST_RGB, &dstRGB);
+    glGetIntegerv(GL_BLEND_SRC_ALPHA, &srcAlpha);
+    glGetIntegerv(GL_BLEND_DST_ALPHA, &dstAlpha);
+    llassert_always(srcRGB == GL_SRC_ALPHA);
+    llassert_always(srcAlpha == GL_SRC_ALPHA);
+    llassert_always(dstRGB == GL_ONE_MINUS_SRC_ALPHA);
+    llassert_always(dstAlpha == GL_ONE_MINUS_SRC_ALPHA);
 
     // disable for now until usage is consistent
     //GLboolean colorMask[4];
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index b3f32fdc83..3735a99109 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -1076,8 +1076,8 @@ void LLGLSLShader::bind()
 
 void LLGLSLShader::bind(U8 variant)
 {
-    llassert(mGLTFVariants.size() == LLGLSLShader::NUM_GLTF_VARIANTS);
-    llassert(variant < LLGLSLShader::NUM_GLTF_VARIANTS);
+    llassert_always(mGLTFVariants.size() == LLGLSLShader::NUM_GLTF_VARIANTS);
+    llassert_always(variant < LLGLSLShader::NUM_GLTF_VARIANTS);
     mGLTFVariants[variant].bind();
 }
 
@@ -1085,7 +1085,7 @@ void LLGLSLShader::bind(bool rigged)
 {
     if (rigged)
     {
-        llassert(mRiggedVariant);
+        llassert_always(mRiggedVariant);
         mRiggedVariant->bind();
     }
     else
@@ -1247,23 +1247,40 @@ S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode)
         llassert(false);
         return -1;
     }
+
     S32 index = mTexture[uniform];
-    if (index != -1 && gGL.getTexUnit(index)->getCurrType() != LLTexUnit::TT_NONE)
+    if (index < 0)
+    {
+        // Invalid texture index - nothing to disable
+        return index;
+    }
+
+    LLTexUnit* tex_unit = gGL.getTexUnit(index);
+    if (!tex_unit)
     {
-        if (gDebugGL && gGL.getTexUnit(index)->getCurrType() != mode)
+        // Invalid texture unit
+        LL_WARNS_ONCE("Shader") << "Invalid texture unit at index: " << index << LL_ENDL;
+        return index;
+    }
+
+    LLTexUnit::eTextureType curr_type = tex_unit->getCurrType();
+    if (curr_type != LLTexUnit::TT_NONE)
+    {
+        if (gDebugGL && curr_type != mode)
         {
             if (gDebugSession)
             {
-                gFailLog << "Texture channel " << index << " texture type corrupted." << std::endl;
+                gFailLog << "Texture channel " << index << " texture type corrupted. Expected: " << mode << ", Found: " << curr_type << std::endl;
                 ll_fail("LLGLSLShader::disableTexture failed");
             }
             else
             {
-                LL_ERRS() << "Texture channel " << index << " texture type corrupted." << LL_ENDL;
+                LL_ERRS() << "Texture channel " << index << " texture type corrupted. Expected: " << mode << ", Found: " << curr_type << LL_ENDL;
             }
         }
-        gGL.getTexUnit(index)->disable();
+        tex_unit->disable();
     }
+
     return index;
 }
 
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index 58c456f134..873ab0cff5 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -59,6 +59,7 @@ public:
     bool attachNothing = false;
     bool hasHeroProbes = false;
     bool isPBRTerrain = false;
+    bool hasTonemap = false;
 };
 
 // ============= Structure for caching shader uniforms ===============
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 0bfcb5d9d2..3f8903ca09 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -330,6 +330,7 @@ S32 LLImageGL::dataFormatBits(S32 dataformat)
     case GL_RGB:                                    return 24;
     case GL_SRGB:                                   return 24;
     case GL_RGB8:                                   return 24;
+    case GL_R11F_G11F_B10F:                         return 32;
     case GL_RGBA:                                   return 32;
     case GL_RGBA8:                                  return 32;
     case GL_RGB10_A2:                               return 32;
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 1d53850f74..1dc87a66ce 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -737,9 +737,8 @@ void LLLightState::setPosition(const LLVector4& position)
     ++gGL.mLightHash;
     mPosition = position;
     //transform position by current modelview matrix
-    glm::vec4 pos(glm::make_vec4(position.mV));
-    const glm::mat4& mat = gGL.getModelviewMatrix();
-    pos = mat * pos;
+    glm::vec4 pos(position);
+    pos = gGL.getModelviewMatrix() * pos;
     mPosition.set(glm::value_ptr(pos));
 }
 
@@ -794,7 +793,7 @@ void LLLightState::setSpotDirection(const LLVector3& direction)
     ++gGL.mLightHash;
 
     //transform direction by current modelview matrix
-    glm::vec3 dir(glm::make_vec3(direction.mV));
+    glm::vec3 dir(direction);
     const glm::mat3 mat(gGL.getModelviewMatrix());
     dir = mat * dir;
 
@@ -2088,12 +2087,14 @@ void set_last_projection(const glm::mat4& mat)
 
 glm::vec3 mul_mat4_vec3(const glm::mat4& mat, const glm::vec3& vec)
 {
-    //const float w = vec[0] * mat[0][3] + vec[1] * mat[1][3] + vec[2] * mat[2][3] + mat[3][3];
-    //return glm::vec3(
-    //    (vec[0] * mat[0][0] + vec[1] * mat[1][0] + vec[2] * mat[2][0] + mat[3][0]) / w,
-    //    (vec[0] * mat[0][1] + vec[1] * mat[1][1] + vec[2] * mat[2][1] + mat[3][1]) / w,
-    //    (vec[0] * mat[0][2] + vec[1] * mat[1][2] + vec[2] * mat[2][2] + mat[3][2]) / w
-    //);
+#if 1 // SIMD path results in strange crashes. Fall back to scalar for now.
+    const float w = vec[0] * mat[0][3] + vec[1] * mat[1][3] + vec[2] * mat[2][3] + mat[3][3];
+    return glm::vec3(
+       (vec[0] * mat[0][0] + vec[1] * mat[1][0] + vec[2] * mat[2][0] + mat[3][0]) / w,
+       (vec[0] * mat[0][1] + vec[1] * mat[1][1] + vec[2] * mat[2][1] + mat[3][1]) / w,
+       (vec[0] * mat[0][2] + vec[1] * mat[1][2] + vec[2] * mat[2][2] + mat[3][2]) / w
+    );
+#else
     LLVector4a x, y, z, s, t, p, q;
 
     x.splat(vec.x);
@@ -2123,4 +2124,5 @@ glm::vec3 mul_mat4_vec3(const glm::mat4& mat, const glm::vec3& vec)
     res.setAdd(x, z);
     res.div(q);
     return glm::make_vec3(res.getF32ptr());
+#endif
 }
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 0885740934..4807c12226 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -291,6 +291,14 @@ bool LLShaderMgr::attachShaderFeatures(LLGLSLShader * shader)
         }
     }
 
+    if (features->hasTonemap)
+    {
+        if (!shader->attachFragmentObject("deferred/tonemapUtilF.glsl"))
+        {
+            return false;
+        }
+    }
+
     // NOTE order of shader object attaching is VERY IMPORTANT!!!
     if (features->hasAtmospherics)
     {
@@ -466,6 +474,7 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev
 
     if (filename.empty())
     {
+        LL_WARNS("ShaderLoading") << "tried loading empty filename" << LL_ENDL;
         return 0;
     }
 
@@ -923,6 +932,8 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev
         }
         LL_WARNS("ShaderLoading") << "Failed to load " << filename << LL_ENDL;
     }
+
+    LL_DEBUGS("ShaderLoading") << "loadShaderFile() completed, ret: " << U32(ret) << LL_ENDL;
     return ret;
 }
 
@@ -1389,6 +1400,7 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("screenTex");
     mReservedUniforms.push_back("screenDepth");
     mReservedUniforms.push_back("refTex");
+    mReservedUniforms.push_back("exclusionTex");
     mReservedUniforms.push_back("eyeVec");
     mReservedUniforms.push_back("time");
     mReservedUniforms.push_back("waveDir1");
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index 34bd73a42e..46788841a5 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -36,6 +36,8 @@ public:
     LLShaderMgr();
     virtual ~LLShaderMgr();
 
+    // Note: although you can use statically hashed strings to just bind a random uniform, it's generally preferably that you use this.
+    // Always document what the actual shader uniform is next to the shader uniform in this struct.
     // clang-format off
     typedef enum
     {                                       // Shader uniform name, set in LLShaderMgr::initAttribsAndUniforms()
@@ -234,6 +236,7 @@ public:
         WATER_SCREENTEX,                    //  "screenTex"
         WATER_SCREENDEPTH,                  //  "screenDepth"
         WATER_REFTEX,                       //  "refTex"
+        WATER_EXCLUSIONTEX,                 //  "exclusionTex"
         WATER_EYEVEC,                       //  "eyeVec"
         WATER_TIME,                         //  "time"
         WATER_WAVE_DIR1,                    //  "waveDir1"