Merge pull request #2513 from secondlife/lua-merge-dev

Merge develop branch into Lua project branch.

29 files changed, 1137 insertions, 537 deletions

diff --git a/indra/llrender/CMakeLists.txt b/indra/llrender/CMakeLists.txt
index 7f881c8bb3..ccff7c7a8c 100644
--- a/indra/llrender/CMakeLists.txt
+++ b/indra/llrender/CMakeLists.txt
@@ -17,6 +17,7 @@ set(llrender_SOURCE_FILES
     llfontfreetype.cpp
     llfontfreetypesvg.cpp
     llfontgl.cpp
+    llfontvertexbuffer.cpp
     llfontregistry.cpp
     llgl.cpp
     llglslshader.cpp
@@ -43,6 +44,7 @@ set(llrender_HEADER_FILES
     llcubemap.h
     llcubemaparray.h
     llfontgl.h
+    llfontvertexbuffer.h
     llfontfreetype.h
     llfontfreetypesvg.h
     llfontbitmapcache.h
diff --git a/indra/llrender/llcubemaparray.cpp b/indra/llrender/llcubemaparray.cpp
index be69b997da..4f5e13765a 100644
--- a/indra/llrender/llcubemaparray.cpp
+++ b/indra/llrender/llcubemaparray.cpp
@@ -125,27 +125,25 @@ void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool us
     mImage->setHasMipMaps(use_mips);
 
     bind(0);
+    free_cur_tex_image();
 
     U32 format = components == 4 ? GL_RGBA16F : GL_RGB16F;
-
     U32 mip = 0;
-
-    free_cur_tex_image();
-
-    while (resolution >= 1)
+    U32 mip_resolution = resolution;
+    while (mip_resolution >= 1)
     {
-        glTexImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, mip, format, resolution, resolution, count * 6, 0,
+        glTexImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, mip, format, mip_resolution, mip_resolution, count * 6, 0,
             GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
 
         if (!use_mips)
         {
             break;
         }
-        resolution /= 2;
+        mip_resolution /= 2;
         ++mip;
     }
 
-    alloc_tex_image(resolution * 6, resolution, format);
+    alloc_tex_image(resolution, resolution, format, count * 6);
 
     mImage->setAddressMode(LLTexUnit::TAM_CLAMP);
 
diff --git a/indra/llrender/llfontbitmapcache.cpp b/indra/llrender/llfontbitmapcache.cpp
index 46c2e89797..ee9cfd0719 100644
--- a/indra/llrender/llfontbitmapcache.cpp
+++ b/indra/llrender/llfontbitmapcache.cpp
@@ -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));
+            mImageGLVec[bitmap_idx].push_back(new LLImageGL(image_raw, false, false));
             LLImageGL* image_gl = getImageGL(bitmap_type, bitmap_num);
 
             // Start at beginning of the new image.
diff --git a/indra/llrender/llfontfreetype.cpp b/indra/llrender/llfontfreetype.cpp
index 741ed993b0..6128e03fa7 100644
--- a/indra/llrender/llfontfreetype.cpp
+++ b/indra/llrender/llfontfreetype.cpp
@@ -178,7 +178,7 @@ unsigned long ft_read_cb(FT_Stream stream, unsigned long offset, unsigned char *
     llifstream *file_stream = static_cast<llifstream *>(stream->descriptor.pointer);
     file_stream->seekg(offset, std::ios::beg);
     file_stream->read((char*)buffer, count);
-    return file_stream->gcount();
+    return (unsigned long)file_stream->gcount();
 }
 
 void ft_close_cb(FT_Stream stream) {
diff --git a/indra/llrender/llfontfreetypesvg.cpp b/indra/llrender/llfontfreetypesvg.cpp
index 355e8432aa..71f751329e 100644
--- a/indra/llrender/llfontfreetypesvg.cpp
+++ b/indra/llrender/llfontfreetypesvg.cpp
@@ -136,18 +136,18 @@ FT_Error LLFontFreeTypeSvgRenderer::OnPresetGlypthSlot(FT_GlyphSlot glyph_slot,
     float svg_scale = llmin(svg_x_scale, svg_y_scale);
     datap->Scale = svg_scale;
 
-    glyph_slot->bitmap.width = floorf(svg_width) * svg_scale;
-    glyph_slot->bitmap.rows = floorf(svg_height) * svg_scale;
+    glyph_slot->bitmap.width = (unsigned int)(floorf(svg_width) * svg_scale);
+    glyph_slot->bitmap.rows = (unsigned int)(floorf(svg_height) * svg_scale);
     glyph_slot->bitmap_left = (document->metrics.x_ppem - glyph_slot->bitmap.width) / 2;
-    glyph_slot->bitmap_top = glyph_slot->face->size->metrics.ascender / 64.f;
+    glyph_slot->bitmap_top = (FT_Int)(glyph_slot->face->size->metrics.ascender / 64.f);
     glyph_slot->bitmap.pitch = glyph_slot->bitmap.width * 4;
     glyph_slot->bitmap.pixel_mode = FT_PIXEL_MODE_BGRA;
 
     /* Copied as-is from fcft (MIT license) */
 
     // Compute all the bearings and set them correctly. The outline is scaled already, we just need to use the bounding box.
-    float horiBearingX = 0.;
-    float horiBearingY = -glyph_slot->bitmap_top;
+    float horiBearingX = 0.f;
+    float horiBearingY = -(float)glyph_slot->bitmap_top;
 
     // XXX parentheses correct?
     float vertBearingX = glyph_slot->metrics.horiBearingX / 64.0f - glyph_slot->metrics.horiAdvance / 64.0f / 2;
@@ -156,13 +156,13 @@ FT_Error LLFontFreeTypeSvgRenderer::OnPresetGlypthSlot(FT_GlyphSlot glyph_slot,
     // Do conversion in two steps to avoid 'bad function cast' warning
     glyph_slot->metrics.width = glyph_slot->bitmap.width * 64;
     glyph_slot->metrics.height = glyph_slot->bitmap.rows * 64;
-    glyph_slot->metrics.horiBearingX = horiBearingX * 64;
-    glyph_slot->metrics.horiBearingY = horiBearingY * 64;
-    glyph_slot->metrics.vertBearingX = vertBearingX * 64;
-    glyph_slot->metrics.vertBearingY = vertBearingY * 64;
+    glyph_slot->metrics.horiBearingX = (FT_Pos)(horiBearingX * 64);
+    glyph_slot->metrics.horiBearingY = (FT_Pos)(horiBearingY * 64);
+    glyph_slot->metrics.vertBearingX = (FT_Pos)(vertBearingX * 64);
+    glyph_slot->metrics.vertBearingY = (FT_Pos)(vertBearingY * 64);
     if (glyph_slot->metrics.vertAdvance == 0)
     {
-        glyph_slot->metrics.vertAdvance = glyph_slot->bitmap.rows * 1.2f * 64;
+        glyph_slot->metrics.vertAdvance = (FT_Pos)(glyph_slot->bitmap.rows * 1.2f * 64);
     }
 
     return FT_Err_Ok;
diff --git a/indra/llrender/llfontgl.cpp b/indra/llrender/llfontgl.cpp
index 59ee8ef84f..9721b020c7 100644
--- a/indra/llrender/llfontgl.cpp
+++ b/indra/llrender/llfontgl.cpp
@@ -112,7 +112,7 @@ S32 LLFontGL::getNumFaces(const std::string& filename)
 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
 {
-    LLRectf rect_float(rect.mLeft, rect.mTop, rect.mRight, rect.mBottom);
+    LLRectf rect_float((F32)rect.mLeft, (F32)rect.mTop, (F32)rect.mRight, (F32)rect.mBottom);
     return render(wstr, begin_offset, rect_float, color, halign, valign, style, shadow, max_chars, right_x, use_ellipses, use_color);
 }
 
@@ -138,7 +138,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, const LLRectf& rec
         y = rect.mBottom;
         break;
     }
-    return render(wstr, begin_offset, x, y, color, halign, valign, style, shadow, max_chars, rect.getWidth(), right_x, use_ellipses, use_color);
+    return render(wstr, begin_offset, x, y, color, halign, valign, style, shadow, max_chars, (S32)rect.getWidth(), right_x, use_ellipses, use_color);
 }
 
 
@@ -270,10 +270,10 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
 
     const LLFontGlyphInfo* next_glyph = NULL;
 
-    const S32 GLYPH_BATCH_SIZE = 30;
-    LLVector3 vertices[GLYPH_BATCH_SIZE * 4];
-    LLVector2 uvs[GLYPH_BATCH_SIZE * 4];
-    LLColor4U colors[GLYPH_BATCH_SIZE * 4];
+    static constexpr S32 GLYPH_BATCH_SIZE = 30;
+    static thread_local LLVector3 vertices[GLYPH_BATCH_SIZE * 4];
+    static thread_local LLVector2 uvs[GLYPH_BATCH_SIZE * 4];
+    static thread_local LLColor4U colors[GLYPH_BATCH_SIZE * 4];
 
     LLColor4U text_color(color);
     // Preserve the transparency to render fading emojis in fading text (e.g.
@@ -402,11 +402,11 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons
 
     if (draw_ellipses)
     {
-
         // recursively render ellipses at end of string
         // we've already reserved enough room
         gGL.pushUIMatrix();
-        renderUTF8(std::string("..."),
+        static LLWString elipses_wstr(utf8string_to_wstring(std::string("...")));
+        render(elipses_wstr,
                 0,
                 (cur_x - origin.mV[VX]) / sScaleX, (F32)y,
                 color,
@@ -503,6 +503,7 @@ F32 LLFontGL::getWidthF32(const std::string& utf8text, S32 begin_offset, S32 max
 
 F32 LLFontGL::getWidthF32(const llwchar* wchars, S32 begin_offset, S32 max_chars, bool no_padding) const
 {
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI;
     const S32 LAST_CHARACTER = LLFontFreetype::LAST_CHAR_FULL;
 
     F32 cur_x = 0;
@@ -560,7 +561,7 @@ F32 LLFontGL::getWidthF32(const llwchar* wchars, S32 begin_offset, S32 max_chars
 
 void LLFontGL::generateASCIIglyphs()
 {
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI;
     for (U32 i = 32; (i < 127); i++)
     {
         mFontFreetype->getGlyphInfo(i, EFontGlyphType::Grayscale);
@@ -570,7 +571,7 @@ void LLFontGL::generateASCIIglyphs()
 // Returns the max number of complete characters from text (up to max_chars) that can be drawn in max_pixels
 S32 LLFontGL::maxDrawableChars(const llwchar* wchars, F32 max_pixels, S32 max_chars, EWordWrapStyle end_on_word_boundary) const
 {
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI;
     if (!wchars || !wchars[0] || max_chars == 0)
     {
         return 0;
@@ -881,7 +882,7 @@ void LLFontGL::dumpFontTextures()
 // static
 bool LLFontGL::loadDefaultFonts()
 {
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI;
     bool succ = true;
     succ &= (NULL != getFontSansSerifSmall());
     succ &= (NULL != getFontSansSerif());
@@ -894,7 +895,7 @@ bool LLFontGL::loadDefaultFonts()
 
 void LLFontGL::loadCommonFonts()
 {
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_UI;
     getFont(LLFontDescriptor("SansSerif", "Small", BOLD));
     getFont(LLFontDescriptor("SansSerif", "Large", BOLD));
     getFont(LLFontDescriptor("SansSerif", "Huge", BOLD));
diff --git a/indra/llrender/llfontregistry.cpp b/indra/llrender/llfontregistry.cpp
index 62f4f35e3d..c48a389f6a 100644
--- a/indra/llrender/llfontregistry.cpp
+++ b/indra/llrender/llfontregistry.cpp
@@ -500,7 +500,7 @@ LLFontGL *LLFontRegistry::createFont(const LLFontDescriptor& desc)
         // *HACK: Fallback fonts don't render, so we can use that to suppress
         // creation of OpenGL textures for test apps. JC
         bool is_fallback = !is_first_found || !mCreateGLTextures;
-        F32 extra_scale = (is_fallback)?fallback_scale:1.0;
+        F32 extra_scale = (is_fallback) ? fallback_scale : 1.0f;
         F32 point_size_scale = extra_scale * point_size;
         bool is_font_loaded = false;
         for(string_vec_t::iterator font_search_path_it = font_search_paths.begin();
diff --git a/indra/llrender/llfontvertexbuffer.cpp b/indra/llrender/llfontvertexbuffer.cpp
new file mode 100644
index 0000000000..392f235aad
--- /dev/null
+++ b/indra/llrender/llfontvertexbuffer.cpp
@@ -0,0 +1,211 @@
+/**
+ * @file llfontvertexbuffer.cpp
+ * @brief Buffer storage for font rendering.
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "llfontvertexbuffer.h"
+
+#include "llvertexbuffer.h"
+
+
+LLFontVertexBuffer::LLFontVertexBuffer()
+{
+}
+
+LLFontVertexBuffer::~LLFontVertexBuffer()
+{
+    reset();
+}
+
+void LLFontVertexBuffer::reset()
+{
+    // Todo: some form of debug only frequecy check&assert to see if this is happening too often.
+    // Regenerating this list is expensive
+    mBufferList.clear();
+}
+
+S32 LLFontVertexBuffer::render(
+    const LLFontGL* fontp,
+    const LLWString& text,
+    S32 begin_offset,
+    LLRect rect,
+    const LLColor4& color,
+    LLFontGL::HAlign halign, LLFontGL::VAlign valign,
+    U8 style,
+    LLFontGL::ShadowType shadow,
+    S32 max_chars, S32 max_pixels,
+    F32* right_x,
+    bool use_ellipses,
+    bool use_color)
+{
+    LLRectf rect_float((F32)rect.mLeft, (F32)rect.mTop, (F32)rect.mRight, (F32)rect.mBottom);
+    return render(fontp, text, begin_offset, rect_float, color, halign, valign, style, shadow, max_chars, right_x, use_ellipses, use_color);
+}
+
+S32 LLFontVertexBuffer::render(
+    const LLFontGL* fontp,
+    const LLWString& text,
+    S32 begin_offset,
+    LLRectf rect,
+    const LLColor4& color,
+    LLFontGL::HAlign halign, LLFontGL::VAlign valign,
+    U8 style,
+    LLFontGL::ShadowType shadow,
+    S32 max_chars,
+    F32* right_x,
+    bool use_ellipses,
+    bool use_color)
+{
+    F32 x = rect.mLeft;
+    F32 y = 0.f;
+
+    switch (valign)
+    {
+    case LLFontGL::TOP:
+        y = rect.mTop;
+        break;
+    case LLFontGL::VCENTER:
+        y = rect.getCenterY();
+        break;
+    case LLFontGL::BASELINE:
+    case LLFontGL::BOTTOM:
+        y = rect.mBottom;
+        break;
+    default:
+        y = rect.mBottom;
+        break;
+    }
+    return render(fontp, text, begin_offset, x, y, color, halign, valign, style, shadow, max_chars, (S32)rect.getWidth(), right_x, use_ellipses, use_color);
+}
+
+S32 LLFontVertexBuffer::render(
+    const LLFontGL* fontp,
+    const LLWString& text,
+    S32 begin_offset,
+    F32 x, F32 y,
+    const LLColor4& color,
+    LLFontGL::HAlign halign, LLFontGL::VAlign valign,
+    U8 style,
+    LLFontGL::ShadowType shadow,
+    S32 max_chars , S32 max_pixels,
+    F32* right_x,
+    bool use_ellipses,
+    bool use_color )
+{
+    if (!LLFontGL::sDisplayFont) //do not display texts
+    {
+        return static_cast<S32>(text.length());
+    }
+    if (mBufferList.empty())
+    {
+        genBuffers(fontp, text, begin_offset, x, y, color, halign, valign,
+            style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color);
+    }
+    else if (mLastX != x
+             || mLastY != y
+             || mLastFont != fontp
+             || mLastColor != color // alphas change often
+             || mLastHalign != halign
+             || mLastValign != valign
+             || mLastOffset != begin_offset
+             || mLastMaxChars != max_chars
+             || mLastMaxPixels != max_pixels
+             || mLastStyle != style
+             || mLastShadow != shadow // ex: buttons change shadow state
+             || mLastScaleX != LLFontGL::sScaleX
+             || mLastScaleY != LLFontGL::sScaleY
+             || mLastOrigin != LLFontGL::sCurOrigin)
+    {
+        genBuffers(fontp, text, begin_offset, x, y, color, halign, valign,
+            style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color);
+    }
+    else
+    {
+        renderBuffers();
+
+        if (right_x)
+        {
+            *right_x = mLastRightX;
+        }
+    }
+    return mChars;
+}
+
+void LLFontVertexBuffer::genBuffers(
+    const LLFontGL* fontp,
+    const LLWString& text,
+    S32 begin_offset,
+    F32 x, F32 y,
+    const LLColor4& color,
+    LLFontGL::HAlign halign, LLFontGL::VAlign valign,
+    U8 style, LLFontGL::ShadowType shadow,
+    S32 max_chars, S32 max_pixels,
+    F32* right_x,
+    bool use_ellipses,
+    bool use_color)
+{
+    // todo: add a debug build assert if this triggers too often for to long?
+    mBufferList.clear();
+
+    gGL.beginList(&mBufferList);
+    mChars = fontp->render(text, begin_offset, x, y, color, halign, valign,
+        style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color);
+    gGL.endList();
+
+    mLastFont = fontp;
+    mLastOffset = begin_offset;
+    mLastMaxChars = max_chars;
+    mLastMaxPixels = max_pixels;
+    mLastX = x;
+    mLastY = y;
+    mLastColor = color;
+    mLastHalign = halign;
+    mLastValign = valign;
+    mLastStyle = style;
+    mLastShadow = shadow;
+
+    mLastScaleX = LLFontGL::sScaleX;
+    mLastScaleY = LLFontGL::sScaleY;
+    mLastOrigin = LLFontGL::sCurOrigin;
+
+    if (right_x)
+    {
+        mLastRightX = *right_x;
+    }
+}
+
+void LLFontVertexBuffer::renderBuffers()
+{
+    gGL.flush(); // deliberately empty pending verts
+    gGL.getTexUnit(0)->enable(LLTexUnit::TT_TEXTURE);
+    gGL.pushUIMatrix();
+    for (LLVertexBufferData& buffer : mBufferList)
+    {
+        buffer.draw();
+    }
+    gGL.popUIMatrix();
+}
+
diff --git a/indra/llrender/llfontvertexbuffer.h b/indra/llrender/llfontvertexbuffer.h
new file mode 100644
index 0000000000..67cf2ca13c
--- /dev/null
+++ b/indra/llrender/llfontvertexbuffer.h
@@ -0,0 +1,119 @@
+/**
+ * @file llfontgl.h
+ * @author Andrii Kleshchev
+ * @brief Buffer storage for font rendering.
+ *
+ * $LicenseInfo:firstyear=2001&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2010, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLFONTVERTEXBUFFER_H
+#define LL_LLFONTVERTEXBUFFER_H
+
+#include "llfontgl.h"
+
+class LLVertexBufferData;
+
+class LLFontVertexBuffer
+{
+public:
+    LLFontVertexBuffer();
+    ~LLFontVertexBuffer();
+
+    void reset();
+
+    S32 render(const LLFontGL* fontp,
+        const LLWString& text,
+        S32 begin_offset,
+        LLRect rect,
+        const LLColor4& color,
+        LLFontGL::HAlign halign = LLFontGL::LEFT, LLFontGL::VAlign valign = LLFontGL::BASELINE,
+        U8 style = LLFontGL::NORMAL,
+        LLFontGL::ShadowType shadow = LLFontGL::NO_SHADOW,
+        S32 max_chars = S32_MAX, S32 max_pixels = S32_MAX,
+        F32* right_x = NULL,
+        bool use_ellipses = false,
+        bool use_color = true);
+
+    S32 render(const LLFontGL* fontp,
+        const LLWString& text,
+        S32 begin_offset,
+        LLRectf rect,
+        const LLColor4& color,
+        LLFontGL::HAlign halign = LLFontGL::LEFT, LLFontGL::VAlign valign = LLFontGL::BASELINE,
+        U8 style = LLFontGL::NORMAL,
+        LLFontGL::ShadowType shadow = LLFontGL::NO_SHADOW,
+        S32 max_chars = S32_MAX,
+        F32* right_x = NULL,
+        bool use_ellipses = false,
+        bool use_color = true);
+
+    S32 render(const LLFontGL* fontp,
+        const LLWString& text,
+        S32 begin_offset,
+        F32 x, F32 y,
+        const LLColor4& color,
+        LLFontGL::HAlign halign = LLFontGL::LEFT, LLFontGL::VAlign valign = LLFontGL::BASELINE,
+        U8 style = LLFontGL::NORMAL,
+        LLFontGL::ShadowType shadow = LLFontGL::NO_SHADOW,
+        S32 max_chars = S32_MAX, S32 max_pixels = S32_MAX,
+        F32* right_x = NULL,
+        bool use_ellipses = false,
+        bool use_color = true);
+private:
+
+    void genBuffers(const LLFontGL* fontp,
+         const LLWString& text,
+         S32 begin_offset,
+         F32 x, F32 y,
+         const LLColor4& color,
+         LLFontGL::HAlign halign, LLFontGL::VAlign valign,
+         U8 style,
+        LLFontGL::ShadowType shadow,
+         S32 max_chars, S32 max_pixels,
+         F32* right_x,
+         bool use_ellipses,
+         bool use_color);
+
+    void renderBuffers();
+
+    std::list<LLVertexBufferData> mBufferList;
+    S32 mChars = 0;
+    const LLFontGL *mLastFont = nullptr;
+    S32 mLastOffset = 0;
+    S32 mLastMaxChars = 0;
+    S32 mLastMaxPixels = 0;
+    F32 mLastX = 0.f;
+    F32 mLastY = 0.f;
+    LLColor4 mLastColor;
+    LLFontGL::HAlign mLastHalign = LLFontGL::LEFT;
+    LLFontGL::VAlign mLastValign = LLFontGL::BASELINE;
+    U8 mLastStyle = LLFontGL::NORMAL;
+    LLFontGL::ShadowType mLastShadow = LLFontGL::NO_SHADOW;
+    F32 mLastRightX = 0.f;
+
+    // LLFontGL's statics
+    F32 mLastScaleX = 1.f;
+    F32 mLastScaleY = 1.f;
+    LLCoordGL mLastOrigin;
+};
+
+#endif
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 7959b3bb57..0db1e27b01 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -1038,6 +1038,7 @@ void LLGLManager::initWGL()
         GLH_EXT_NAME(wglGetGPUIDsAMD) = (PFNWGLGETGPUIDSAMDPROC)GLH_EXT_GET_PROC_ADDRESS("wglGetGPUIDsAMD");
         GLH_EXT_NAME(wglGetGPUInfoAMD) = (PFNWGLGETGPUINFOAMDPROC)GLH_EXT_GET_PROC_ADDRESS("wglGetGPUInfoAMD");
     }
+    mHasNVXGpuMemoryInfo = ExtensionExists("GL_NVX_gpu_memory_info", gGLHExts.mSysExts);
 
     if (ExtensionExists("WGL_EXT_swap_control", gGLHExts.mSysExts))
     {
@@ -1205,6 +1206,17 @@ bool LLGLManager::initGL()
             LL_WARNS("RenderInit") << "VRAM Detected (AMDAssociations):" << mVRAM << LL_ENDL;
         }
     }
+    else if (mHasNVXGpuMemoryInfo)
+    {
+        GLint mem_kb = 0;
+        glGetIntegerv(GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX, &mem_kb);
+        mVRAM = mem_kb / 1024;
+
+        if (mVRAM != 0)
+        {
+            LL_WARNS("RenderInit") << "VRAM Detected (NVXGpuMemoryInfo):" << mVRAM << LL_ENDL;
+        }
+    }
 #endif
 
 #if LL_WINDOWS
@@ -2742,7 +2754,7 @@ LLGLDepthTest::LLGLDepthTest(GLboolean depth_enabled, GLboolean write_enabled, G
 : mPrevDepthEnabled(sDepthEnabled), mPrevDepthFunc(sDepthFunc), mPrevWriteEnabled(sWriteEnabled)
 {
     stop_glerror();
-
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
     checkState();
 
     if (!depth_enabled)
@@ -2775,6 +2787,7 @@ LLGLDepthTest::LLGLDepthTest(GLboolean depth_enabled, GLboolean write_enabled, G
 
 LLGLDepthTest::~LLGLDepthTest()
 {
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
     checkState();
     if (sDepthEnabled != mPrevDepthEnabled )
     {
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 909dad2e85..cd1ba55b16 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -97,11 +97,15 @@ public:
 
     // Vendor-specific extensions
     bool mHasAMDAssociations = false;
+    bool mHasNVXGpuMemoryInfo = false;
 
     bool mIsAMD;
     bool mIsNVIDIA;
     bool mIsIntel;
 
+    // hints to the render pipe
+    U32 mDownScaleMethod = 0; // see settings.xml RenderDownScaleMethod
+
 #if LL_DARWIN
     // Needed to distinguish problem cards on older Macs that break with Materials
     bool mIsMobileGF;
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index 25e4a88f28..a157bfee21 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -190,7 +190,7 @@ void LLGLSLShader::dumpStats()
         tris_sec /= seconds;
 
         F32 pct_samples = (F32)((F64)mSamplesDrawn / (F64)sTotalSamplesDrawn) * 100.f;
-        F32 samples_sec = (F32)mSamplesDrawn / 1000000000.0;
+        F32 samples_sec = (F32)(mSamplesDrawn / 1000000000.0);
         samples_sec /= seconds;
 
         F32 pct_binds = (F32)mBinds / (F32)sTotalBinds * 100.f;
@@ -1087,17 +1087,17 @@ void LLGLSLShader::unbind(void)
     sCurBoundShaderPtr = NULL;
 }
 
-S32 LLGLSLShader::bindTexture(const std::string& uniform, LLTexture* texture, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace colorspace)
+S32 LLGLSLShader::bindTexture(const std::string& uniform, LLTexture* texture, LLTexUnit::eTextureType mode)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
 
     S32 channel = 0;
     channel = getUniformLocation(uniform);
 
-    return bindTexture(channel, texture, mode, colorspace);
+    return bindTexture(channel, texture, mode);
 }
 
-S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace colorspace)
+S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextureType mode)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
 
@@ -1113,7 +1113,6 @@ S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextu
     if (uniform > -1)
     {
         gGL.getTexUnit(uniform)->bindFast(texture);
-        gGL.getTexUnit(uniform)->setTextureColorSpace(colorspace);
     }
 
     return uniform;
@@ -1194,7 +1193,7 @@ S32 LLGLSLShader::getTextureChannel(S32 uniform) const
     return mTexture[uniform];
 }
 
-S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace space)
+S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
 
@@ -1211,12 +1210,11 @@ S32 LLGLSLShader::enableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTex
     {
         gGL.getTexUnit(index)->activate();
         gGL.getTexUnit(index)->enable(mode);
-        gGL.getTexUnit(index)->setTextureColorSpace(space);
     }
     return index;
 }
 
-S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTexUnit::eTextureColorSpace space)
+S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
 
@@ -1229,7 +1227,7 @@ S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTe
     S32 index = mTexture[uniform];
     if (index != -1 && gGL.getTexUnit(index)->getCurrType() != LLTexUnit::TT_NONE)
     {
-        if (gDebugGL && gGL.getTexUnit(index)->getCurrType() != mode && gGL.getTexUnit(index)->getCurrColorSpace() != space)
+        if (gDebugGL && gGL.getTexUnit(index)->getCurrType() != mode)
         {
             if (gDebugSession)
             {
@@ -1265,7 +1263,7 @@ void LLGLSLShader::uniform1i(U32 index, GLint x)
             if (iter == mValue.end() || iter->second.mV[0] != x)
             {
                 glUniform1i(mUniform[index], x);
-                mValue[mUniform[index]] = LLVector4(x, 0.f, 0.f, 0.f);
+                mValue[mUniform[index]] = LLVector4((F32)x, 0.f, 0.f, 0.f);
             }
         }
     }
@@ -1405,7 +1403,7 @@ void LLGLSLShader::uniform1iv(U32 index, U32 count, const GLint* v)
         if (mUniform[index] >= 0)
         {
             const auto& iter = mValue.find(mUniform[index]);
-            LLVector4 vec(v[0], 0.f, 0.f, 0.f);
+            LLVector4 vec((F32)v[0], 0.f, 0.f, 0.f);
             if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
             {
                 glUniform1iv(mUniform[index], count, v);
@@ -1432,7 +1430,7 @@ void LLGLSLShader::uniform4iv(U32 index, U32 count, const GLint* v)
         if (mUniform[index] >= 0)
         {
             const auto& iter = mValue.find(mUniform[index]);
-            LLVector4 vec(v[0], v[1], v[2], v[3]);
+            LLVector4 vec((F32)v[0], (F32)v[1], (F32)v[2], (F32)v[3]);
             if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
             {
                 glUniform1iv(mUniform[index], count, v);
@@ -1552,6 +1550,34 @@ void LLGLSLShader::uniform4fv(U32 index, U32 count, const GLfloat* v)
     }
 }
 
+void LLGLSLShader::uniform4uiv(U32 index, U32 count, const GLuint* v)
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
+    llassert(sCurBoundShaderPtr == this);
+
+    if (mProgramObject)
+    {
+        if (mUniform.size() <= index)
+        {
+            LL_WARNS_ONCE("Shader") << "Uniform index out of bounds. Size: " << (S32)mUniform.size() << " index: " << index << LL_ENDL;
+            llassert(false);
+            return;
+        }
+
+        if (mUniform[index] >= 0)
+        {
+            const auto& iter = mValue.find(mUniform[index]);
+            LLVector4 vec((F32)v[0], (F32)v[1], (F32)v[2], (F32)v[3]);
+            if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
+            {
+                LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
+                glUniform4uiv(mUniform[index], count, v);
+                mValue[mUniform[index]] = vec;
+            }
+        }
+    }
+}
+
 void LLGLSLShader::uniformMatrix2fv(U32 index, U32 count, GLboolean transpose, const GLfloat* v)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
@@ -1702,7 +1728,7 @@ void LLGLSLShader::uniform1i(const LLStaticHashedString& uniform, GLint v)
     if (location >= 0)
     {
         const auto& iter = mValue.find(location);
-        LLVector4 vec(v, 0.f, 0.f, 0.f);
+        LLVector4 vec((F32)v, 0.f, 0.f, 0.f);
         if (iter == mValue.end() || shouldChange(iter->second, vec))
         {
             glUniform1i(location, v);
@@ -1718,7 +1744,7 @@ void LLGLSLShader::uniform1iv(const LLStaticHashedString& uniform, U32 count, co
 
     if (location >= 0)
     {
-        LLVector4 vec(v[0], 0, 0, 0);
+        LLVector4 vec((F32)v[0], 0.f, 0.f, 0.f);
         const auto& iter = mValue.find(location);
         if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
         {
@@ -1736,7 +1762,7 @@ void LLGLSLShader::uniform4iv(const LLStaticHashedString& uniform, U32 count, co
 
     if (location >= 0)
     {
-        LLVector4 vec(v[0], v[1], v[2], v[3]);
+        LLVector4 vec((F32)v[0], (F32)v[1], (F32)v[2], (F32)v[3]);
         const auto& iter = mValue.find(location);
         if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
         {
@@ -1755,7 +1781,7 @@ void LLGLSLShader::uniform2i(const LLStaticHashedString& uniform, GLint i, GLint
     if (location >= 0)
     {
         const auto& iter = mValue.find(location);
-        LLVector4 vec(i, j, 0.f, 0.f);
+        LLVector4 vec((F32)i, (F32)j, 0.f, 0.f);
         if (iter == mValue.end() || shouldChange(iter->second, vec))
         {
             glUniform2i(location, i, j);
@@ -1886,6 +1912,24 @@ void LLGLSLShader::uniform4fv(const LLStaticHashedString& uniform, U32 count, co
     }
 }
 
+void LLGLSLShader::uniform4uiv(const LLStaticHashedString& uniform, U32 count, const GLuint* v)
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
+    GLint location = getUniformLocation(uniform);
+
+    if (location >= 0)
+    {
+        LLVector4 vec((F32)v[0], (F32)v[1], (F32)v[2], (F32)v[3]);
+        const auto& iter = mValue.find(location);
+        if (iter == mValue.end() || shouldChange(iter->second, vec) || count != 1)
+        {
+            LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
+            glUniform4uiv(location, count, v);
+            mValue[location] = vec;
+        }
+    }
+}
+
 void LLGLSLShader::uniformMatrix4fv(const LLStaticHashedString& uniform, U32 count, GLboolean transpose, const GLfloat* v)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER;
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index 86e5625dca..27c8f0b7d0 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -208,6 +208,7 @@ public:
     void uniform2fv(U32 index, U32 count, const GLfloat* v);
     void uniform3fv(U32 index, U32 count, const GLfloat* v);
     void uniform4fv(U32 index, U32 count, const GLfloat* v);
+    void uniform4uiv(U32 index, U32 count, const GLuint* v);
     void uniform2i(const LLStaticHashedString& uniform, GLint i, GLint j);
     void uniformMatrix2fv(U32 index, U32 count, GLboolean transpose, const GLfloat* v);
     void uniformMatrix3fv(U32 index, U32 count, GLboolean transpose, const GLfloat* v);
@@ -223,6 +224,7 @@ public:
     void uniform2fv(const LLStaticHashedString& uniform, U32 count, const GLfloat* v);
     void uniform3fv(const LLStaticHashedString& uniform, U32 count, const GLfloat* v);
     void uniform4fv(const LLStaticHashedString& uniform, U32 count, const GLfloat* v);
+    void uniform4uiv(const LLStaticHashedString& uniform, U32 count, const GLuint* v);
     void uniformMatrix4fv(const LLStaticHashedString& uniform, U32 count, GLboolean transpose, const GLfloat* v);
 
     void setMinimumAlpha(F32 minimum);
@@ -239,6 +241,10 @@ public:
 
     void clearPermutations();
     void addPermutation(std::string name, std::string value);
+    void addPermutations(const std::map<std::string, std::string>& defines)
+    {
+        mDefines.insert(defines.begin(), defines.end());
+    }
     void removePermutation(std::string name);
 
     void addConstant(const LLGLSLShader::eShaderConsts shader_const);
@@ -247,16 +253,16 @@ public:
     //if given texture uniform is active in the shader,
     //the corresponding channel will be active upon return
     //returns channel texture is enabled in from [0-MAX)
-    S32 enableTexture(S32 uniform, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE, LLTexUnit::eTextureColorSpace space = LLTexUnit::TCS_LINEAR);
-    S32 disableTexture(S32 uniform, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE, LLTexUnit::eTextureColorSpace space = LLTexUnit::TCS_LINEAR);
+    S32 enableTexture(S32 uniform, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE);
+    S32 disableTexture(S32 uniform, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE);
 
     // get the texture channel of the given uniform, or -1 if uniform is not used as a texture
     S32 getTextureChannel(S32 uniform) const;
 
     // bindTexture returns the texture unit we've bound the texture to.
     // You can reuse the return value to unbind a texture when required.
-    S32 bindTexture(const std::string& uniform, LLTexture* texture, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE, LLTexUnit::eTextureColorSpace space = LLTexUnit::TCS_LINEAR);
-    S32 bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE, LLTexUnit::eTextureColorSpace space = LLTexUnit::TCS_LINEAR);
+    S32 bindTexture(const std::string& uniform, LLTexture* texture, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE);
+    S32 bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE);
     S32 bindTexture(const std::string& uniform, LLRenderTarget* texture, bool depth = false, LLTexUnit::eTextureFilterOptions mode = LLTexUnit::TFO_BILINEAR);
     S32 bindTexture(S32 uniform, LLRenderTarget* texture, bool depth = false, LLTexUnit::eTextureFilterOptions mode = LLTexUnit::TFO_BILINEAR, U32 index = 0);
     S32 unbindTexture(const std::string& uniform, LLTexUnit::eTextureType mode = LLTexUnit::TT_TEXTURE);
diff --git a/indra/llrender/llgltexture.cpp b/indra/llrender/llgltexture.cpp
index e614f45986..4dcca5a726 100644
--- a/indra/llrender/llgltexture.cpp
+++ b/indra/llrender/llgltexture.cpp
@@ -351,20 +351,6 @@ void LLGLTexture::forceUpdateBindStats(void) const
     return mGLTexturep->forceUpdateBindStats() ;
 }
 
-U32 LLGLTexture::getTexelsInAtlas() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getTexelsInAtlas() ;
-}
-
-U32 LLGLTexture::getTexelsInGLTexture() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getTexelsInGLTexture() ;
-}
-
 bool LLGLTexture::isGLTextureCreated() const
 {
     llassert(mGLTexturep.notNull()) ;
@@ -372,13 +358,6 @@ bool LLGLTexture::isGLTextureCreated() const
     return mGLTexturep->isGLTextureCreated() ;
 }
 
-S32  LLGLTexture::getDiscardLevelInAtlas() const
-{
-    llassert(mGLTexturep.notNull()) ;
-
-    return mGLTexturep->getDiscardLevelInAtlas() ;
-}
-
 void LLGLTexture::destroyGLTexture()
 {
     if(mGLTexturep.notNull() && mGLTexturep->getHasGLTexture())
diff --git a/indra/llrender/llgltexture.h b/indra/llrender/llgltexture.h
index 0901243f8f..122d2a7f9c 100644
--- a/indra/llrender/llgltexture.h
+++ b/indra/llrender/llgltexture.h
@@ -51,10 +51,10 @@ public:
         BOOST_NONE          = 0,
         BOOST_AVATAR        ,
         BOOST_AVATAR_BAKED  ,
-        BOOST_SCULPTED      ,
         BOOST_TERRAIN       , // Needed for minimap generation for now. Lower than BOOST_HIGH so the texture stats don't get forced, i.e. texture stats are manually managed by minimap/terrain instead.
 
         BOOST_HIGH          = 10,
+        BOOST_SCULPTED      ,
         BOOST_BUMP          ,
         BOOST_UNUSED_1      , // Placeholder to avoid disrupting habits around texture debug
         BOOST_SELECTED      ,
@@ -75,7 +75,6 @@ public:
         AVATAR_SCRATCH_TEX,
         DYNAMIC_TEX,
         MEDIA,
-        ATLAS,
         OTHER,
         MAX_GL_IMAGE_CATEGORY
     };
@@ -83,8 +82,6 @@ public:
     typedef enum
     {
         DELETED = 0,         //removed from memory
-        DELETION_CANDIDATE,  //ready to be removed from memory
-        INACTIVE,            //not be used for the last certain period (i.e., 30 seconds).
         ACTIVE,              //just being used, can become inactive if not being used for a certain time (10 seconds).
         NO_DELETE = 99       //stay in memory, can not be removed.
     } LLGLTextureState;
@@ -156,10 +153,7 @@ public:
     bool       isJustBound()const ;
     void       forceUpdateBindStats(void) const;
 
-    U32        getTexelsInAtlas() const ;
-    U32        getTexelsInGLTexture() const ;
     bool       isGLTextureCreated() const ;
-    S32        getDiscardLevelInAtlas() const ;
     LLGLTextureState getTextureState() const { return mTextureState; }
 
     //---------------------------------------------------------------------------------------------
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 7e5cd628c1..68c20048ec 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -41,6 +41,7 @@
 #include "llrender.h"
 #include "llwindow.h"
 #include "llframetimer.h"
+#include <unordered_set>
 
 extern LL_COMMON_API bool on_main_thread();
 
@@ -56,6 +57,9 @@ const F32 MIN_TEXTURE_LIFETIME = 10.f;
 U32 wpo2(U32 i);
 
 
+U32 LLImageGL::sFrameCount = 0;
+
+
 // texture memory accounting (for macOS)
 static LLMutex sTexMemMutex;
 static std::unordered_map<U32, U64> sTextureAllocs;
@@ -63,15 +67,19 @@ static U64 sTextureBytes = 0;
 
 // track a texture alloc on the currently bound texture.
 // asserts that no currently tracked alloc exists
-void LLImageGLMemory::alloc_tex_image(U32 width, U32 height, U32 pixformat)
+void LLImageGLMemory::alloc_tex_image(U32 width, U32 height, U32 intformat, U32 count)
 {
     U32 texUnit = gGL.getCurrentTexUnitIndex();
+    llassert(texUnit == 0); // allocations should always be done on tex unit 0
     U32 texName = gGL.getTexUnit(texUnit)->getCurrTexture();
-    U64 size = LLImageGL::dataFormatBytes(pixformat, width, height);
+    U64 size = LLImageGL::dataFormatBytes(intformat, width, height);
+    size *= count;
 
     llassert(size >= 0);
 
     sTexMemMutex.lock();
+
+    // it is a precondition that no existing allocation exists for this texture
     llassert(sTextureAllocs.find(texName) == sTextureAllocs.end());
 
     sTextureAllocs[texName] = size;
@@ -85,7 +93,7 @@ void LLImageGLMemory::free_tex_image(U32 texName)
 {
     sTexMemMutex.lock();
     auto iter = sTextureAllocs.find(texName);
-    if (iter != sTextureAllocs.end())
+    if (iter != sTextureAllocs.end()) // sometimes a texName will be "freed" before allocated (e.g. first call to setManualImage for a given texName)
     {
         llassert(iter->second <= sTextureBytes); // sTextureBytes MUST NOT go below zero
 
@@ -110,6 +118,7 @@ void LLImageGLMemory::free_tex_images(U32 count, const U32* texNames)
 void LLImageGLMemory::free_cur_tex_image()
 {
     U32 texUnit = gGL.getCurrentTexUnitIndex();
+    llassert(texUnit == 0); // frees should always be done on tex unit 0
     U32 texName = gGL.getTexUnit(texUnit)->getCurrTexture();
     free_tex_image(texName);
 }
@@ -130,10 +139,9 @@ S32 LLImageGL::sCount                   = 0;
 
 bool LLImageGL::sGlobalUseAnisotropic   = false;
 F32 LLImageGL::sLastFrameTime           = 0.f;
-bool LLImageGL::sAllowReadBackRaw       = false ;
 LLImageGL* LLImageGL::sDefaultGLTexture = NULL ;
 bool LLImageGL::sCompressTextures = false;
-std::set<LLImageGL*> LLImageGL::sImageList;
+std::unordered_set<LLImageGL*> LLImageGL::sImageList;
 
 
 bool LLImageGLThread::sEnabledTextures = false;
@@ -150,6 +158,9 @@ S32 LLImageGL::sMaxCategories = 1 ;
 
 //optimization for when we don't need to calculate mIsMask
 bool LLImageGL::sSkipAnalyzeAlpha;
+U32  LLImageGL::sScratchPBO = 0;
+U32  LLImageGL::sScratchPBOSize = 0;
+
 
 //------------------------
 //****************************************************************************************************
@@ -159,20 +170,6 @@ bool LLImageGL::sSkipAnalyzeAlpha;
 //**************************************************************************************
 //below are functions for debug use
 //do not delete them even though they are not currently being used.
-void check_all_images()
-{
-    for (std::set<LLImageGL*>::iterator iter = LLImageGL::sImageList.begin();
-         iter != LLImageGL::sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if (glimage->getTexName() && glimage->isGLTextureCreated())
-        {
-            gGL.getTexUnit(0)->bind(glimage) ;
-            glimage->checkTexSize() ;
-            gGL.getTexUnit(0)->unbind(glimage->getTarget()) ;
-        }
-    }
-}
 
 void LLImageGL::checkTexSize(bool forced) const
 {
@@ -252,6 +249,11 @@ void LLImageGL::initClass(LLWindow* window, S32 num_catagories, bool skip_analyz
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     sSkipAnalyzeAlpha = skip_analyze_alpha;
 
+    if (sScratchPBO == 0)
+    {
+        glGenBuffers(1, &sScratchPBO);
+    }
+
     if (thread_texture_loads || thread_media_updates)
     {
         LLImageGLThread::createInstance(window);
@@ -265,6 +267,12 @@ void LLImageGL::cleanupClass()
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
     LLImageGLThread::deleteSingleton();
+    if (sScratchPBO != 0)
+    {
+        glDeleteBuffers(1, &sScratchPBO);
+        sScratchPBO = 0;
+        sScratchPBOSize = 0;
+    }
 }
 
 
@@ -273,6 +281,15 @@ S32 LLImageGL::dataFormatBits(S32 dataformat)
 {
     switch (dataformat)
     {
+    case GL_COMPRESSED_RED:                         return 8;
+    case GL_COMPRESSED_RG:                          return 16;
+    case GL_COMPRESSED_RGB:                         return 24;
+    case GL_COMPRESSED_SRGB:                        return 32;
+    case GL_COMPRESSED_RGBA:                        return 32;
+    case GL_COMPRESSED_SRGB_ALPHA:                  return 32;
+    case GL_COMPRESSED_LUMINANCE:                   return 8;
+    case GL_COMPRESSED_LUMINANCE_ALPHA:             return 16;
+    case GL_COMPRESSED_ALPHA:                       return 8;
     case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:          return 4;
     case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:    return 4;
     case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:          return 8;
@@ -280,21 +297,35 @@ S32 LLImageGL::dataFormatBits(S32 dataformat)
     case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:          return 8;
     case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:    return 8;
     case GL_LUMINANCE:                              return 8;
+    case GL_LUMINANCE8:                             return 8;
     case GL_ALPHA:                                  return 8;
+    case GL_ALPHA8:                                 return 8;
     case GL_RED:                                    return 8;
+    case GL_R8:                                     return 8;
     case GL_COLOR_INDEX:                            return 8;
     case GL_LUMINANCE_ALPHA:                        return 16;
+    case GL_LUMINANCE8_ALPHA8:                      return 16;
+    case GL_RG:                                     return 16;
+    case GL_RG8:                                    return 16;
     case GL_RGB:                                    return 24;
     case GL_SRGB:                                   return 24;
     case GL_RGB8:                                   return 24;
     case GL_RGBA:                                   return 32;
+    case GL_RGBA8:                                  return 32;
     case GL_SRGB_ALPHA:                             return 32;
     case GL_BGRA:                                   return 32;      // Used for QuickTime media textures on the Mac
     case GL_DEPTH_COMPONENT:                        return 24;
+    case GL_DEPTH_COMPONENT24:                      return 24;
+    case GL_R16F:                                   return 16;
+    case GL_RG16F:                                  return 32;
     case GL_RGB16F:                                 return 48;
     case GL_RGBA16F:                                return 64;
+    case GL_R32F:                                   return 32;
+    case GL_RG32F:                                  return 64;
+    case GL_RGB32F:                                 return 96;
+    case GL_RGBA32F:                                return 128;
     default:
-        LL_ERRS() << "LLImageGL::Unknown format: " << dataformat << LL_ENDL;
+        LL_ERRS() << "LLImageGL::Unknown format: " << std::hex << dataformat << std::dec << LL_ENDL;
         return 0;
     }
 }
@@ -337,13 +368,14 @@ S32 LLImageGL::dataFormatComponents(S32 dataformat)
       case GL_RED:                              return 1;
       case GL_COLOR_INDEX:                      return 1;
       case GL_LUMINANCE_ALPHA:                  return 2;
+      case GL_RG:                               return 2;
       case GL_RGB:                              return 3;
       case GL_SRGB:                             return 3;
       case GL_RGBA:                             return 4;
       case GL_SRGB_ALPHA:                       return 4;
       case GL_BGRA:                             return 4;       // Used for QuickTime media textures on the Mac
       default:
-        LL_ERRS() << "LLImageGL::Unknown format: " << dataformat << LL_ENDL;
+        LL_ERRS() << "LLImageGL::Unknown format: " << std::hex << dataformat << std::dec << LL_ENDL;
         return 0;
     }
 }
@@ -360,66 +392,19 @@ void LLImageGL::updateStats(F32 current_time)
 //----------------------------------------------------------------------------
 
 //static
-void LLImageGL::destroyGL(bool save_state)
+void LLImageGL::destroyGL()
 {
     for (S32 stage = 0; stage < gGLManager.mNumTextureImageUnits; stage++)
     {
         gGL.getTexUnit(stage)->unbind(LLTexUnit::TT_TEXTURE);
     }
-
-    sAllowReadBackRaw = true ;
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if (glimage->mTexName)
-        {
-            if (save_state && glimage->isGLTextureCreated() && glimage->mComponents)
-            {
-                glimage->mSaveData = new LLImageRaw;
-                if(!glimage->readBackRaw(glimage->mCurrentDiscardLevel, glimage->mSaveData, false)) //necessary, keep it.
-                {
-                    glimage->mSaveData = NULL ;
-                }
-            }
-
-            glimage->destroyGLTexture();
-            stop_glerror();
-        }
-    }
-    sAllowReadBackRaw = false ;
-}
-
-//static
-void LLImageGL::restoreGL()
-{
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
-    {
-        LLImageGL* glimage = *iter;
-        if(glimage->getTexName())
-        {
-            LL_ERRS() << "tex name is not 0." << LL_ENDL ;
-        }
-        if (glimage->mSaveData.notNull())
-        {
-            if (glimage->getComponents() && glimage->mSaveData->getComponents())
-            {
-                glimage->createGLTexture(glimage->mCurrentDiscardLevel, glimage->mSaveData, 0, true, glimage->getCategory());
-                stop_glerror();
-            }
-            glimage->mSaveData = NULL; // deletes data
-        }
-    }
 }
 
 //static
 void LLImageGL::dirtyTexOptions()
 {
-    for (std::set<LLImageGL*>::iterator iter = sImageList.begin();
-         iter != sImageList.end(); iter++)
+    for (auto& glimage : sImageList)
     {
-        LLImageGL* glimage = *iter;
         glimage->mTexOptionsDirty = true;
         stop_glerror();
     }
@@ -451,29 +436,29 @@ bool LLImageGL::create(LLPointer<LLImageGL>& dest, const LLImageRaw* imageraw, b
 
 //----------------------------------------------------------------------------
 
-LLImageGL::LLImageGL(bool usemipmaps)
+LLImageGL::LLImageGL(bool usemipmaps/* = true*/, bool allow_compression/* = true*/)
 :   mSaveData(0), mExternalTexture(false)
 {
-    init(usemipmaps);
+    init(usemipmaps, allow_compression);
     setSize(0, 0, 0);
     sImageList.insert(this);
     sCount++;
 }
 
-LLImageGL::LLImageGL(U32 width, U32 height, U8 components, bool usemipmaps)
+LLImageGL::LLImageGL(U32 width, U32 height, U8 components, bool usemipmaps/* = true*/, bool allow_compression/* = true*/)
 :   mSaveData(0), mExternalTexture(false)
 {
     llassert( components <= 4 );
-    init(usemipmaps);
+    init(usemipmaps, allow_compression);
     setSize(width, height, components);
     sImageList.insert(this);
     sCount++;
 }
 
-LLImageGL::LLImageGL(const LLImageRaw* imageraw, bool usemipmaps)
+LLImageGL::LLImageGL(const LLImageRaw* imageraw, bool usemipmaps/* = true*/, bool allow_compression/* = true*/)
 :   mSaveData(0), mExternalTexture(false)
 {
-    init(usemipmaps);
+    init(usemipmaps, allow_compression);
     setSize(0, 0, 0);
     sImageList.insert(this);
     sCount++;
@@ -490,7 +475,7 @@ LLImageGL::LLImageGL(
     LLGLenum formatType,
     LLTexUnit::eTextureAddressMode addressMode)
 {
-    init(false);
+    init(false, true);
     mTexName = texName;
     mTarget = target;
     mComponents = components;
@@ -512,7 +497,7 @@ LLImageGL::~LLImageGL()
     }
 }
 
-void LLImageGL::init(bool usemipmaps)
+void LLImageGL::init(bool usemipmaps, bool allow_compression)
 {
 #if LL_IMAGEGL_THREAD_CHECK
     mActiveThread = LLThread::currentID();
@@ -542,11 +527,7 @@ void LLImageGL::init(bool usemipmaps)
     mHeight = 0;
     mCurrentDiscardLevel = -1;
 
-    mDiscardLevelInAtlas = -1 ;
-    mTexelsInAtlas = 0 ;
-    mTexelsInGLTexture = 0 ;
-
-    mAllowCompression = true;
+    mAllowCompression = allow_compression;
 
     mTarget = GL_TEXTURE_2D;
     mBindTarget = LLTexUnit::TT_TEXTURE;
@@ -622,9 +603,6 @@ bool LLImageGL::setSize(S32 width, S32 height, S32 ncomponents, S32 discard_leve
             return false;
         }
 
-        // pickmask validity depends on old image size, delete it
-        freePickMask();
-
         mWidth = width;
         mHeight = height;
         mComponents = ncomponents;
@@ -789,7 +767,7 @@ bool LLImageGL::setImage(const U8* data_in, bool data_hasmips /* = false */, S32
                 }
                 if (is_compressed)
                 {
-                    S32 tex_size = dataFormatBytes(mFormatPrimary, w, h);
+                    GLsizei tex_size = (GLsizei)dataFormatBytes(mFormatPrimary, w, h);
                     glCompressedTexImage2D(mTarget, gl_level, mFormatPrimary, w, h, 0, tex_size, (GLvoid *)data_in);
                     stop_glerror();
                 }
@@ -992,7 +970,7 @@ bool LLImageGL::setImage(const U8* data_in, bool data_hasmips /* = false */, S32
         S32 h = getHeight();
         if (is_compressed)
         {
-            S32 tex_size = dataFormatBytes(mFormatPrimary, w, h);
+            GLsizei tex_size = (GLsizei)dataFormatBytes(mFormatPrimary, w, h);
             glCompressedTexImage2D(mTarget, 0, mFormatPrimary, w, h, 0, tex_size, (GLvoid *)data_in);
             stop_glerror();
         }
@@ -1025,98 +1003,6 @@ bool LLImageGL::setImage(const U8* data_in, bool data_hasmips /* = false */, S32
     return true;
 }
 
-bool LLImageGL::preAddToAtlas(S32 discard_level, const LLImageRaw* raw_image)
-{
-    //not compatible with core GL profile
-    llassert(!LLRender::sGLCoreProfile);
-
-    if (gGLManager.mIsDisabled)
-    {
-        LL_WARNS() << "Trying to create a texture while GL is disabled!" << LL_ENDL;
-        return false;
-    }
-    llassert(gGLManager.mInited);
-    stop_glerror();
-
-    if (discard_level < 0)
-    {
-        llassert(mCurrentDiscardLevel >= 0);
-        discard_level = mCurrentDiscardLevel;
-    }
-
-    // Actual image width/height = raw image width/height * 2^discard_level
-    S32 w = raw_image->getWidth() << discard_level;
-    S32 h = raw_image->getHeight() << discard_level;
-
-    // setSize may call destroyGLTexture if the size does not match
-    if (!setSize(w, h, raw_image->getComponents(), discard_level))
-    {
-        LL_WARNS() << "Trying to create a texture with incorrect dimensions!" << LL_ENDL;
-        return false;
-    }
-
-    if (!mHasExplicitFormat)
-    {
-        switch (mComponents)
-        {
-            case 1:
-                // Use luminance alpha (for fonts)
-                mFormatInternal = GL_LUMINANCE8;
-                mFormatPrimary  = GL_LUMINANCE;
-                mFormatType     = GL_UNSIGNED_BYTE;
-                break;
-            case 2:
-                // Use luminance alpha (for fonts)
-                mFormatInternal = GL_LUMINANCE8_ALPHA8;
-                mFormatPrimary  = GL_LUMINANCE_ALPHA;
-                mFormatType     = GL_UNSIGNED_BYTE;
-                break;
-            case 3:
-                mFormatInternal = GL_RGB8;
-                mFormatPrimary = GL_RGB;
-                mFormatType    = GL_UNSIGNED_BYTE;
-                break;
-            case 4:
-                mFormatInternal = GL_RGBA8;
-                mFormatPrimary = GL_RGBA;
-                mFormatType    = GL_UNSIGNED_BYTE;
-                break;
-            default:
-                LL_ERRS() << "Bad number of components for texture: " << (U32) getComponents() << LL_ENDL;
-        }
-    }
-
-    mCurrentDiscardLevel = discard_level;
-    mDiscardLevelInAtlas = discard_level;
-    mTexelsInAtlas = raw_image->getWidth() * raw_image->getHeight() ;
-    mLastBindTime = sLastFrameTime;
-    mGLTextureCreated = false ;
-
-    glPixelStorei(GL_UNPACK_ROW_LENGTH, raw_image->getWidth());
-    stop_glerror();
-
-    if(mFormatSwapBytes)
-    {
-        glPixelStorei(GL_UNPACK_SWAP_BYTES, 1);
-        stop_glerror();
-    }
-
-    return true ;
-}
-
-void LLImageGL::postAddToAtlas()
-{
-    if(mFormatSwapBytes)
-    {
-        glPixelStorei(GL_UNPACK_SWAP_BYTES, 0);
-        stop_glerror();
-    }
-
-    glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
-    gGL.getTexUnit(0)->setTextureFilteringOption(mFilterOption);
-    stop_glerror();
-}
-
 U32 type_width_from_pixtype(U32 pixtype)
 {
     U32 type_width = 0;
@@ -1149,7 +1035,7 @@ bool should_stagger_image_set(bool compressed)
 #else
     // glTexSubImage2D doesn't work with compressed textures on select tested Nvidia GPUs on Windows 10 -Cosmic,2023-03-08
     // Setting media textures off-thread seems faster when not using sub_image_lines (Nvidia/Windows 10) -Cosmic,2023-03-31
-    return !compressed && on_main_thread();
+    return !compressed && on_main_thread() && !gGLManager.mIsIntel;
 #endif
 }
 
@@ -1332,12 +1218,36 @@ void LLImageGL::generateTextures(S32 numTextures, U32 *textures)
 }
 
 // static
+void LLImageGL::updateClass()
+{
+    sFrameCount++;
+}
+
+// static
 void LLImageGL::deleteTextures(S32 numTextures, const U32 *textures)
 {
+    // wait a few frames before actually deleting the textures to avoid
+    // synchronization issues with the GPU
+    static std::vector<U32> sFreeList[4];
+
     if (gGLManager.mInited)
     {
-        free_tex_images(numTextures, textures);
-        glDeleteTextures(numTextures, textures);
+        LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+        U32 idx = sFrameCount % 4;
+
+        for (S32 i = 0; i < numTextures; ++i)
+        {
+            sFreeList[idx].push_back(textures[i]);
+        }
+
+        idx = (sFrameCount + 3) % 4;
+
+        if (!sFreeList[idx].empty())
+        {
+            free_tex_images((GLsizei) sFreeList[idx].size(), sFreeList[idx].data());
+            glDeleteTextures((GLsizei)sFreeList[idx].size(), sFreeList[idx].data());
+            sFreeList[idx].resize(0);
+        }
     }
 }
 
@@ -1345,90 +1255,122 @@ void LLImageGL::deleteTextures(S32 numTextures, const U32 *textures)
 void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void* pixels, bool allow_compression)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
-    bool use_scratch = false;
-    U32* scratch = NULL;
+    std::unique_ptr<U32[]> scratch;
     if (LLRender::sGLCoreProfile)
     {
-        if (pixformat == GL_ALPHA && pixtype == GL_UNSIGNED_BYTE)
-        { //GL_ALPHA is deprecated, convert to RGBA
-            if (pixels != nullptr)
-            {
-                use_scratch = true;
-                scratch = new(std::nothrow) U32[width * height];
-                if (!scratch)
-                {
-                    LLError::LLUserWarningMsg::showOutOfMemory();
-                    LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
-                              << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
-                }
+        LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+        if (gGLManager.mGLVersion >= 3.29f)
+        {
+            if (pixformat == GL_ALPHA)
+            { //GL_ALPHA is deprecated, convert to RGBA
+                const GLint mask[] = { GL_ZERO, GL_ZERO, GL_ZERO, GL_RED };
+                glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask);
+                pixformat = GL_RED;
+                intformat = GL_R8;
+            }
 
-                U32 pixel_count = (U32)(width * height);
-                for (U32 i = 0; i < pixel_count; i++)
-                {
-                    U8* pix = (U8*)&scratch[i];
-                    pix[0] = pix[1] = pix[2] = 0;
-                    pix[3] = ((U8*)pixels)[i];
-                }
+            if (pixformat == GL_LUMINANCE)
+            { //GL_LUMINANCE is deprecated, convert to GL_RGBA
+                const GLint mask[] = { GL_RED, GL_RED, GL_RED, GL_ONE };
+                glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask);
+                pixformat = GL_RED;
+                intformat = GL_R8;
             }
 
-            pixformat = GL_RGBA;
-            intformat = GL_RGBA8;
+            if (pixformat == GL_LUMINANCE_ALPHA)
+            { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA
+                const GLint mask[] = { GL_RED, GL_RED, GL_RED, GL_GREEN };
+                glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask);
+                pixformat = GL_RG;
+                intformat = GL_RG8;
+            }
         }
-
-        if (pixformat == GL_LUMINANCE_ALPHA && pixtype == GL_UNSIGNED_BYTE)
-        { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA
-            if (pixels != nullptr)
-            {
-                use_scratch = true;
-                scratch = new(std::nothrow) U32[width * height];
-                if (!scratch)
+        else
+        {
+            if (pixformat == GL_ALPHA && pixtype == GL_UNSIGNED_BYTE)
+            { //GL_ALPHA is deprecated, convert to RGBA
+                if (pixels != nullptr)
                 {
-                    LLError::LLUserWarningMsg::showOutOfMemory();
-                    LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
-                        << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
-                }
+                    scratch.reset(new(std::nothrow) U32[width * height]);
+                    if (!scratch)
+                    {
+                        LLError::LLUserWarningMsg::showOutOfMemory();
+                        LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
+                            << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
+                    }
 
-                U32 pixel_count = (U32)(width * height);
-                for (U32 i = 0; i < pixel_count; i++)
-                {
-                    U8 lum = ((U8*)pixels)[i * 2 + 0];
-                    U8 alpha = ((U8*)pixels)[i * 2 + 1];
+                    U32 pixel_count = (U32)(width * height);
+                    for (U32 i = 0; i < pixel_count; i++)
+                    {
+                        U8* pix = (U8*)&scratch[i];
+                        pix[0] = pix[1] = pix[2] = 0;
+                        pix[3] = ((U8*)pixels)[i];
+                    }
 
-                    U8* pix = (U8*)&scratch[i];
-                    pix[0] = pix[1] = pix[2] = lum;
-                    pix[3] = alpha;
+                    pixels = scratch.get();
                 }
-            }
 
-            pixformat = GL_RGBA;
-            intformat = GL_RGBA8;
-        }
+                pixformat = GL_RGBA;
+                intformat = GL_RGBA8;
+            }
 
-        if (pixformat == GL_LUMINANCE && pixtype == GL_UNSIGNED_BYTE)
-        { //GL_LUMINANCE_ALPHA is deprecated, convert to RGB
-            if (pixels != nullptr)
-            {
-                use_scratch = true;
-                scratch = new(std::nothrow) U32[width * height];
-                if (!scratch)
+            if (pixformat == GL_LUMINANCE_ALPHA && pixtype == GL_UNSIGNED_BYTE)
+            { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA
+                if (pixels != nullptr)
                 {
-                    LLError::LLUserWarningMsg::showOutOfMemory();
-                    LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
-                        << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
+                    scratch.reset(new(std::nothrow) U32[width * height]);
+                    if (!scratch)
+                    {
+                        LLError::LLUserWarningMsg::showOutOfMemory();
+                        LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
+                            << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
+                    }
+
+                    U32 pixel_count = (U32)(width * height);
+                    for (U32 i = 0; i < pixel_count; i++)
+                    {
+                        U8 lum = ((U8*)pixels)[i * 2 + 0];
+                        U8 alpha = ((U8*)pixels)[i * 2 + 1];
+
+                        U8* pix = (U8*)&scratch[i];
+                        pix[0] = pix[1] = pix[2] = lum;
+                        pix[3] = alpha;
+                    }
+
+                    pixels = scratch.get();
                 }
 
-                U32 pixel_count = (U32)(width * height);
-                for (U32 i = 0; i < pixel_count; i++)
+                pixformat = GL_RGBA;
+                intformat = GL_RGBA8;
+            }
+
+            if (pixformat == GL_LUMINANCE && pixtype == GL_UNSIGNED_BYTE)
+            { //GL_LUMINANCE_ALPHA is deprecated, convert to RGB
+                if (pixels != nullptr)
                 {
-                    U8 lum = ((U8*)pixels)[i];
+                    scratch.reset(new(std::nothrow) U32[width * height]);
+                    if (!scratch)
+                    {
+                        LLError::LLUserWarningMsg::showOutOfMemory();
+                        LL_ERRS() << "Failed to allocate " << (U32)(width * height * sizeof(U32))
+                            << " bytes for a manual image W" << width << " H" << height << LL_ENDL;
+                    }
+
+                    U32 pixel_count = (U32)(width * height);
+                    for (U32 i = 0; i < pixel_count; i++)
+                    {
+                        U8 lum = ((U8*)pixels)[i];
 
-                    U8* pix = (U8*)&scratch[i];
-                    pix[0] = pix[1] = pix[2] = lum;
-                    pix[3] = 255;
+                        U8* pix = (U8*)&scratch[i];
+                        pix[0] = pix[1] = pix[2] = lum;
+                        pix[3] = 255;
+                    }
+
+                    pixels = scratch.get();
                 }
+                pixformat = GL_RGBA;
+                intformat = GL_RGB8;
             }
-            pixformat = GL_RGBA;
-            intformat = GL_RGB8;
         }
     }
 
@@ -1437,6 +1379,14 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
     {
         switch (intformat)
         {
+        case GL_RED:
+        case GL_R8:
+            intformat = GL_COMPRESSED_RED;
+            break;
+        case GL_RG:
+        case GL_RG8:
+            intformat = GL_COMPRESSED_RG;
+            break;
         case GL_RGB:
         case GL_RGB8:
             intformat = GL_COMPRESSED_RGB;
@@ -1465,12 +1415,8 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
         case GL_ALPHA8:
             intformat = GL_COMPRESSED_ALPHA;
             break;
-        case GL_RED:
-        case GL_R8:
-            intformat = GL_COMPRESSED_RED;
-            break;
         default:
-            LL_WARNS() << "Could not compress format: " << std::hex << intformat << LL_ENDL;
+            LL_WARNS() << "Could not compress format: " << std::hex << intformat << std::dec << LL_ENDL;
             break;
         }
     }
@@ -1486,7 +1432,7 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
         if (!use_sub_image)
         {
             LL_PROFILE_ZONE_NAMED("glTexImage2D alloc + copy");
-            glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels);
+            glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, pixels);
         }
         else
         {
@@ -1496,21 +1442,16 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt
                 glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, nullptr);
             }
 
-            U8* src = (U8*)(use_scratch ? scratch : pixels);
+            U8* src = (U8*)(pixels);
             if (src)
             {
                 LL_PROFILE_ZONE_NAMED("glTexImage2D copy");
                 sub_image_lines(target, miplevel, 0, 0, width, height, pixformat, pixtype, src, width);
             }
         }
-        alloc_tex_image(width, height, pixformat);
+        alloc_tex_image(width, height, intformat, 1);
     }
     stop_glerror();
-
-    if (use_scratch)
-    {
-        delete[] scratch;
-    }
 }
 
 //create an empty GL texture: just create a texture name
@@ -1752,7 +1693,6 @@ bool LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, bool data_
 
 
     mTextureMemory = (S64Bytes)getMipBytes(mCurrentDiscardLevel);
-    mTexelsInGLTexture = getWidth() * getHeight();
 
     // mark this as bound at this point, so we don't throw it out immediately
     mLastBindTime = sLastFrameTime;
@@ -1830,8 +1770,7 @@ void LLImageGL::syncTexName(LLGLuint texname)
 
 bool LLImageGL::readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const
 {
-    llassert_always(sAllowReadBackRaw) ;
-    //LL_ERRS() << "should not call this function!" << LL_ENDL ;
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
 
     if (discard_level < 0)
     {
@@ -2297,6 +2236,8 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h)
 //----------------------------------------------------------------------------
 U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight)
 {
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+    freePickMask();
     U32 pick_width = pWidth/2 + 1;
     U32 pick_height = pHeight/2 + 1;
 
@@ -2314,7 +2255,6 @@ U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight)
 //----------------------------------------------------------------------------
 void LLImageGL::freePickMask()
 {
-    // pickmask validity depends on old image size, delete it
     if (mPickMask != NULL)
     {
         delete [] mPickMask;
@@ -2352,16 +2292,16 @@ void LLImageGL::updatePickMask(S32 width, S32 height, const U8* data_in)
         return ;
     }
 
-    freePickMask();
-
     if (mFormatType != GL_UNSIGNED_BYTE ||
         ((mFormatPrimary != GL_RGBA)
       && (mFormatPrimary != GL_SRGB_ALPHA)))
     {
         //cannot generate a pick mask for this texture
+        freePickMask();
         return;
     }
 
+
 #ifdef SHOW_ASSERT
     const U32 pickSize = createPickMask(width, height);
 #else // SHOW_ASSERT
@@ -2460,6 +2400,114 @@ void LLImageGL::resetCurTexSizebar()
     sCurTexSizeBar = -1 ;
     sCurTexPickSize = -1 ;
 }
+
+bool LLImageGL::scaleDown(S32 desired_discard)
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE;
+
+    if (mTarget != GL_TEXTURE_2D)
+    {
+        return false;
+    }
+
+    desired_discard = llmin(desired_discard, mMaxDiscardLevel);
+
+    if (desired_discard <= mCurrentDiscardLevel)
+    {
+        return false;
+    }
+
+    S32 mip = desired_discard - mCurrentDiscardLevel;
+
+    S32 desired_width = getWidth(desired_discard);
+    S32 desired_height = getHeight(desired_discard);
+
+    if (gGLManager.mDownScaleMethod == 0)
+    { // use an FBO to downscale the texture
+        // allocate new texture
+        U32 temp_texname = 0;
+        generateTextures(1, &temp_texname);
+        gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, temp_texname, true);
+        {
+            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("scaleDown - glTexImage2D");
+            glTexImage2D(mTarget, 0, mFormatInternal, desired_width, desired_height, 0, mFormatPrimary, mFormatType, NULL);
+        }
+
+        // account for new texture getting created
+        alloc_tex_image(desired_width, desired_height, mFormatInternal, 1);
+
+        // Use render-to-texture to scale down the texture
+        {
+            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("scaleDown - glFramebufferTexture2D");
+            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, mTarget, temp_texname, 0);
+        }
+
+        glViewport(0, 0, desired_width, desired_height);
+
+        // draw a full screen triangle
+        gGL.getTexUnit(0)->bind(this);
+        glDrawArrays(GL_TRIANGLES, 0, 3);
+        gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+
+        // delete old texture and assign new texture name
+        deleteTextures(1, &mTexName);
+        mTexName = temp_texname;
+
+        if (mHasMipMaps)
+        { // generate mipmaps if needed
+            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("scaleDown - glGenerateMipmap");
+            gGL.getTexUnit(0)->bind(this);
+            glGenerateMipmap(mTarget);
+            gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+        }
+    }
+    else
+    { // use a PBO to downscale the texture
+        U64 size = getBytes(desired_discard);
+        llassert(size <= 2048 * 2048 * 4); // we shouldn't be using this method to downscale huge textures, but it'll work
+        gGL.getTexUnit(0)->bind(this, false, true);
+
+        if (sScratchPBO == 0)
+        {
+            glGenBuffers(1, &sScratchPBO);
+            sScratchPBOSize = 0;
+        }
+
+        glBindBuffer(GL_PIXEL_PACK_BUFFER, sScratchPBO);
+
+        if (size > sScratchPBOSize)
+        {
+            glBufferData(GL_PIXEL_PACK_BUFFER, size, NULL, GL_STREAM_COPY);
+            sScratchPBOSize = (U32)size;
+        }
+
+        glGetTexImage(mTarget, mip, mFormatPrimary, mFormatType, nullptr);
+
+        free_tex_image(mTexName);
+
+        glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+
+        glBindBuffer(GL_PIXEL_UNPACK_BUFFER, sScratchPBO);
+        glTexImage2D(mTarget, 0, mFormatInternal, desired_width, desired_height, 0, mFormatPrimary, mFormatType, nullptr);
+        glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+
+        alloc_tex_image(desired_width, desired_height, mFormatInternal, 1);
+
+        if (mHasMipMaps)
+        {
+            LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("scaleDown - glGenerateMipmap");
+            glGenerateMipmap(mTarget);
+        }
+
+        gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+    }
+
+    mCurrentDiscardLevel = desired_discard;
+
+    return true;
+}
+
+
 //----------------------------------------------------------------------------
 #if LL_IMAGEGL_THREAD_CHECK
 void LLImageGL::checkActiveThread()
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index 5c7a5ce821..a8b94bd5b0 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -39,6 +39,7 @@
 #include "llrender.h"
 #include "threadpool.h"
 #include "workqueue.h"
+#include <unordered_set>
 
 #define LL_IMAGEGL_THREAD_CHECK 0 //set to 1 to enable thread debugging for ImageGL
 
@@ -49,7 +50,7 @@ class LLWindow;
 
 namespace LLImageGLMemory
 {
-    void alloc_tex_image(U32 width, U32 height, U32 pixformat);
+    void alloc_tex_image(U32 width, U32 height, U32 intformat, U32 count);
     void free_tex_image(U32 texName);
     void free_tex_images(U32 count, const U32* texNames);
     void free_cur_tex_image();
@@ -61,6 +62,9 @@ class LLImageGL : public LLRefCount
     friend class LLTexUnit;
 public:
 
+    // call once per frame
+    static void updateClass();
+
     // Get an estimate of how many bytes have been allocated in vram for textures.
     // Does not include mipmaps.
     // NOTE: multiplying this number by two gives a good estimate for total
@@ -83,9 +87,8 @@ public:
     // needs to be called every frame
     static void updateStats(F32 current_time);
 
-    // Save off / restore GL textures
-    static void destroyGL(bool save_state = true);
-    static void restoreGL();
+    // cleanup GL state
+    static void destroyGL();
     static void dirtyTexOptions();
 
     static bool checkSize(S32 width, S32 height);
@@ -98,9 +101,9 @@ public:
     static bool create(LLPointer<LLImageGL>& dest, const LLImageRaw* imageraw, bool usemipmaps = true);
 
 public:
-    LLImageGL(bool usemipmaps = true);
-    LLImageGL(U32 width, U32 height, U8 components, bool usemipmaps = true);
-    LLImageGL(const LLImageRaw* imageraw, bool usemipmaps = true);
+    LLImageGL(bool usemipmaps = true, bool allow_compression = true);
+    LLImageGL(U32 width, U32 height, U8 components, bool usemipmaps = true, bool allow_compression = true);
+    LLImageGL(const LLImageRaw* imageraw, bool usemipmaps = true, bool allow_compression = true);
 
     // For wrapping textures created via GL elsewhere with our API only. Use with caution.
     LLImageGL(LLGLuint mTexName, U32 components, LLGLenum target, LLGLint  formatInternal, LLGLenum formatPrimary, LLGLenum formatType, LLTexUnit::eTextureAddressMode addressMode);
@@ -148,6 +151,10 @@ public:
     S32  getDiscardLevel() const        { return mCurrentDiscardLevel; }
     S32  getMaxDiscardLevel() const     { return mMaxDiscardLevel; }
 
+    // override the current discard level
+    // should only be used for local textures where you know exactly what you're doing
+    void setDiscardLevel(S32 level) { mCurrentDiscardLevel = level; }
+
     S32  getCurrentWidth() const { return mWidth ;}
     S32  getCurrentHeight() const { return mHeight ;}
     S32  getWidth(S32 discard_level = -1) const;
@@ -194,26 +201,26 @@ public:
     void setFilteringOption(LLTexUnit::eTextureFilterOptions option);
     LLTexUnit::eTextureFilterOptions getFilteringOption(void) const { return mFilterOption; }
 
-    LLGLenum getTexTarget()const { return mTarget ;}
-    S8       getDiscardLevelInAtlas()const {return mDiscardLevelInAtlas;}
-    U32      getTexelsInAtlas()const { return mTexelsInAtlas ;}
-    U32      getTexelsInGLTexture()const {return mTexelsInGLTexture;}
+    LLGLenum getTexTarget()const { return mTarget; }
 
-
-    void init(bool usemipmaps);
+    void init(bool usemipmaps, bool allow_compression);
     virtual void cleanup(); // Clean up the LLImageGL so it can be reinitialized.  Be careful when using this in derived class destructors
 
     void setNeedsAlphaAndPickMask(bool need_mask);
 
-    bool preAddToAtlas(S32 discard_level, const LLImageRaw* raw_image);
-    void postAddToAtlas() ;
-
 #if LL_IMAGEGL_THREAD_CHECK
     // thread debugging
     std::thread::id mActiveThread;
     void checkActiveThread();
 #endif
 
+    // scale down to the desired discard level using GPU
+    // returns true if texture was scaled down
+    // desired discard will be clamped to max discard
+    // if desired discard is less than or equal to current discard, no scaling will occur
+    // only works for GL_TEXTURE_2D target
+    bool scaleDown(S32 desired_discard);
+
 public:
     // Various GL/Rendering options
     S64Bytes mTextureMemory;
@@ -240,15 +247,10 @@ private:
 
     bool     mGLTextureCreated ;
     LLGLuint mTexName;
-    //LLGLuint mNewTexName = 0; // tex name set by background thread to be applied in main thread
     U16      mWidth;
     U16      mHeight;
     S8       mCurrentDiscardLevel;
 
-    S8       mDiscardLevelInAtlas;
-    U32      mTexelsInAtlas ;
-    U32      mTexelsInGLTexture;
-
     bool mAllowCompression;
 
 protected:
@@ -275,9 +277,9 @@ protected:
 
     // STATICS
 public:
-    static std::set<LLImageGL*> sImageList;
+    static std::unordered_set<LLImageGL*> sImageList;
     static S32 sCount;
-
+    static U32 sFrameCount;
     static F32 sLastFrameTime;
 
     // Global memory statistics
@@ -301,6 +303,8 @@ public:
 private:
     static S32 sMaxCategories;
     static bool sSkipAnalyzeAlpha;
+    static U32 sScratchPBO;
+    static U32 sScratchPBOSize;
 
     //the flag to allow to call readBackRaw(...).
     //can be removed if we do not use that function at all.
diff --git a/indra/llrender/llpostprocess.cpp b/indra/llrender/llpostprocess.cpp
index 8ebd09f20d..eef7193c92 100644
--- a/indra/llrender/llpostprocess.cpp
+++ b/indra/llrender/llpostprocess.cpp
@@ -343,7 +343,7 @@ void LLPostProcess::viewOrthogonal(unsigned int width, unsigned int height)
     gGL.matrixMode(LLRender::MM_PROJECTION);
     gGL.pushMatrix();
     gGL.loadIdentity();
-    gGL.ortho( 0.f, (GLdouble) width , (GLdouble) height , 0.f, -1.f, 1.f );
+    gGL.ortho( 0.f, (GLfloat) width , (GLfloat) height , 0.f, -1.f, 1.f );
     gGL.matrixMode(LLRender::MM_MODELVIEW);
     gGL.pushMatrix();
     gGL.loadIdentity();
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index a0209fab43..828a509971 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -77,6 +77,7 @@ struct LLVBCache
 };
 
 static std::unordered_map<U64, LLVBCache> sVBCache;
+static thread_local std::list<LLVertexBufferData> *sBufferDataList = nullptr;
 
 static const GLenum sGLTextureType[] =
 {
@@ -115,7 +116,7 @@ static const GLenum sGLBlendFactor[] =
 
 LLTexUnit::LLTexUnit(S32 index)
     : mCurrTexType(TT_NONE),
-    mCurrColorScale(1), mCurrAlphaScale(1), mCurrTexture(0), mTexColorSpace(TCS_LINEAR),
+    mCurrColorScale(1), mCurrAlphaScale(1), mCurrTexture(0),
     mHasMipMaps(false),
     mIndex(index)
 {
@@ -145,8 +146,6 @@ void LLTexUnit::refreshState(void)
     {
         glBindTexture(GL_TEXTURE_2D, 0);
     }
-
-    setTextureColorSpace(mTexColorSpace);
 }
 
 void LLTexUnit::activate(void)
@@ -241,7 +240,6 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind)
                         setTextureAddressMode(gl_tex->mAddressMode);
                         setTextureFilteringOption(gl_tex->mFilterOption);
                     }
-                    setTextureColorSpace(mTexColorSpace);
                 }
             }
             else
@@ -318,7 +316,6 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind, S32
             setTextureFilteringOption(texture->mFilterOption);
             stop_glerror();
         }
-        setTextureColorSpace(mTexColorSpace);
     }
 
     stop_glerror();
@@ -354,7 +351,6 @@ bool LLTexUnit::bind(LLCubeMap* cubeMap)
                 setTextureAddressMode(cubeMap->mImages[0]->mAddressMode);
                 setTextureFilteringOption(cubeMap->mImages[0]->mFilterOption);
             }
-            setTextureColorSpace(mTexColorSpace);
             return true;
         }
         else
@@ -403,7 +399,6 @@ bool LLTexUnit::bindManual(eTextureType type, U32 texture, bool hasMips)
         mCurrTexture = texture;
         glBindTexture(sGLTextureType[type], texture);
         mHasMipMaps = hasMips;
-        setTextureColorSpace(mTexColorSpace);
     }
     return true;
 }
@@ -424,8 +419,6 @@ void LLTexUnit::unbind(eTextureType type)
     {
         mCurrTexture = 0;
 
-        // Always make sure our texture color space is reset to linear.  SRGB sampling should be opt-in in the vast majority of cases.  Also prevents color space "popping".
-        mTexColorSpace = TCS_LINEAR;
         if (type == LLTexUnit::TT_TEXTURE)
         {
             glBindTexture(sGLTextureType[type], sWhiteTexture);
@@ -447,8 +440,6 @@ void LLTexUnit::unbindFast(eTextureType type)
     {
         mCurrTexture = 0;
 
-        // Always make sure our texture color space is reset to linear.  SRGB sampling should be opt-in in the vast majority of cases.  Also prevents color space "popping".
-        mTexColorSpace = TCS_LINEAR;
         if (type == LLTexUnit::TT_TEXTURE)
         {
             glBindTexture(sGLTextureType[type], sWhiteTexture);
@@ -642,11 +633,6 @@ void LLTexUnit::debugTextureUnit(void)
     }
 }
 
-void LLTexUnit::setTextureColorSpace(eTextureColorSpace space)
-{
-    mTexColorSpace = space;
-}
-
 LLLightState::LLLightState(S32 index)
 : mIndex(index),
   mEnabled(false),
@@ -1514,7 +1500,7 @@ LLLightState* LLRender::getLight(U32 index)
 
 void LLRender::setAmbientLightColor(const LLColor4& color)
 {
-    LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE;
     if (color != mAmbientLightColor)
     {
         ++mLightHash;
@@ -1543,6 +1529,30 @@ void LLRender::clearErrors()
     }
 }
 
+void LLRender::beginList(std::list<LLVertexBufferData> *list)
+{
+    if (sBufferDataList)
+    {
+        LL_ERRS() << "beginList called while another list is open." << LL_ENDL;
+    }
+    llassert(LLGLSLShader::sCurBoundShaderPtr == &gUIProgram);
+    flush();
+    sBufferDataList = list;
+}
+
+void LLRender::endList()
+{
+    if (sBufferDataList)
+    {
+        flush();
+        sBufferDataList = nullptr;
+    }
+    else
+    {
+        llassert(false); // endList called without an open list
+    }
+}
+
 void LLRender::begin(const GLuint& mode)
 {
     if (mode != mMode)
@@ -1585,6 +1595,7 @@ void LLRender::end()
         flush();
     }
 }
+
 void LLRender::flush()
 {
     STOP_GLERROR;
@@ -1633,127 +1644,166 @@ void LLRender::flush()
         if (mBuffer)
         {
 
-            HBXXH64 hash;
+            LLVertexBuffer *vb;
+
             U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask;
 
+            if (sBufferDataList)
+            {
+                vb = genBuffer(attribute_mask, count);
+                sBufferDataList->emplace_back(
+                    vb,
+                    mMode,
+                    count,
+                    gGL.getTexUnit(0)->mCurrTexture,
+                    mMatrix[MM_MODELVIEW][mMatIdx[MM_MODELVIEW]],
+                    mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]],
+                    mMatrix[MM_TEXTURE0][mMatIdx[MM_TEXTURE0]]
+                    );
+            }
+            else
             {
-                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash");
+                vb = bufferfromCache(attribute_mask, count);
+            }
 
-                hash.update((U8*)mVerticesp.get(), count * sizeof(LLVector4a));
-                if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
-                {
-                    hash.update((U8*)mTexcoordsp.get(), count * sizeof(LLVector2));
-                }
+            drawBuffer(vb, mMode, count);
+        }
+        else
+        {
+            // mBuffer is present in main thread and not present in an image thread
+            LL_ERRS() << "A flush call from outside main rendering thread" << LL_ENDL;
+        }
 
-                if (attribute_mask & LLVertexBuffer::MAP_COLOR)
-                {
-                    hash.update((U8*)mColorsp.get(), count * sizeof(LLColor4U));
-                }
+        resetStriders(count);
+    }
+}
 
-                hash.finalize();
-            }
+LLVertexBuffer* LLRender::bufferfromCache(U32 attribute_mask, U32 count)
+{
+    LLVertexBuffer *vb = nullptr;
+    HBXXH64 hash;
 
+    {
+        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash");
 
-            U64 vhash = hash.digest();
+        hash.update((U8*)mVerticesp.get(), count * sizeof(LLVector4a));
+        if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
+        {
+            hash.update((U8*)mTexcoordsp.get(), count * sizeof(LLVector2));
+        }
 
-            // check the VB cache before making a new vertex buffer
-            // This is a giant hack to deal with (mostly) our terrible UI rendering code
-            // that was built on top of OpenGL immediate mode.  Huge performance wins
-            // can be had by not uploading geometry to VRAM unless absolutely necessary.
-            // Most of our usage of the "immediate mode" style draw calls is actually
-            // sending the same geometry over and over again.
-            // To leverage this, we maintain a running hash of the vertex stream being
-            // built up before a flush, and then check that hash against a VB
-            // cache just before creating a vertex buffer in VRAM
-            std::unordered_map<U64, LLVBCache>::iterator cache = sVBCache.find(vhash);
+        if (attribute_mask & LLVertexBuffer::MAP_COLOR)
+        {
+            hash.update((U8*)mColorsp.get(), count * sizeof(LLColor4U));
+        }
 
-            LLPointer<LLVertexBuffer> vb;
+        hash.finalize();
+    }
 
-            if (cache != sVBCache.end())
-            {
-                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hit");
-                // cache hit, just use the cached buffer
-                vb = cache->second.vb;
-                cache->second.touched = std::chrono::steady_clock::now();
-            }
-            else
-            {
-                LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss");
-                vb = new LLVertexBuffer(attribute_mask);
-                vb->allocateBuffer(count, 0);
+    U64 vhash = hash.digest();
 
-                vb->setBuffer();
+    // check the VB cache before making a new vertex buffer
+    // This is a giant hack to deal with (mostly) our terrible UI rendering code
+    // that was built on top of OpenGL immediate mode.  Huge performance wins
+    // can be had by not uploading geometry to VRAM unless absolutely necessary.
+    // Most of our usage of the "immediate mode" style draw calls is actually
+    // sending the same geometry over and over again.
+    // To leverage this, we maintain a running hash of the vertex stream being
+    // built up before a flush, and then check that hash against a VB
+    // cache just before creating a vertex buffer in VRAM
+    std::unordered_map<U64, LLVBCache>::iterator cache = sVBCache.find(vhash);
 
-                vb->setPositionData((LLVector4a*) mVerticesp.get());
+    if (cache != sVBCache.end())
+    {
+        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hit");
+        // cache hit, just use the cached buffer
+        vb = cache->second.vb;
+        cache->second.touched = std::chrono::steady_clock::now();
+    }
+    else
+    {
+        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss");
+        vb = genBuffer(attribute_mask, count);
+
+        sVBCache[vhash] = { vb , std::chrono::steady_clock::now() };
+
+        static U32 miss_count = 0;
+        miss_count++;
+        if (miss_count > 1024)
+        {
+            LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache clean");
+            miss_count = 0;
+            auto now = std::chrono::steady_clock::now();
 
-                if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
+            using namespace std::chrono_literals;
+            // every 1024 misses, clean the cache of any VBs that haven't been touched in the last second
+            for (std::unordered_map<U64, LLVBCache>::iterator iter = sVBCache.begin(); iter != sVBCache.end(); )
+            {
+                if (now - iter->second.touched > 1s)
                 {
-                    vb->setTexCoord0Data(mTexcoordsp.get());
+                    iter = sVBCache.erase(iter);
                 }
-
-                if (attribute_mask & LLVertexBuffer::MAP_COLOR)
+                else
                 {
-                    vb->setColorData(mColorsp.get());
+                    ++iter;
                 }
+            }
+        }
+    }
+    return vb;
+}
 
-#if LL_DARWIN
-                vb->unmapBuffer();
-#endif
-                vb->unbind();
+LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count)
+{
+    LLVertexBuffer * vb = new LLVertexBuffer(attribute_mask);
+    vb->allocateBuffer(count, 0);
 
-                sVBCache[vhash] = { vb , std::chrono::steady_clock::now() };
+    vb->setBuffer();
 
-                static U32 miss_count = 0;
-                miss_count++;
-                if (miss_count > 1024)
-                {
-                    LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache clean");
-                    miss_count = 0;
-                    auto now = std::chrono::steady_clock::now();
+    vb->setPositionData((LLVector4a*)mVerticesp.get());
 
-                    using namespace std::chrono_literals;
-                    // every 1024 misses, clean the cache of any VBs that haven't been touched in the last second
-                    for (std::unordered_map<U64, LLVBCache>::iterator iter = sVBCache.begin(); iter != sVBCache.end(); )
-                    {
-                        if (now - iter->second.touched > 1s)
-                        {
-                            iter = sVBCache.erase(iter);
-                        }
-                        else
-                        {
-                            ++iter;
-                        }
-                    }
-                }
-            }
+    if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0)
+    {
+        vb->setTexCoord0Data(mTexcoordsp.get());
+    }
 
-            vb->setBuffer();
+    if (attribute_mask & LLVertexBuffer::MAP_COLOR)
+    {
+        vb->setColorData(mColorsp.get());
+    }
 
-            if (mMode == LLRender::QUADS && sGLCoreProfile)
-            {
-                vb->drawArrays(LLRender::TRIANGLES, 0, count);
-                mQuadCycle = 1;
-            }
-            else
-            {
-                vb->drawArrays(mMode, 0, count);
-            }
-        }
-        else
-        {
-            // mBuffer is present in main thread and not present in an image thread
-            LL_ERRS() << "A flush call from outside main rendering thread" << LL_ENDL;
-        }
+#if LL_DARWIN
+    vb->unmapBuffer();
+#endif
+    vb->unbind();
 
+    return vb;
+}
 
-        mVerticesp[0] = mVerticesp[count];
-        mTexcoordsp[0] = mTexcoordsp[count];
-        mColorsp[0] = mColorsp[count];
+void LLRender::drawBuffer(LLVertexBuffer* vb, U32 mode, S32 count)
+{
+    vb->setBuffer();
 
-        mCount = 0;
+    if (mode == LLRender::QUADS && sGLCoreProfile)
+    {
+        vb->drawArrays(LLRender::TRIANGLES, 0, count);
+        mQuadCycle = 1;
+    }
+    else
+    {
+        vb->drawArrays(mode, 0, count);
     }
 }
 
+void LLRender::resetStriders(S32 count)
+{
+    mVerticesp[0] = mVerticesp[count];
+    mTexcoordsp[0] = mTexcoordsp[count];
+    mColorsp[0] = mColorsp[count];
+
+    mCount = 0;
+}
+
 void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z)
 {
     //the range of mVerticesp, mColorsp and mTexcoordsp is [0, 4095]
diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h
index be9f3895e7..39c13e328a 100644
--- a/indra/llrender/llrender.h
+++ b/indra/llrender/llrender.h
@@ -45,12 +45,14 @@
 #include "glh/glh_linear.h"
 
 #include <array>
+#include <list>
 
 class LLVertexBuffer;
 class LLCubeMap;
 class LLImageGL;
 class LLRenderTarget;
-class LLTexture ;
+class LLTexture;
+class LLVertexBufferData;
 
 #define LL_MATRIX_STACK_DEPTH 32
 
@@ -219,17 +221,12 @@ public:
 
     void setHasMipMaps(bool hasMips) { mHasMipMaps = hasMips; }
 
-    void setTextureColorSpace(eTextureColorSpace space);
-
-    eTextureColorSpace getCurrColorSpace() { return mTexColorSpace; }
-
 protected:
     friend class LLRender;
 
     S32                 mIndex;
     U32                 mCurrTexture;
     eTextureType        mCurrTexType;
-    eTextureColorSpace  mTexColorSpace;
     S32                 mCurrColorScale;
     S32                 mCurrAlphaScale;
     bool                mHasMipMaps;
@@ -293,11 +290,18 @@ public:
 
     enum eTexIndex : U8
     {
-        DIFFUSE_MAP           = 0,
-        ALTERNATE_DIFFUSE_MAP = 1,
-        NORMAL_MAP            = 1,
-        SPECULAR_MAP          = 2,
-        NUM_TEXTURE_CHANNELS  = 3,
+        // Channels for material textures
+        DIFFUSE_MAP            = 0,
+        ALTERNATE_DIFFUSE_MAP  = 1,
+        NORMAL_MAP             = 1,
+        SPECULAR_MAP           = 2,
+        // Channels for PBR textures
+        BASECOLOR_MAP          = 3,
+        METALLIC_ROUGHNESS_MAP = 4,
+        GLTF_NORMAL_MAP        = 5,
+        EMISSIVE_MAP           = 6,
+        // Total number of channels
+        NUM_TEXTURE_CHANNELS   = 7,
     };
 
     enum eVolumeTexIndex : U8
@@ -413,8 +417,15 @@ public:
 
     void flush();
 
+    // if list is set, will store buffers in list for later use, if list isn't set, will use cache
+    void beginList(std::list<LLVertexBufferData> *list);
+    void endList();
+
     void begin(const GLuint& mode);
     void end();
+
+    U8 getMode() const { return mMode; }
+
     void vertex2i(const GLint& x, const GLint& y);
     void vertex2f(const GLfloat& x, const GLfloat& y);
     void vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z);
@@ -483,6 +494,11 @@ public:
 private:
     friend class LLLightState;
 
+    LLVertexBuffer* bufferfromCache(U32 attribute_mask, U32 count);
+    LLVertexBuffer* genBuffer(U32 attribute_mask, S32 count);
+    void drawBuffer(LLVertexBuffer* vb, U32 mode, S32 count);
+    void resetStriders(S32 count);
+
     eMatrixMode mMatrixMode;
     U32 mMatIdx[NUM_MATRIX_MODES];
     U32 mMatHash[NUM_MATRIX_MODES];
@@ -513,7 +529,6 @@ private:
 
     std::vector<LLVector3> mUIOffset;
     std::vector<LLVector3> mUIScale;
-
 };
 
 extern F32 gGLModelView[16];
diff --git a/indra/llrender/llrender2dutils.cpp b/indra/llrender/llrender2dutils.cpp
index ef02fbd071..428370057e 100644
--- a/indra/llrender/llrender2dutils.cpp
+++ b/indra/llrender/llrender2dutils.cpp
@@ -364,7 +364,7 @@ void gl_draw_scaled_image_with_border(S32 x, S32 y, S32 width, S32 height, LLTex
     {
         // add in offset of current image to current UI translation
         const LLVector3 ui_scale = gGL.getUIScale();
-        const LLVector3 ui_translation = (gGL.getUITranslation() + LLVector3(x, y, 0.f)).scaledVec(ui_scale);
+        const LLVector3 ui_translation = (gGL.getUITranslation() + LLVector3((F32)x, (F32)y, 0.f)).scaledVec(ui_scale);
 
         F32 uv_width = uv_outer_rect.getWidth();
         F32 uv_height = uv_outer_rect.getHeight();
@@ -375,8 +375,8 @@ void gl_draw_scaled_image_with_border(S32 x, S32 y, S32 width, S32 height, LLTex
                                 uv_outer_rect.mLeft + (center_rect.mRight * uv_width),
                                 uv_outer_rect.mBottom + (center_rect.mBottom * uv_height));
 
-        F32 image_width = image->getWidth(0);
-        F32 image_height = image->getHeight(0);
+        F32 image_width = (F32)image->getWidth(0);
+        F32 image_height = (F32)image->getHeight(0);
 
         S32 image_natural_width = ll_round(image_width * uv_width);
         S32 image_natural_height = ll_round(image_height * uv_height);
@@ -413,10 +413,10 @@ void gl_draw_scaled_image_with_border(S32 x, S32 y, S32 width, S32 height, LLTex
             draw_center_rect.setCenterAndSize(uv_center_rect.getCenterX() * width, uv_center_rect.getCenterY() * height, scaled_width, scaled_height);
         }
 
-        draw_center_rect.mLeft   = ll_round(ui_translation.mV[VX] + (F32)draw_center_rect.mLeft * ui_scale.mV[VX]);
-        draw_center_rect.mTop    = ll_round(ui_translation.mV[VY] + (F32)draw_center_rect.mTop * ui_scale.mV[VY]);
-        draw_center_rect.mRight  = ll_round(ui_translation.mV[VX] + (F32)draw_center_rect.mRight * ui_scale.mV[VX]);
-        draw_center_rect.mBottom = ll_round(ui_translation.mV[VY] + (F32)draw_center_rect.mBottom * ui_scale.mV[VY]);
+        draw_center_rect.mLeft   = (F32)ll_round(ui_translation.mV[VX] + (F32)draw_center_rect.mLeft * ui_scale.mV[VX]);
+        draw_center_rect.mTop    = (F32)ll_round(ui_translation.mV[VY] + (F32)draw_center_rect.mTop * ui_scale.mV[VY]);
+        draw_center_rect.mRight  = (F32)ll_round(ui_translation.mV[VX] + (F32)draw_center_rect.mRight * ui_scale.mV[VX]);
+        draw_center_rect.mBottom = (F32)ll_round(ui_translation.mV[VY] + (F32)draw_center_rect.mBottom * ui_scale.mV[VY]);
 
         LLRectf draw_outer_rect(ui_translation.mV[VX],
                                 ui_translation.mV[VY] + height * ui_scale.mV[VY],
@@ -726,7 +726,7 @@ void gl_line_3d( const LLVector3& start, const LLVector3& end, const LLColor4& c
     }
     gGL.end();
 
-    LLRender2D::getInstance()->setLineWidth(1.f);
+    LLRender2D::setLineWidth(1.f);
 }
 
 void gl_arc_2d(F32 center_x, F32 center_y, F32 radius, S32 steps, bool filled, F32 start_angle, F32 end_angle)
diff --git a/indra/llrender/llrender2dutils.h b/indra/llrender/llrender2dutils.h
index 0d3efc38d6..096e7584f1 100644
--- a/indra/llrender/llrender2dutils.h
+++ b/indra/llrender/llrender2dutils.h
@@ -122,12 +122,13 @@ inline void gl_rect_2d_offset_local( const LLRect& rect, S32 pixel_offset, bool
 
 class LLImageProviderInterface;
 
-class LLRender2D : public LLParamSingleton<LLRender2D>
+class LLRender2D : public LLSimpleton<LLRender2D>
 {
-    LLSINGLETON(LLRender2D, LLImageProviderInterface* image_provider);
     LOG_CLASS(LLRender2D);
-    ~LLRender2D();
 public:
+    LLRender2D(LLImageProviderInterface* image_provider);
+    ~LLRender2D();
+
     static void pushMatrix();
     static void popMatrix();
     static void loadIdentity();
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index efdde77a32..38bc5ff331 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -123,7 +123,7 @@ bool LLRenderTarget::allocate(U32 resx, U32 resy, U32 color_fmt, bool depth, LLT
 
     if (mGenerateMipMaps != LLTexUnit::TMG_NONE) {
         // Calculate the number of mip levels based upon resolution that we should have.
-        mMipLevels = 1 + floor(log10((float)llmax(mResX, mResY))/log10(2.0));
+        mMipLevels = 1 + (U32)floor(log10((float)llmax(mResX, mResY)) / log10(2.0));
     }
 
     if (depth)
@@ -426,14 +426,17 @@ void LLRenderTarget::bindTarget()
                             GL_COLOR_ATTACHMENT1,
                             GL_COLOR_ATTACHMENT2,
                             GL_COLOR_ATTACHMENT3};
-    glDrawBuffers(static_cast<GLsizei>(mTex.size()), drawbuffers);
 
     if (mTex.empty())
     { //no color buffer to draw to
         glDrawBuffer(GL_NONE);
         glReadBuffer(GL_NONE);
     }
-
+    else
+    {
+        glDrawBuffers(static_cast<GLsizei>(mTex.size()), drawbuffers);
+        glReadBuffer(GL_COLOR_ATTACHMENT0);
+    }
     check_framebuffer_status();
 
     glViewport(0, 0, mResX, mResY);
@@ -472,12 +475,10 @@ void LLRenderTarget::clear(U32 mask_in)
 
 U32 LLRenderTarget::getTexture(U32 attachment) const
 {
-    if (attachment > mTex.size()-1)
-    {
-        LL_ERRS() << "Invalid attachment index." << LL_ENDL;
-    }
-    if (mTex.empty())
+    if (attachment >= mTex.size())
     {
+        LL_WARNS() << "Invalid attachment index " << attachment << " for size " << mTex.size() << LL_ENDL;
+        llassert(false);
         return 0;
     }
     return mTex[attachment];
@@ -508,7 +509,6 @@ void LLRenderTarget::bindTexture(U32 index, S32 channel, LLTexUnit::eTextureFilt
     }
 
     gGL.getTexUnit(channel)->setTextureFilteringOption(filter_options);
-    gGL.getTexUnit(channel)->setTextureColorSpace(isSRGB ? LLTexUnit::TCS_SRGB : LLTexUnit::TCS_LINEAR);
 }
 
 void LLRenderTarget::flush()
@@ -519,7 +519,8 @@ void LLRenderTarget::flush()
     llassert(sCurFBO == mFBO);
     llassert(sBoundTarget == this);
 
-    if (mGenerateMipMaps == LLTexUnit::TMG_AUTO) {
+    if (mGenerateMipMaps == LLTexUnit::TMG_AUTO)
+    {
         LL_PROFILE_GPU_ZONE("rt generate mipmaps");
         bindTexture(0, 0, LLTexUnit::TFO_TRILINEAR);
         glGenerateMipmap(GL_TEXTURE_2D);
@@ -540,6 +541,8 @@ void LLRenderTarget::flush()
         glViewport(gGLViewport[0], gGLViewport[1], gGLViewport[2], gGLViewport[3]);
         sCurResX = gGLViewport[2];
         sCurResY = gGLViewport[3];
+        glReadBuffer(GL_BACK);
+        glDrawBuffer(GL_BACK);
     }
 }
 
diff --git a/indra/llrender/llrendertarget.h b/indra/llrender/llrendertarget.h
index 340276a752..cd3290cf66 100644
--- a/indra/llrender/llrendertarget.h
+++ b/indra/llrender/llrendertarget.h
@@ -61,7 +61,7 @@
 class LLRenderTarget
 {
 public:
-    //whether or not to use FBO implementation
+    // Whether or not to use FBO implementation
     static bool sUseFBO;
     static U32 sBytesAllocated;
     static U32 sCurFBO;
@@ -172,6 +172,8 @@ public:
     // *HACK
     void swapFBORefs(LLRenderTarget& other);
 
+    static LLRenderTarget* sBoundTarget;
+
 protected:
     U32 mResX;
     U32 mResY;
@@ -186,8 +188,6 @@ protected:
     U32 mMipLevels;
 
     LLTexUnit::eTextureType mUsage;
-
-    static LLRenderTarget* sBoundTarget;
 };
 
 #endif
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index a8e9f20b40..150277c8df 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -44,7 +44,6 @@ using std::make_pair;
 using std::string;
 
 LLShaderMgr * LLShaderMgr::sInstance = NULL;
-bool LLShaderMgr::sMirrorsEnabled = false;
 
 LLShaderMgr::LLShaderMgr()
 {
@@ -596,13 +595,15 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev
                 extra_code_text[extra_code_count++] = strdup("precision highp float;\n");
             }
         }
-
-        extra_code_text[extra_code_count++] = strdup("#define FXAA_GLSL_130 1\n");
     }
 
-    if (sMirrorsEnabled)
+    if (type == GL_FRAGMENT_SHADER)
+    {
+        extra_code_text[extra_code_count++] = strdup("#define FRAGMENT_SHADER 1\n");
+    }
+    else
     {
-        extra_code_text[extra_code_count++] = strdup("#define HERO_PROBES 1\n");
+        extra_code_text[extra_code_count++] = strdup("#define VERTEX_SHADER 1\n");
     }
 
     // Use alpha float to store bit flags
@@ -1003,7 +1004,7 @@ void LLShaderMgr::initShaderCache(bool enabled, const LLUUID& old_cache_version,
                     ProgramBinaryData binary_info = ProgramBinaryData();
                     binary_info.mBinaryFormat = data_pair.second["binary_format"].asInteger();
                     binary_info.mBinaryLength = data_pair.second["binary_size"].asInteger();
-                    binary_info.mLastUsedTime = data_pair.second["last_used"].asReal();
+                    binary_info.mLastUsedTime = (F32)data_pair.second["last_used"].asReal();
                     mShaderBinaryCache.insert_or_assign(LLUUID(data_pair.first), binary_info);
                 }
             }
@@ -1034,7 +1035,7 @@ void LLShaderMgr::persistShaderCacheMetadata()
     LLSD out = LLSD::emptyMap();
 
     static const F32 LRU_TIME = (60.f * 60.f) * 24.f * 7.f; // 14 days
-    const F32 current_time = LLTimer::getTotalSeconds();
+    const F32 current_time = (F32)LLTimer::getTotalSeconds();
     for (auto it = mShaderBinaryCache.begin(); it != mShaderBinaryCache.end();)
     {
         const ProgramBinaryData& shader_metadata = it->second;
@@ -1093,7 +1094,7 @@ bool LLShaderMgr::loadCachedProgramBinary(LLGLSLShader* shader)
                     glGetProgramiv(shader->mProgramObject, GL_LINK_STATUS, &success);
                     if (error == GL_NO_ERROR && success == GL_TRUE)
                     {
-                        binary_iter->second.mLastUsedTime = LLTimer::getTotalSeconds();
+                        binary_iter->second.mLastUsedTime = (F32)LLTimer::getTotalSeconds();
                         LL_INFOS() << "Loaded cached binary for shader: " << shader->mName << LL_ENDL;
                         return true;
                     }
@@ -1131,7 +1132,7 @@ bool LLShaderMgr::saveCachedProgramBinary(LLGLSLShader* shader)
                 fwrite(program_binary.data(), sizeof(U8), program_binary.size(), outfile);
                 outfile.close();
 
-                binary_info.mLastUsedTime = LLTimer::getTotalSeconds();
+                binary_info.mLastUsedTime = (F32)LLTimer::getTotalSeconds();
 
                 mShaderBinaryCache.insert_or_assign(shader->mShaderHash, binary_info);
                 return true;
@@ -1415,6 +1416,7 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("detail_3");
 
     mReservedUniforms.push_back("alpha_ramp");
+    mReservedUniforms.push_back("paint_map");
 
     mReservedUniforms.push_back("detail_0_base_color");
     mReservedUniforms.push_back("detail_1_base_color");
@@ -1439,6 +1441,8 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("emissiveColors");
     mReservedUniforms.push_back("minimum_alphas");
 
+    mReservedUniforms.push_back("region_scale");
+
     mReservedUniforms.push_back("origin");
     mReservedUniforms.push_back("display_gamma");
 
@@ -1473,6 +1477,11 @@ void LLShaderMgr::initAttribsAndUniforms()
 
     mReservedUniforms.push_back("debug_normal_draw_length");
 
+    mReservedUniforms.push_back("edgesTex");
+    mReservedUniforms.push_back("areaTex");
+    mReservedUniforms.push_back("searchTex");
+    mReservedUniforms.push_back("blendTex");
+
     llassert(mReservedUniforms.size() == END_RESERVED_UNIFORMS);
 
     std::set<std::string> dupe_check;
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index fe6137c448..0eb9db6715 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -267,6 +267,7 @@ public:
         TERRAIN_DETAIL3,                    //  "detail_3"
 
         TERRAIN_ALPHARAMP,                  //  "alpha_ramp"
+        TERRAIN_PAINTMAP,                   //  "paint_map"
 
         TERRAIN_DETAIL0_BASE_COLOR,                //  "detail_0_base_color" (GLTF)
         TERRAIN_DETAIL1_BASE_COLOR,                //  "detail_1_base_color" (GLTF)
@@ -291,6 +292,8 @@ public:
         TERRAIN_EMISSIVE_COLORS,                   //  "emissiveColors" (GLTF)
         TERRAIN_MINIMUM_ALPHAS,                    //  "minimum_alphas" (GLTF)
 
+        REGION_SCALE,                              //  "region_scale" (GLTF)
+
         SHINY_ORIGIN,                       //  "origin"
         DISPLAY_GAMMA,                      //  "display_gamma"
 
@@ -329,6 +332,11 @@ public:
 
         DEBUG_NORMAL_DRAW_LENGTH,           //  "debug_normal_draw_length"
 
+        SMAA_EDGE_TEX,                      //  "edgesTex"
+        SMAA_AREA_TEX,                      //  "areaTex"
+        SMAA_SEARCH_TEX,                    //  "searchTex"
+        SMAA_BLEND_TEX,                     //  "blendTex"
+
         END_RESERVED_UNIFORMS
     } eGLSLReservedUniforms;
     // clang-format on
@@ -378,7 +386,6 @@ public:
     bool mShaderCacheInitialized = false;
     bool mShaderCacheEnabled = false;
     std::string mShaderCacheDir;
-    static bool sMirrorsEnabled;
 
 protected:
 
diff --git a/indra/llrender/lluiimage.cpp b/indra/llrender/lluiimage.cpp
index bcf665ca18..dc18bf16bf 100644
--- a/indra/llrender/lluiimage.cpp
+++ b/indra/llrender/lluiimage.cpp
@@ -81,10 +81,10 @@ void LLUIImage::draw3D(const LLVector3& origin_agent, const LLVector3& x_axis, c
          }
     }
 
-    LLRender2D::getInstance()->pushMatrix();
+    LLRender2D::pushMatrix();
     {
-        LLVector3 rect_origin = origin_agent + (rect.mLeft * x_axis) + (rect.mBottom * y_axis);
-        LLRender2D::getInstance()->translate(rect_origin.mV[VX],
+        LLVector3 rect_origin = origin_agent + ((F32)rect.mLeft * x_axis) + ((F32)rect.mBottom * y_axis);
+        LLRender2D::translate(rect_origin.mV[VX],
                                             rect_origin.mV[VY],
                                             rect_origin.mV[VZ]);
         gGL.getTexUnit(0)->bind(getImage());
@@ -100,10 +100,10 @@ void LLUIImage::draw3D(const LLVector3& origin_agent, const LLVector3& x_axis, c
                                         (rect.getHeight() - (border_height * border_scale * 0.5f)) / (F32)rect.getHeight(),
                                         (rect.getWidth() - (border_width * border_scale * 0.5f)) / (F32)rect.getWidth(),
                                         (border_height * border_scale * 0.5f) / (F32)rect.getHeight()),
-                                rect.getWidth() * x_axis,
-                                rect.getHeight() * y_axis);
+                                (F32)rect.getWidth() * x_axis,
+                                (F32)rect.getHeight() * y_axis);
 
-    } LLRender2D::getInstance()->popMatrix();
+    } LLRender2D::popMatrix();
 }
 
 //#include "lluiimage.inl"
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 2eb7c21f77..156e300853 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -570,6 +570,54 @@ public:
 
 static LLVBOPool* sVBOPool = nullptr;
 
+void LLVertexBufferData::draw()
+{
+    if (!mVB)
+    {
+        llassert(false);
+        // Not supposed to happen, check buffer generation
+        return;
+    }
+
+    if (mTexName)
+    {
+        gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mTexName);
+    }
+    else
+    {
+        gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+    }
+
+    gGL.matrixMode(LLRender::MM_MODELVIEW);
+    gGL.pushMatrix();
+    gGL.loadMatrix(mModelView.m);
+    gGL.matrixMode(LLRender::MM_PROJECTION);
+    gGL.pushMatrix();
+    gGL.loadMatrix(mProjection.m);
+    gGL.matrixMode(LLRender::MM_TEXTURE0);
+    gGL.pushMatrix();
+    gGL.loadMatrix(mTexture0.m);
+
+    mVB->setBuffer();
+
+    if (mMode == LLRender::QUADS && LLRender::sGLCoreProfile)
+    {
+        mVB->drawArrays(LLRender::TRIANGLES, 0, mCount);
+    }
+    else
+    {
+        mVB->drawArrays(mMode, 0, mCount);
+    }
+
+    gGL.popMatrix();
+    gGL.matrixMode(LLRender::MM_PROJECTION);
+    gGL.popMatrix();
+    gGL.matrixMode(LLRender::MM_MODELVIEW);
+    gGL.popMatrix();
+}
+
+//============================================================================
+
 //static
 U64 LLVertexBuffer::getBytesAllocated()
 {
@@ -794,6 +842,18 @@ void LLVertexBuffer::setLabel(const char* label) {
 }
 #endif
 
+void LLVertexBuffer::clone(LLVertexBuffer& target) const
+{
+    target.mTypeMask = mTypeMask;
+    target.mIndicesType = mIndicesType;
+    target.mIndicesStride = mIndicesStride;
+    if (target.getNumVerts() != getNumVerts() ||
+        target.getNumIndices() != getNumIndices())
+    {
+        target.allocateBuffer(getNumVerts(), getNumIndices());
+    }
+}
+
 void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const
 {
     llassert(validateRange(start, end, count, indices_offset));
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 49500e28ce..2a4affdc60 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -53,6 +53,37 @@
 //============================================================================
 // base class
 class LLPrivateMemoryPool;
+class LLVertexBuffer;
+
+class LLVertexBufferData
+{
+public:
+    LLVertexBufferData()
+        : mVB(nullptr)
+        , mMode(0)
+        , mCount(0)
+        , mTexName(0)
+    {}
+    LLVertexBufferData(LLVertexBuffer* buffer, U8 mode, U32 count, U32 tex_name, glh::matrix4f model_view, glh::matrix4f projection, glh::matrix4f texture0)
+        : mVB(buffer)
+        , mMode(mode)
+        , mCount(count)
+        , mTexName(tex_name)
+        , mProjection(model_view)
+        , mModelView(projection)
+        , mTexture0(texture0)
+    {}
+    void draw();
+    LLPointer<LLVertexBuffer> mVB;
+    U8 mMode;
+    U32 mCount;
+    U32 mTexName;
+    glh::matrix4f mProjection;
+    glh::matrix4f mModelView;
+    glh::matrix4f mTexture0;
+};
+typedef std::list<LLVertexBufferData> buffer_data_list_t;
+
 class LLVertexBuffer final : public LLRefCount
 {
 public:
@@ -163,13 +194,13 @@ public:
 
     // set for rendering
     // assumes (and will assert on) the following:
-    //      - this buffer has no pending unampBuffer call
+    //      - this buffer has no pending unmapBuffer call
     //      - a shader is currently bound
     //      - This buffer has sufficient attributes within it to satisfy the needs of the currently bound shader
     void    setBuffer();
 
     // Only call each getVertexPointer, etc, once before calling unmapBuffer()
-    // call unmapBuffer() after calls to getXXXStrider() before any cals to setBuffer()
+    // call unmapBuffer() after calls to getXXXStrider() before any calls to setBuffer()
     // example:
     //   vb->getVertexBuffer(verts);
     //   vb->getNormalStrider(norms);
@@ -218,12 +249,12 @@ public:
     U32 getNumIndices() const               { return mNumIndices; }
 
     U32 getTypeMask() const                 { return mTypeMask; }
-    bool hasDataType(AttributeType type) const      { return ((1 << type) & getTypeMask()); }
+    bool hasDataType(AttributeType type) const { return ((1 << type) & getTypeMask()); }
     U32 getSize() const                     { return mSize; }
     U32 getIndicesSize() const              { return mIndicesSize; }
     U8* getMappedData() const               { return mMappedData; }
     U8* getMappedIndices() const            { return mMappedIndexData; }
-    U32 getOffset(AttributeType type) const         { return mOffsets[type]; }
+    U32 getOffset(AttributeType type) const { return mOffsets[type]; }
 
     // these functions assume (and assert on) the current VBO being bound
     // Detailed error checking can be enabled by setting gDebugGL to true
@@ -242,6 +273,7 @@ public:
     void setLabel(const char* label);
     #endif
 
+    void clone(LLVertexBuffer& target) const;
 
 protected:
     U32     mGLBuffer = 0;      // GL VBO handle