diff options
Diffstat (limited to 'indra/llrender')
36 files changed, 2283 insertions, 1507 deletions
diff --git a/indra/llrender/CMakeLists.txt b/indra/llrender/CMakeLists.txt index 7424356057..d48b508ddc 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 0242fe60ce..fb35e002df 100644 --- a/indra/llrender/llcubemaparray.cpp +++ b/indra/llrender/llcubemaparray.cpp @@ -105,16 +105,48 @@ LLCubeMapArray::LLCubeMapArray() } +LLCubeMapArray::LLCubeMapArray(LLCubeMapArray& lhs, U32 width, U32 count) : mTextureStage(0) +{ + mWidth = width; + mCount = count; + + // Allocate a new cubemap array with the same criteria as the incoming cubemap array + allocate(mWidth, lhs.mImage->getComponents(), count, lhs.mImage->getUseMipMaps(), lhs.mHDR); + + // Copy each cubemap from the incoming array to the new array + U32 min_count = std::min(count, lhs.mCount); + for (U32 i = 0; i < min_count * 6; ++i) + { + U32 src_resolution = lhs.mWidth; + U32 dst_resolution = mWidth; + { + GLint components = GL_RGB; + if (mImage->getComponents() == 4) + components = GL_RGBA; + GLint format = GL_RGB; + + // Handle different resolutions by scaling the image + LLPointer<LLImageRaw> src_image = new LLImageRaw(lhs.mWidth, lhs.mWidth, lhs.mImage->getComponents()); + glGetTexImage(GL_TEXTURE_CUBE_MAP_ARRAY, 0, components, GL_UNSIGNED_BYTE, src_image->getData()); + + LLPointer<LLImageRaw> scaled_image = src_image->scaled(mWidth, mWidth); + glTexSubImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, 0, 0, 0, i, mWidth, mWidth, 1, components, GL_UNSIGNED_BYTE, scaled_image->getData()); + } + } +} + LLCubeMapArray::~LLCubeMapArray() { } -void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool use_mips) +void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool use_mips, bool hdr) { U32 texname = 0; mWidth = resolution; mCount = count; + mHDR = hdr; + LLImageGL::generateTextures(1, &texname); mImage = new LLImageGL(resolution, resolution, components, use_mips); @@ -125,17 +157,19 @@ void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool us mImage->setHasMipMaps(use_mips); bind(0); - - U32 format = components == 4 ? GL_RGBA16F : GL_RGB16F; - - U32 mip = 0; - free_cur_tex_image(); - while (resolution >= 1) + U32 format = components == 4 ? GL_RGBA16F : GL_R11F_G11F_B10F; + if (!hdr) + { + format = components == 4 ? GL_RGBA8 : GL_RGB8; + } + U32 mip = 0; + U32 mip_resolution = resolution; + while (mip_resolution >= 1) { #if GL_VERSION_4_0 - 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); #endif @@ -143,11 +177,11 @@ void LLCubeMapArray::allocate(U32 resolution, U32 components, U32 count, bool us { 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/llcubemaparray.h b/indra/llrender/llcubemaparray.h index 675aaaf07c..6b4288cb23 100644 --- a/indra/llrender/llcubemaparray.h +++ b/indra/llrender/llcubemaparray.h @@ -36,6 +36,7 @@ class LLCubeMapArray : public LLRefCount { public: LLCubeMapArray(); + LLCubeMapArray(LLCubeMapArray& lhs, U32 width, U32 count); static GLenum sTargets[6]; @@ -52,7 +53,7 @@ public: // components - number of components per pixel // count - number of cube maps in the array // use_mips - if true, mipmaps will be allocated for this cube map array and anisotropic filtering will be used - void allocate(U32 res, U32 components, U32 count, bool use_mips = true); + void allocate(U32 res, U32 components, U32 count, bool use_mips = true, bool hdr = true); void bind(S32 stage); void unbind(); @@ -73,4 +74,5 @@ protected: U32 mWidth = 0; U32 mCount = 0; S32 mTextureStage; + bool mHDR; }; diff --git a/indra/llrender/llfontbitmapcache.cpp b/indra/llrender/llfontbitmapcache.cpp index 46c2e89797..6a3af1e608 100644 --- a/indra/llrender/llfontbitmapcache.cpp +++ b/indra/llrender/llfontbitmapcache.cpp @@ -107,7 +107,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp mBitmapHeight = image_height; S32 num_components = getNumComponents(bitmap_type); - mImageRawVec[bitmap_idx].push_back(new LLImageRaw(mBitmapWidth, mBitmapHeight, num_components)); + mImageRawVec[bitmap_idx].emplace_back(new LLImageRaw(mBitmapWidth, mBitmapHeight, num_components)); bitmap_num = static_cast<U32>(mImageRawVec[bitmap_idx].size()) - 1; LLImageRaw* image_raw = getImageRaw(bitmap_type, bitmap_num); @@ -117,7 +117,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp } // Make corresponding GL image. - mImageGLVec[bitmap_idx].push_back(new LLImageGL(image_raw, false)); + mImageGLVec[bitmap_idx].emplace_back(new LLImageGL(image_raw, false, false)); LLImageGL* image_gl = getImageGL(bitmap_type, bitmap_num); // Start at beginning of the new image. @@ -141,6 +141,7 @@ bool LLFontBitmapCache::nextOpenPos(S32 width, S32& pos_x, S32& pos_y, EFontGlyp bitmap_num = getNumBitmaps(bitmap_type) - 1; mCurrentOffsetX[bitmap_idx] += width + 1; + mGeneration++; return true; } @@ -168,6 +169,7 @@ void LLFontBitmapCache::reset() mBitmapWidth = 0; mBitmapHeight = 0; + mGeneration++; } //static diff --git a/indra/llrender/llfontbitmapcache.h b/indra/llrender/llfontbitmapcache.h index f2dfd87877..0ae4e6bed0 100644 --- a/indra/llrender/llfontbitmapcache.h +++ b/indra/llrender/llfontbitmapcache.h @@ -63,6 +63,7 @@ public: U32 getNumBitmaps(EFontGlyphType bitmapType) const { return (bitmapType < EFontGlyphType::Count) ? static_cast<U32>(mImageRawVec[static_cast<U32>(bitmapType)].size()) : 0U; } S32 getBitmapWidth() const { return mBitmapWidth; } S32 getBitmapHeight() const { return mBitmapHeight; } + S32 getCacheGeneration() const { return mGeneration; } protected: static U32 getNumComponents(EFontGlyphType bitmap_type); @@ -74,6 +75,7 @@ private: S32 mCurrentOffsetY[static_cast<U32>(EFontGlyphType::Count)] = { 1 }; S32 mMaxCharWidth = 0; S32 mMaxCharHeight = 0; + S32 mGeneration = 0; std::vector<LLPointer<LLImageRaw>> mImageRawVec[static_cast<U32>(EFontGlyphType::Count)]; std::vector<LLPointer<LLImageGL>> mImageGLVec[static_cast<U32>(EFontGlyphType::Count)]; }; diff --git a/indra/llrender/llfontfreetype.cpp b/indra/llrender/llfontfreetype.cpp index 9766de1dfa..e93c98970b 100644 --- a/indra/llrender/llfontfreetype.cpp +++ b/indra/llrender/llfontfreetype.cpp @@ -32,7 +32,7 @@ // Freetype stuff #include <ft2build.h> #ifdef LL_WINDOWS -#include <freetype2\freetype\ftsystem.h> +#include <freetype/ftsystem.h> #endif #include "llfontfreetypesvg.h" @@ -148,7 +148,6 @@ LLFontFreetype::LLFontFreetype() mIsFallback(false), mFTFace(NULL), mRenderGlyphCount(0), - mAddGlyphCount(0), mStyle(0), mPointSize(0) { @@ -180,7 +179,7 @@ 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) { @@ -554,7 +553,7 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l return NULL; llassert(!mIsFallback); - fontp->renderGlyph(requested_glyph_type, glyph_index); + fontp->renderGlyph(requested_glyph_type, glyph_index, wch); EFontGlyphType bitmap_glyph_type = EFontGlyphType::Unspecified; switch (fontp->mFTFace->glyph->bitmap.pixel_mode) @@ -576,7 +575,6 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l S32 pos_x, pos_y; U32 bitmap_num; mFontBitmapCachep->nextOpenPos(width, pos_x, pos_y, bitmap_glyph_type, bitmap_num); - mAddGlyphCount++; LLFontGlyphInfo* gi = new LLFontGlyphInfo(glyph_index, requested_glyph_type); gi->mXBitmapOffset = pos_x; @@ -658,7 +656,14 @@ LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, l LLImageGL *image_gl = mFontBitmapCachep->getImageGL(bitmap_glyph_type, bitmap_num); LLImageRaw *image_raw = mFontBitmapCachep->getImageRaw(bitmap_glyph_type, bitmap_num); - image_gl->setSubImage(image_raw, 0, 0, image_gl->getWidth(), image_gl->getHeight()); + if (image_gl && image_raw) + { + image_gl->setSubImage(image_raw, 0, 0, image_gl->getWidth(), image_gl->getHeight()); + } + else + { + llassert(false); //images were just inserted by nextOpenPos, they shouldn't be missing + } return gi; } @@ -699,7 +704,7 @@ void LLFontFreetype::insertGlyphInfo(llwchar wch, LLFontGlyphInfo* gi) const } } -void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) const +void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index, llwchar wch) const { if (mFTFace == NULL) return; @@ -714,11 +719,28 @@ void LLFontFreetype::renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) co FT_Error error = FT_Load_Glyph(mFTFace, glyph_index, load_flags); if (FT_Err_Ok != error) { + if (error == FT_Err_Out_Of_Memory) + { + LLError::LLUserWarningMsg::showOutOfMemory(); + LL_ERRS() << "Out of memory loading glyph for character " << llformat("U+%xu", U32(wch)) << LL_ENDL; + } + std::string message = llformat( - "Error %d (%s) loading glyph %u: bitmap_type=%u, load_flags=%d", - error, FT_Error_String(error), glyph_index, bitmap_type, load_flags); + "Error %d (%s) loading wchar %u glyph %u/%u: bitmap_type=%u, load_flags=%d", + error, FT_Error_String(error), wch, glyph_index, mFTFace->num_glyphs, bitmap_type, load_flags); LL_WARNS_ONCE() << message << LL_ENDL; error = FT_Load_Glyph(mFTFace, glyph_index, load_flags ^ FT_LOAD_COLOR); + if (FT_Err_Invalid_Outline == error + || FT_Err_Invalid_Composite == error + || (FT_Err_Ok != error && LLStringOps::isEmoji(wch))) + { + // value~0 always corresponds to the 'missing glyph' + error = FT_Load_Glyph(mFTFace, 0, FT_LOAD_FORCE_AUTOHINT); + if (FT_Err_Ok != error) + { + LL_ERRS() << "Loading fallback for char '" << (U32)wch << "', glyph " << glyph_index << " failed with error : " << (S32)error << LL_ENDL; + } + } llassert_always_msg(FT_Err_Ok == error, message.c_str()); } @@ -825,7 +847,12 @@ bool LLFontFreetype::setSubImageBGRA(U32 x, U32 y, U32 bitmap_num, U16 width, U1 { LLImageRaw* image_raw = mFontBitmapCachep->getImageRaw(EFontGlyphType::Color, bitmap_num); llassert(!mIsFallback); - llassert(image_raw && (image_raw->getComponents() == 4)); + if (!image_raw) + { + llassert(false); + return false; + } + llassert(image_raw->getComponents() == 4); // NOTE: inspired by LLImageRaw::setSubImage() U32* image_data = (U32*)image_raw->getData(); @@ -853,10 +880,17 @@ bool LLFontFreetype::setSubImageBGRA(U32 x, U32 y, U32 bitmap_num, U16 width, U1 void LLFontFreetype::setSubImageLuminanceAlpha(U32 x, U32 y, U32 bitmap_num, U32 width, U32 height, U8 *data, S32 stride) const { LLImageRaw *image_raw = mFontBitmapCachep->getImageRaw(EFontGlyphType::Grayscale, bitmap_num); - LLImageDataLock lock(image_raw); llassert(!mIsFallback); - llassert(image_raw && (image_raw->getComponents() == 2)); + if (!image_raw) + { + llassert(false); + return; + } + + LLImageDataLock lock(image_raw); + + llassert(image_raw->getComponents() == 2); U8 *target = image_raw->getData(); llassert(target); diff --git a/indra/llrender/llfontfreetype.h b/indra/llrender/llfontfreetype.h index eba89f5def..783bf4a4b3 100644 --- a/indra/llrender/llfontfreetype.h +++ b/indra/llrender/llfontfreetype.h @@ -156,7 +156,7 @@ private: bool hasGlyph(llwchar wch) const; // Has a glyph for this character LLFontGlyphInfo* addGlyph(llwchar wch, EFontGlyphType glyph_type) const; // Add a new character to the font if necessary LLFontGlyphInfo* addGlyphFromFont(const LLFontFreetype *fontp, llwchar wch, U32 glyph_index, EFontGlyphType bitmap_type) const; // Add a glyph from this font to the other (returns the glyph_index, 0 if not found) - void renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index) const; + void renderGlyph(EFontGlyphType bitmap_type, U32 glyph_index, llwchar wch) const; void insertGlyphInfo(llwchar wch, LLFontGlyphInfo* gi) const; std::string mName; @@ -187,7 +187,6 @@ private: mutable LLFontBitmapCache* mFontBitmapCachep; mutable S32 mRenderGlyphCount; - mutable S32 mAddGlyphCount; }; #endif // LL_FONTFREETYPE_H diff --git a/indra/llrender/llfontfreetypesvg.cpp b/indra/llrender/llfontfreetypesvg.cpp index 45cbc5dbd7..04825ae8a3 100644 --- a/indra/llrender/llfontfreetypesvg.cpp +++ b/indra/llrender/llfontfreetypesvg.cpp @@ -145,18 +145,18 @@ 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; @@ -165,13 +165,13 @@ 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 9482987970..16eec1fdd2 100644 --- a/indra/llrender/llfontgl.cpp +++ b/indra/llrender/llfontgl.cpp @@ -58,6 +58,7 @@ F32 LLFontGL::sVertDPI = 96.f; F32 LLFontGL::sHorizDPI = 96.f; F32 LLFontGL::sScaleX = 1.f; F32 LLFontGL::sScaleY = 1.f; +S32 LLFontGL::sResolutionGeneration = 0; bool LLFontGL::sDisplayFont = true ; std::string LLFontGL::sAppDir; @@ -109,10 +110,16 @@ S32 LLFontGL::getNumFaces(const std::string& filename) return mFontFreetype->getNumFaces(filename); } +S32 LLFontGL::getCacheGeneration() const +{ + const LLFontBitmapCache* font_bitmap_cache = mFontFreetype->getFontBitmapCache(); + return font_bitmap_cache->getCacheGeneration(); +} + S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, const LLRect& rect, const LLColor4 &color, HAlign halign, VAlign valign, U8 style, ShadowType shadow, S32 max_chars, F32* right_x, bool use_ellipses, bool use_color) const { - 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 +145,7 @@ 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); } @@ -249,6 +256,10 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons const LLFontBitmapCache* font_bitmap_cache = mFontFreetype->getFontBitmapCache(); + // This looks wrong, value is dynamic. + // LLFontBitmapCache::nextOpenPos can alter these values when + // new characters get added to cache, which affects whole string. + // Todo: Perhaps value should update after symbols were added? F32 inv_width = 1.f / font_bitmap_cache->getBitmapWidth(); F32 inv_height = 1.f / font_bitmap_cache->getBitmapHeight(); @@ -270,10 +281,14 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons const LLFontGlyphInfo* next_glyph = NULL; - const S32 GLYPH_BATCH_SIZE = 120; - LLVector4a vertices[GLYPH_BATCH_SIZE * 6]; - LLVector2 uvs[GLYPH_BATCH_SIZE * 6]; - LLColor4U colors[GLYPH_BATCH_SIZE * 6]; + // string can have more than one glyph per char (ex: bold or shadow), + // make sure that GLYPH_BATCH_SIZE won't end up with half a symbol. + // See drawGlyph. + // Ex: with shadows it's 6 glyps per char. 30 fits exactly 5 chars. + static constexpr S32 GLYPH_BATCH_SIZE = 30; + static thread_local LLVector4a vertices[GLYPH_BATCH_SIZE * 6]; + static thread_local LLVector2 uvs[GLYPH_BATCH_SIZE * 6]; + static thread_local LLColor4U colors[GLYPH_BATCH_SIZE * 6]; LLColor4U text_color(color); // Preserve the transparency to render fading emojis in fading text (e.g. @@ -282,6 +297,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons std::pair<EFontGlyphType, S32> bitmap_entry = std::make_pair(EFontGlyphType::Grayscale, -1); S32 glyph_count = 0; + llwchar last_char = wstr[begin_offset]; for (i = begin_offset; i < begin_offset + length; i++) { llwchar wch = wstr[i]; @@ -299,7 +315,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons } // Per-glyph bitmap texture. std::pair<EFontGlyphType, S32> next_bitmap_entry = fgi->mBitmapEntry; - if (next_bitmap_entry != bitmap_entry) + if (next_bitmap_entry != bitmap_entry || last_char != wch) { // Actually draw the queued glyphs before switching their texture; // otherwise the queued glyphs will be taken from wrong textures. @@ -316,6 +332,11 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons bitmap_entry = next_bitmap_entry; LLImageGL* font_image = font_bitmap_cache->getImageGL(bitmap_entry.first, bitmap_entry.second); gGL.getTexUnit(0)->bind(font_image); + + // For some reason it's not enough to compare by bitmap_entry. + // Issue hits emojis, japenese and chinese glyphs, only on first run. + // Todo: figure it out, there might be a bug with raw image data. + last_char = wch; } if ((start_x + scaled_max_pixels) < (cur_x + fgi->mXBearing + fgi->mWidth)) @@ -343,6 +364,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons gGL.vertexBatchPreTransformed(vertices, uvs, colors, glyph_count * 6); } gGL.end(); + glyph_count = 0; } @@ -401,11 +423,10 @@ 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, @@ -416,7 +437,6 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, cons right_x, false, use_color); - gGL.popUIMatrix(); } gGL.popUIMatrix(); @@ -502,6 +522,7 @@ 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; @@ -559,7 +580,7 @@ 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); @@ -569,7 +590,7 @@ 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; @@ -880,7 +901,7 @@ 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()); @@ -893,7 +914,7 @@ 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)); @@ -1229,34 +1250,34 @@ void LLFontGL::renderTriangle(LLVector4a* vertex_out, LLVector2* uv_out, LLColor { S32 index = 0; - vertex_out[index] = LLVector4a(screen_rect.mRight, screen_rect.mTop, 0.f); - uv_out[index] = LLVector2(uv_rect.mRight, uv_rect.mTop); + vertex_out[index].set(screen_rect.mRight, screen_rect.mTop, 0.f); + uv_out[index].set(uv_rect.mRight, uv_rect.mTop); colors_out[index] = color; index++; - vertex_out[index] = LLVector4a(screen_rect.mLeft, screen_rect.mTop, 0.f); - uv_out[index] = LLVector2(uv_rect.mLeft, uv_rect.mTop); + vertex_out[index].set(screen_rect.mLeft, screen_rect.mTop, 0.f); + uv_out[index].set(uv_rect.mLeft, uv_rect.mTop); colors_out[index] = color; index++; - vertex_out[index] = LLVector4a(screen_rect.mLeft, screen_rect.mBottom, 0.f); - uv_out[index] = LLVector2(uv_rect.mLeft, uv_rect.mBottom); + vertex_out[index].set(screen_rect.mLeft, screen_rect.mBottom, 0.f); + uv_out[index].set(uv_rect.mLeft, uv_rect.mBottom); colors_out[index] = color; index++; - vertex_out[index] = LLVector4a(screen_rect.mRight, screen_rect.mTop, 0.f); - uv_out[index] = LLVector2(uv_rect.mRight, uv_rect.mTop); + vertex_out[index].set(screen_rect.mRight, screen_rect.mTop, 0.f); + uv_out[index].set(uv_rect.mRight, uv_rect.mTop); colors_out[index] = color; index++; - vertex_out[index] = LLVector4a(screen_rect.mLeft, screen_rect.mBottom, 0.f); - uv_out[index] = LLVector2(uv_rect.mLeft, uv_rect.mBottom); + vertex_out[index].set(screen_rect.mLeft, screen_rect.mBottom, 0.f); + uv_out[index].set(uv_rect.mLeft, uv_rect.mBottom); colors_out[index] = color; index++; - vertex_out[index] = LLVector4a(screen_rect.mRight, screen_rect.mBottom, 0.f); - uv_out[index] = LLVector2(uv_rect.mRight, uv_rect.mBottom); + vertex_out[index].set(screen_rect.mRight, screen_rect.mBottom, 0.f); + uv_out[index].set(uv_rect.mRight, uv_rect.mBottom); colors_out[index] = color; } diff --git a/indra/llrender/llfontgl.h b/indra/llrender/llfontgl.h index 4bb6c55c65..1c8e036f58 100644 --- a/indra/llrender/llfontgl.h +++ b/indra/llrender/llfontgl.h @@ -90,6 +90,7 @@ public: bool loadFace(const std::string& filename, F32 point_size, const F32 vert_dpi, const F32 horz_dpi, bool is_fallback, S32 face_n); S32 getNumFaces(const std::string& filename); + S32 getCacheGeneration() const; S32 render(const LLWString &text, S32 begin_offset, const LLRect& rect, @@ -224,6 +225,7 @@ public: static F32 sHorizDPI; static F32 sScaleX; static F32 sScaleY; + static S32 sResolutionGeneration; static bool sDisplayFont ; static std::string sAppDir; // For loading fonts diff --git a/indra/llrender/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..a223509d30 --- /dev/null +++ b/indra/llrender/llfontvertexbuffer.cpp @@ -0,0 +1,239 @@ +/** + * @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" + + +bool LLFontVertexBuffer::sEnableBufferCollection = true; + +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 (!sEnableBufferCollection) + { + // For debug purposes and performance testing + return fontp->render(text, begin_offset, x, y, color, halign, valign, style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color); + } + 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 + || mLastVertDPI != LLFontGL::sVertDPI + || mLastHorizDPI != LLFontGL::sHorizDPI + || mLastOrigin != LLFontGL::sCurOrigin + || mLastResGeneration != LLFontGL::sResolutionGeneration + || mLastFontCacheGen != fontp->getCacheGeneration()) + { + genBuffers(fontp, text, begin_offset, x, y, color, halign, valign, + style, shadow, max_chars, max_pixels, right_x, use_ellipses, use_color); + } + 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(); + // Save before rendreing, it can change mid-render, + // so will need to rerender previous characters + mLastFontCacheGen = fontp->getCacheGeneration(); + + gGL.beginList(&mBufferList); + mChars = fontp->render(text, begin_offset, x, y, color, halign, valign, + 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; + mLastVertDPI = LLFontGL::sVertDPI; + mLastHorizDPI = LLFontGL::sHorizDPI; + mLastOrigin = LLFontGL::sCurOrigin; + mLastResGeneration = LLFontGL::sResolutionGeneration; + + if (right_x) + { + mLastRightX = *right_x; + } +} + +void LLFontVertexBuffer::renderBuffers() +{ + gGL.flush(); // deliberately empty pending verts + gGL.getTexUnit(0)->enable(LLTexUnit::TT_TEXTURE); + gGL.pushUIMatrix(); + + gGL.loadUIIdentity(); + + // Depth translation, so that floating text appears 'in-world' + // and is correctly occluded. + gGL.translatef(0.f, 0.f, LLFontGL::sCurDepth); + gGL.setSceneBlendType(LLRender::BT_ALPHA); + + // Note: ellipses should technically be covered by push/load/translate of their own + // but it's more complexity, values do not change, skipping doesn't appear to break + // anything, so we can skip that until it proves to cause issues. + 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..a9e1e2337c --- /dev/null +++ b/indra/llrender/llfontvertexbuffer.h @@ -0,0 +1,130 @@ +/** + * @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); + + static void enableBufferCollection(bool enable) { sEnableBufferCollection = enable; } +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; + F32 mLastVertDPI = 0.f; + F32 mLastHorizDPI = 0.f; + S32 mLastResGeneration = 0; + LLCoordGL mLastOrigin; + + // Adding new characters to bitmap cache can alter value from getBitmapWidth(); + // which alters whole string. So rerender when new characters were added to cache. + S32 mLastFontCacheGen = 0; + + static bool sEnableBufferCollection; +}; + +#endif diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp index 2a7cc3d141..ac66faaf5a 100644 --- a/indra/llrender/llgl.cpp +++ b/indra/llrender/llgl.cpp @@ -50,10 +50,9 @@ #include "llglheaders.h" #include "llglslshader.h" -#include "llvertexbuffer.h" -#include "llcontrol.h" -extern LLControlGroup gSavedSettings; - +#include "glm/glm.hpp" +#include <glm/gtc/matrix_access.hpp> +#include "glm/gtc/type_ptr.hpp" #if LL_WINDOWS #include "lldxhardware.h" @@ -239,8 +238,6 @@ PFNWGLBLITCONTEXTFRAMEBUFFERAMDPROC wglBlitContextFramebufferAMD = n PFNWGLSWAPINTERVALEXTPROC wglSwapIntervalEXT = nullptr; PFNWGLGETSWAPINTERVALEXTPROC wglGetSwapIntervalEXT = nullptr; -/* - // GL_VERSION_1_2 //PFNGLDRAWRANGEELEMENTSPROC glDrawRangeElements = nullptr; //PFNGLTEXIMAGE3DPROC glTexImage3D = nullptr; @@ -985,7 +982,6 @@ PFNGLMULTIDRAWARRAYSINDIRECTCOUNTPROC glMultiDrawArraysIndirectCount = nullpt PFNGLMULTIDRAWELEMENTSINDIRECTCOUNTPROC glMultiDrawElementsIndirectCount = nullptr; PFNGLPOLYGONOFFSETCLAMPPROC glPolygonOffsetClamp = nullptr; -*/ #endif LLGLManager gGLManager; @@ -1002,7 +998,6 @@ LLGLManager::LLGLManager() : mIsAMD(false), mIsNVIDIA(false), mIsIntel(false), - mIsApple(false), #if LL_DARWIN mIsMobileGF(false), #endif @@ -1047,6 +1042,7 @@ 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)) { @@ -1176,10 +1172,10 @@ bool LLGLManager::initGL() mGLVendorShort = "INTEL"; mIsIntel = true; } - else if(mGLVendor.find("APPLE") != std::string::npos) + else if (mGLVendor.find("APPLE") != std::string::npos) { mGLVendorShort = "APPLE"; - mIsApple = TRUE; + mIsApple = true; } else { @@ -1221,21 +1217,15 @@ bool LLGLManager::initGL() LL_WARNS("RenderInit") << "VRAM Detected (AMDAssociations):" << mVRAM << LL_ENDL; } } -#endif + else if (mHasNVXGpuMemoryInfo) + { + GLint mem_kb = 0; + glGetIntegerv(GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX, &mem_kb); + mVRAM = mem_kb / 1024; -#if LL_WINDOWS - if (mVRAM < 256) - { - // Something likely went wrong using the above extensions - // try WMI first and fall back to old method (from dxdiag) if all else fails - // Function will check all GPUs WMI knows of and will pick up the one with most - // memory. We need to check all GPUs because system can switch active GPU to - // weaker one, to preserve power when not under load. - U32 mem = LLDXHardware::getMBVideoMemoryViaWMI(); - if (mem != 0) + if (mVRAM != 0) { - mVRAM = mem; - LL_WARNS("RenderInit") << "VRAM Detected (WMI):" << mVRAM<< LL_ENDL; + LL_WARNS("RenderInit") << "VRAM Detected (NVXGpuMemoryInfo):" << mVRAM << LL_ENDL; } } #endif @@ -1243,8 +1233,6 @@ bool LLGLManager::initGL() if (mVRAM < 256 && old_vram > 0) { // fall back to old method - // Note: on Windows value will be from LLDXHardware. - // Either received via dxdiag or via WMI by id from dxdiag. mVRAM = old_vram; } @@ -1256,31 +1244,20 @@ bool LLGLManager::initGL() glGetIntegerv(GL_MAX_SAMPLE_MASK_WORDS, &mMaxSampleMaskWords); #endif glGetIntegerv(GL_MAX_SAMPLES, &mMaxSamples); + glGetIntegerv(GL_MAX_VARYING_VECTORS, &mMaxVaryingVectors); glGetIntegerv(GL_MAX_UNIFORM_BLOCK_SIZE, &mMaxUniformBlockSize); // sanity clamp max uniform block size to 64k just in case // there's some implementation that reports a crazy value mMaxUniformBlockSize = llmin(mMaxUniformBlockSize, 65536); - if (mGLVersion >= 4.59f) + if (mHasAnisotropic) { glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &mMaxAnisotropy); } initGLStates(); - U32 MPVBufferOptiMode = gSavedSettings.getU32("MPVBufferOptiMode"); - if (MPVBufferOptiMode == 0) - { - if(mIsApple) MPVBufferOptiMode = 2; - else MPVBufferOptiMode = 1; - gSavedSettings.setU32("MPVBufferOptiMode",MPVBufferOptiMode); - } - LLVertexBuffer::sMappingMode = MPVBufferOptiMode; - //LLRender::sMappingMode = MPVBufferOptiMode; - - LL_INFOS() << "milo init sMappingMode " << MPVBufferOptiMode << LL_ENDL; - return true; } @@ -1397,7 +1374,6 @@ void LLGLManager::asLLSD(LLSD& info) info["is_ati"] = mIsAMD; // note, do not rename is_ati to is_amd without coordinating with DW info["is_nvidia"] = mIsNVIDIA; info["is_intel"] = mIsIntel; - info["is_apple"] = mIsApple; info["gl_renderer"] = mGLRenderer; } @@ -1442,6 +1418,11 @@ void LLGLManager::initExtensions() mHasCubeMapArray = mGLVersion >= 3.99f; mHasTransformFeedback = mGLVersion >= 3.99f; mHasDebugOutput = mGLVersion >= 4.29f; + mHasAnisotropic = mGLVersion >= 4.59f; + if(!mHasAnisotropic && gGLHExts.mSysExts) + { + mHasAnisotropic = ExtensionExists("GL_EXT_texture_filter_anisotropic", gGLHExts.mSysExts); + } // Misc glGetIntegerv(GL_MAX_ELEMENTS_VERTICES, (GLint*) &mGLMaxVertexRange); @@ -1450,7 +1431,6 @@ void LLGLManager::initExtensions() mInited = true; -/* #if LL_WINDOWS LL_DEBUGS("RenderInit") << "GL Probe: Getting symbols" << LL_ENDL; @@ -2288,7 +2268,6 @@ void LLGLManager::initExtensions() glPolygonOffsetClamp = (PFNGLPOLYGONOFFSETCLAMPPROC)GLH_EXT_GET_PROC_ADDRESS("glPolygonOffsetClamp"); #endif -*/ } void rotate_quat(LLQuaternion& rotation) @@ -2477,12 +2456,15 @@ void LLGLState::checkStates(GLboolean writeAlpha) return; } - GLint src; - GLint dst; - glGetIntegerv(GL_BLEND_SRC, &src); - glGetIntegerv(GL_BLEND_DST, &dst); - llassert_always(src == GL_SRC_ALPHA); - llassert_always(dst == GL_ONE_MINUS_SRC_ALPHA); + GLint srcRGB, dstRGB, srcAlpha, dstAlpha; + glGetIntegerv(GL_BLEND_SRC_RGB, &srcRGB); + glGetIntegerv(GL_BLEND_DST_RGB, &dstRGB); + glGetIntegerv(GL_BLEND_SRC_ALPHA, &srcAlpha); + glGetIntegerv(GL_BLEND_DST_ALPHA, &dstAlpha); + llassert_always(srcRGB == GL_SRC_ALPHA); + llassert_always(srcAlpha == GL_SRC_ALPHA); + llassert_always(dstRGB == GL_ONE_MINUS_SRC_ALPHA); + llassert_always(dstAlpha == GL_ONE_MINUS_SRC_ALPHA); // disable for now until usage is consistent //GLboolean colorMask[4]; @@ -2724,7 +2706,7 @@ void parse_glsl_version(S32& major, S32& minor) LLStringUtil::convertToS32(minor_str, minor); } -LLGLUserClipPlane::LLGLUserClipPlane(const LLPlane& p, const glh::matrix4f& modelview, const glh::matrix4f& projection, bool apply) +LLGLUserClipPlane::LLGLUserClipPlane(const LLPlane& p, const glm::mat4& modelview, const glm::mat4& projection, bool apply) { mApply = apply; @@ -2751,13 +2733,12 @@ void LLGLUserClipPlane::disable() void LLGLUserClipPlane::setPlane(F32 a, F32 b, F32 c, F32 d) { - glh::matrix4f& P = mProjection; - glh::matrix4f& M = mModelview; + const glm::mat4& P = mProjection; + const glm::mat4& M = mModelview; - glh::matrix4f invtrans_MVP = (P * M).inverse().transpose(); - glh::vec4f oplane(a,b,c,d); - glh::vec4f cplane; - invtrans_MVP.mult_matrix_vec(oplane, cplane); + glm::mat4 invtrans_MVP = glm::transpose(glm::inverse(P*M)); + glm::vec4 oplane(a,b,c,d); + glm::vec4 cplane = invtrans_MVP * oplane; cplane /= fabs(cplane[2]); // normalize such that depth is not scaled cplane[3] -= 1; @@ -2765,13 +2746,13 @@ void LLGLUserClipPlane::setPlane(F32 a, F32 b, F32 c, F32 d) if(cplane[2] < 0) cplane *= -1; - glh::matrix4f suffix; - suffix.set_row(2, cplane); - glh::matrix4f newP = suffix * P; + glm::mat4 suffix; + suffix = glm::row(suffix, 2, cplane); + glm::mat4 newP = suffix * P; gGL.matrixMode(LLRender::MM_PROJECTION); gGL.pushMatrix(); - gGL.loadMatrix(newP.m); - gGLObliqueProjectionInverse = LLMatrix4(newP.inverse().transpose().m); + gGL.loadMatrix(glm::value_ptr(newP)); + gGLObliqueProjectionInverse = LLMatrix4(glm::value_ptr(glm::transpose(glm::inverse(newP)))); gGL.matrixMode(LLRender::MM_MODELVIEW); } @@ -2784,7 +2765,7 @@ 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) @@ -2817,6 +2798,7 @@ LLGLDepthTest::LLGLDepthTest(GLboolean depth_enabled, GLboolean write_enabled, G LLGLDepthTest::~LLGLDepthTest() { + LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE; checkState(); if (sDepthEnabled != mPrevDepthEnabled ) { @@ -2867,31 +2849,27 @@ void LLGLDepthTest::checkState() LLGLSquashToFarClip::LLGLSquashToFarClip() { - glh::matrix4f proj = get_current_projection(); + glm::mat4 proj = get_current_projection(); setProjectionMatrix(proj, 0); } -LLGLSquashToFarClip::LLGLSquashToFarClip(glh::matrix4f& P, U32 layer) +LLGLSquashToFarClip::LLGLSquashToFarClip(const glm::mat4& P, U32 layer) { setProjectionMatrix(P, layer); } - -void LLGLSquashToFarClip::setProjectionMatrix(glh::matrix4f& projection, U32 layer) +void LLGLSquashToFarClip::setProjectionMatrix(glm::mat4 projection, U32 layer) { - F32 depth = 0.99999f - 0.0001f * layer; - for (U32 i = 0; i < 4; i++) - { - projection.element(2, i) = projection.element(3, i) * depth; - } + glm::vec4 P_row_3 = glm::row(projection, 3) * depth; + projection = glm::row(projection, 2, P_row_3); LLRender::eMatrixMode last_matrix_mode = gGL.getMatrixMode(); gGL.matrixMode(LLRender::MM_PROJECTION); gGL.pushMatrix(); - gGL.loadMatrix(projection.m); + gGL.loadMatrix(glm::value_ptr(projection)); gGL.matrixMode(last_matrix_mode); } diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h index a3d87e001c..bf8368a7b3 100644 --- a/indra/llrender/llgl.h +++ b/indra/llrender/llgl.h @@ -43,7 +43,7 @@ #include "llinstancetracker.h" #include "llglheaders.h" -#include "glh/glh_linear.h" +#include "glm/mat4x4.hpp" extern bool gDebugGL; extern bool gDebugSession; @@ -88,6 +88,7 @@ public: S32 mGLMaxTextureSize; F32 mMaxAnisotropy = 0.f; S32 mMaxUniformBlockSize = 0; + S32 mMaxVaryingVectors = 0; // GL 4.x capabilities bool mHasCubeMapArray = false; @@ -97,11 +98,15 @@ public: // Vendor-specific extensions bool mHasAMDAssociations = false; + bool mHasNVXGpuMemoryInfo = false; bool mIsAMD; bool mIsNVIDIA; bool mIsIntel; - bool mIsApple; + bool mIsApple = false; + + // 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 @@ -155,18 +160,17 @@ void assert_glerror(); void clear_glerror(); +#if !LL_RELEASE_FOR_DOWNLOAD # define stop_glerror() assert_glerror() # define llglassertok() assert_glerror() - -// stop_glerror is still needed on OS X but has performance implications -// use macro below to conditionally add stop_glerror to non-release builds -// on OS X -#if LL_DARWIN && !LL_RELEASE_FOR_DOWNLOAD -#define STOP_GLERROR stop_glerror() +# define STOP_GLERROR stop_glerror() #else -#define STOP_GLERROR +# define stop_glerror() +# define llglassertok() +# define STOP_GLERROR #endif + #define llglassertok_always() assert_glerror() //////////////////////// @@ -314,7 +318,7 @@ class LLGLUserClipPlane { public: - LLGLUserClipPlane(const LLPlane& plane, const glh::matrix4f& modelview, const glh::matrix4f& projection, bool apply = true); + LLGLUserClipPlane(const LLPlane& plane, const glm::mat4& modelview, const glm::mat4& projection, bool apply = true); ~LLGLUserClipPlane(); void setPlane(F32 a, F32 b, F32 c, F32 d); @@ -323,8 +327,8 @@ public: private: bool mApply; - glh::matrix4f mProjection; - glh::matrix4f mModelview; + glm::mat4 mProjection; + glm::mat4 mModelview; }; /* @@ -338,9 +342,9 @@ class LLGLSquashToFarClip { public: LLGLSquashToFarClip(); - LLGLSquashToFarClip(glh::matrix4f& projection, U32 layer = 0); + LLGLSquashToFarClip(const glm::mat4& projection, U32 layer = 0); - void setProjectionMatrix(glh::matrix4f& projection, U32 layer); + void setProjectionMatrix(glm::mat4 projection, U32 layer); ~LLGLSquashToFarClip(); }; diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h index 7d56f12766..a2685376cd 100644 --- a/indra/llrender/llglheaders.h +++ b/indra/llrender/llglheaders.h @@ -62,7 +62,7 @@ // LL_WINDOWS // windows gl headers depend on things like APIENTRY, so include windows. -#include "llwin32headerslean.h" +#include "llwin32headers.h" //---------------------------------------------------------------------------- #include <GL/gl.h> diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 880b491253..aac5e3abc8 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -63,6 +63,7 @@ U64 LLGLSLShader::sTotalTimeElapsed = 0; U32 LLGLSLShader::sTotalTrianglesDrawn = 0; U64 LLGLSLShader::sTotalSamplesDrawn = 0; U32 LLGLSLShader::sTotalBinds = 0; +boost::json::value LLGLSLShader::sDefaultStats; //UI shader -- declared here so llui_libtest will link properly LLGLSLShader gUIProgram; @@ -101,9 +102,9 @@ void LLGLSLShader::initProfile() sTotalSamplesDrawn = 0; sTotalBinds = 0; - for (std::set<LLGLSLShader*>::iterator iter = sInstances.begin(); iter != sInstances.end(); ++iter) + for (auto ptr : sInstances) { - (*iter)->clearStats(); + ptr->clearStats(); } } @@ -117,45 +118,57 @@ struct LLGLSLShaderCompareTimeElapsed }; //static -void LLGLSLShader::finishProfile(bool emit_report) +void LLGLSLShader::finishProfile(boost::json::value& statsv) { sProfileEnabled = false; - if (emit_report) + if (! statsv.is_null()) { - std::vector<LLGLSLShader*> sorted; - - for (std::set<LLGLSLShader*>::iterator iter = sInstances.begin(); iter != sInstances.end(); ++iter) - { - sorted.push_back(*iter); - } - + std::vector<LLGLSLShader*> sorted(sInstances.begin(), sInstances.end()); std::sort(sorted.begin(), sorted.end(), LLGLSLShaderCompareTimeElapsed()); + auto& stats = statsv.as_object(); + auto shadersit = stats.emplace("shaders", boost::json::array_kind).first; + auto& shaders = shadersit->value().as_array(); bool unbound = false; - for (std::vector<LLGLSLShader*>::iterator iter = sorted.begin(); iter != sorted.end(); ++iter) + for (auto ptr : sorted) { - (*iter)->dumpStats(); - if ((*iter)->mBinds == 0) + if (ptr->mBinds == 0) { unbound = true; } + else + { + auto& shaderit = shaders.emplace_back(boost::json::object_kind); + ptr->dumpStats(shaderit.as_object()); + } } + constexpr float mega = 1'000'000.f; + float totalTimeMs = sTotalTimeElapsed / mega; LL_INFOS() << "-----------------------------------" << LL_ENDL; - LL_INFOS() << "Total rendering time: " << llformat("%.4f ms", sTotalTimeElapsed / 1000000.f) << LL_ENDL; - LL_INFOS() << "Total samples drawn: " << llformat("%.4f million", sTotalSamplesDrawn / 1000000.f) << LL_ENDL; - LL_INFOS() << "Total triangles drawn: " << llformat("%.3f million", sTotalTrianglesDrawn / 1000000.f) << LL_ENDL; + LL_INFOS() << "Total rendering time: " << llformat("%.4f ms", totalTimeMs) << LL_ENDL; + LL_INFOS() << "Total samples drawn: " << llformat("%.4f million", sTotalSamplesDrawn / mega) << LL_ENDL; + LL_INFOS() << "Total triangles drawn: " << llformat("%.3f million", sTotalTrianglesDrawn / mega) << LL_ENDL; LL_INFOS() << "-----------------------------------" << LL_ENDL; - + auto totalsit = stats.emplace("totals", boost::json::object_kind).first; + auto& totals = totalsit->value().as_object(); + totals.emplace("time", totalTimeMs / 1000.0); + totals.emplace("binds", sTotalBinds); + totals.emplace("samples", sTotalSamplesDrawn); + totals.emplace("triangles", sTotalTrianglesDrawn); + + auto unusedit = stats.emplace("unused", boost::json::array_kind).first; + auto& unused = unusedit->value().as_array(); if (unbound) { LL_INFOS() << "The following shaders were unused: " << LL_ENDL; - for (std::vector<LLGLSLShader*>::iterator iter = sorted.begin(); iter != sorted.end(); ++iter) + for (auto ptr : sorted) { - if ((*iter)->mBinds == 0) + if (ptr->mBinds == 0) { - LL_INFOS() << (*iter)->mName << LL_ENDL; + LL_INFOS() << ptr->mName << LL_ENDL; + unused.emplace_back(ptr->mName); } } } @@ -170,36 +183,43 @@ void LLGLSLShader::clearStats() mBinds = 0; } -void LLGLSLShader::dumpStats() +void LLGLSLShader::dumpStats(boost::json::object& stats) { - if (mBinds > 0) + stats.emplace("name", mName); + auto filesit = stats.emplace("files", boost::json::array_kind).first; + auto& files = filesit->value().as_array(); + LL_INFOS() << "=============================================" << LL_ENDL; + LL_INFOS() << mName << LL_ENDL; + for (U32 i = 0; i < mShaderFiles.size(); ++i) { - LL_INFOS() << "=============================================" << LL_ENDL; - LL_INFOS() << mName << LL_ENDL; - for (U32 i = 0; i < mShaderFiles.size(); ++i) - { - LL_INFOS() << mShaderFiles[i].first << LL_ENDL; - } - LL_INFOS() << "=============================================" << LL_ENDL; + LL_INFOS() << mShaderFiles[i].first << LL_ENDL; + files.emplace_back(mShaderFiles[i].first); + } + LL_INFOS() << "=============================================" << LL_ENDL; - F32 ms = mTimeElapsed / 1000000.f; - F32 seconds = ms / 1000.f; + constexpr float mega = 1'000'000.f; + constexpr double giga = 1'000'000'000.0; + F32 ms = mTimeElapsed / mega; + F32 seconds = ms / 1000.f; - F32 pct_tris = (F32)mTrianglesDrawn / (F32)sTotalTrianglesDrawn * 100.f; - F32 tris_sec = (F32)(mTrianglesDrawn / 1000000.0); - tris_sec /= seconds; + F32 pct_tris = (F32)mTrianglesDrawn / (F32)sTotalTrianglesDrawn * 100.f; + F32 tris_sec = (F32)(mTrianglesDrawn / mega); + tris_sec /= seconds; - F32 pct_samples = (F32)((F64)mSamplesDrawn / (F64)sTotalSamplesDrawn) * 100.f; - F32 samples_sec = (F32)mSamplesDrawn / 1000000000.0; - samples_sec /= seconds; + F32 pct_samples = (F32)((F64)mSamplesDrawn / (F64)sTotalSamplesDrawn) * 100.f; + F32 samples_sec = (F32)(mSamplesDrawn / giga); + samples_sec /= seconds; - F32 pct_binds = (F32)mBinds / (F32)sTotalBinds * 100.f; + F32 pct_binds = (F32)mBinds / (F32)sTotalBinds * 100.f; - LL_INFOS() << "Triangles Drawn: " << mTrianglesDrawn << " " << llformat("(%.2f pct of total, %.3f million/sec)", pct_tris, tris_sec) << LL_ENDL; - LL_INFOS() << "Binds: " << mBinds << " " << llformat("(%.2f pct of total)", pct_binds) << LL_ENDL; - LL_INFOS() << "SamplesDrawn: " << mSamplesDrawn << " " << llformat("(%.2f pct of total, %.3f billion/sec)", pct_samples, samples_sec) << LL_ENDL; - LL_INFOS() << "Time Elapsed: " << mTimeElapsed << " " << llformat("(%.2f pct of total, %.5f ms)\n", (F32)((F64)mTimeElapsed / (F64)sTotalTimeElapsed) * 100.f, ms) << LL_ENDL; - } + LL_INFOS() << "Triangles Drawn: " << mTrianglesDrawn << " " << llformat("(%.2f pct of total, %.3f million/sec)", pct_tris, tris_sec) << LL_ENDL; + LL_INFOS() << "Binds: " << mBinds << " " << llformat("(%.2f pct of total)", pct_binds) << LL_ENDL; + LL_INFOS() << "SamplesDrawn: " << mSamplesDrawn << " " << llformat("(%.2f pct of total, %.3f billion/sec)", pct_samples, samples_sec) << LL_ENDL; + LL_INFOS() << "Time Elapsed: " << mTimeElapsed << " " << llformat("(%.2f pct of total, %.5f ms)\n", (F32)((F64)mTimeElapsed / (F64)sTotalTimeElapsed) * 100.f, ms) << LL_ENDL; + stats.emplace("time", seconds); + stats.emplace("binds", mBinds); + stats.emplace("samples", mSamplesDrawn); + stats.emplace("triangles", mTrianglesDrawn); } //static @@ -307,16 +327,15 @@ bool LLGLSLShader::readProfileQuery(bool for_runtime, bool force_read) GLuint64 primitives_generated = 0; glGetQueryObjectui64v(mPrimitivesQuery, GL_QUERY_RESULT, &primitives_generated); #else - GLuint samples_passed = 0; - glGetQueryObjectuiv(mSamplesQuery, GL_QUERY_RESULT, &samples_passed); - GLuint primitives_generated = 0; glGetQueryObjectuiv(mPrimitivesQuery, GL_QUERY_RESULT, &primitives_generated); #endif sTotalTimeElapsed += time_elapsed; +#if GL_VERSION_1_5 sTotalSamplesDrawn += samples_passed; mSamplesDrawn += samples_passed; +#endif U32 tri_count = (U32)primitives_generated / 3; @@ -1065,7 +1084,7 @@ void LLGLSLShader::bind() { LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER; - llassert(mProgramObject != 0); + llassert_always(mProgramObject != 0); gGL.flush(); @@ -1088,12 +1107,15 @@ void LLGLSLShader::bind() LLShaderMgr::instance()->updateShaderUniforms(this); mUniformsDirty = false; } + + llassert_always(sCurBoundShaderPtr != nullptr); + llassert_always(sCurBoundShader == mProgramObject); } void LLGLSLShader::bind(U8 variant) { - llassert(mGLTFVariants.size() == LLGLSLShader::NUM_GLTF_VARIANTS); - llassert(variant < LLGLSLShader::NUM_GLTF_VARIANTS); + llassert_always(mGLTFVariants.size() == LLGLSLShader::NUM_GLTF_VARIANTS); + llassert_always(variant < LLGLSLShader::NUM_GLTF_VARIANTS); mGLTFVariants[variant].bind(); } @@ -1101,7 +1123,7 @@ void LLGLSLShader::bind(bool rigged) { if (rigged) { - llassert(mRiggedVariant); + llassert_always(mRiggedVariant); mRiggedVariant->bind(); } else @@ -1126,17 +1148,17 @@ 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; @@ -1152,7 +1174,6 @@ S32 LLGLSLShader::bindTexture(S32 uniform, LLTexture* texture, LLTexUnit::eTextu if (uniform > -1) { gGL.getTexUnit(uniform)->bindFast(texture); - gGL.getTexUnit(uniform)->setTextureColorSpace(colorspace); } return uniform; @@ -1233,7 +1254,7 @@ 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; @@ -1250,12 +1271,11 @@ 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; @@ -1265,23 +1285,40 @@ S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTe llassert(false); return -1; } + S32 index = mTexture[uniform]; - if (index != -1 && gGL.getTexUnit(index)->getCurrType() != LLTexUnit::TT_NONE) + if (index < 0) { - if (gDebugGL && gGL.getTexUnit(index)->getCurrType() != mode && gGL.getTexUnit(index)->getCurrColorSpace() != space) + // Invalid texture index - nothing to disable + return index; + } + + LLTexUnit* tex_unit = gGL.getTexUnit(index); + if (!tex_unit) + { + // Invalid texture unit + LL_WARNS_ONCE("Shader") << "Invalid texture unit at index: " << index << LL_ENDL; + return index; + } + + LLTexUnit::eTextureType curr_type = tex_unit->getCurrType(); + if (curr_type != LLTexUnit::TT_NONE) + { + if (gDebugGL && curr_type != mode) { if (gDebugSession) { - gFailLog << "Texture channel " << index << " texture type corrupted." << std::endl; + gFailLog << "Texture channel " << index << " texture type corrupted. Expected: " << mode << ", Found: " << curr_type << std::endl; ll_fail("LLGLSLShader::disableTexture failed"); } else { - LL_ERRS() << "Texture channel " << index << " texture type corrupted." << LL_ENDL; + LL_ERRS() << "Texture channel " << index << " texture type corrupted. Expected: " << mode << ", Found: " << curr_type << LL_ENDL; } } - gGL.getTexUnit(index)->disable(); + tex_unit->disable(); } + return index; } @@ -1304,7 +1341,7 @@ 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); } } } @@ -1444,7 +1481,7 @@ 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); @@ -1471,7 +1508,7 @@ 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); @@ -1591,6 +1628,34 @@ 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; @@ -1741,7 +1806,7 @@ 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); @@ -1757,7 +1822,7 @@ 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) { @@ -1775,7 +1840,7 @@ 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) { @@ -1794,7 +1859,7 @@ 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); @@ -1856,6 +1921,23 @@ void LLGLSLShader::uniform3f(const LLStaticHashedString& uniform, GLfloat x, GLf } } +void LLGLSLShader::uniform4f(const LLStaticHashedString& uniform, GLfloat x, GLfloat y, GLfloat z, GLfloat w) +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER; + GLint location = getUniformLocation(uniform); + + if (location >= 0) + { + const auto& iter = mValue.find(location); + LLVector4 vec(x, y, z, w); + if (iter == mValue.end() || shouldChange(iter->second, vec)) + { + glUniform4f(location, x, y, z, w); + mValue[location] = vec; + } + } +} + void LLGLSLShader::uniform1fv(const LLStaticHashedString& uniform, U32 count, const GLfloat* v) { LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER; @@ -1925,6 +2007,24 @@ 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..873ab0cff5 100644 --- a/indra/llrender/llglslshader.h +++ b/indra/llrender/llglslshader.h @@ -30,6 +30,7 @@ #include "llgl.h" #include "llrender.h" #include "llstaticstringtable.h" +#include <boost/json.hpp> #include <unordered_map> class LLShaderFeatures @@ -51,12 +52,14 @@ public: bool hasAmbientOcclusion = false; bool hasSrgb = false; bool isDeferred = false; + bool hasFullGBuffer = false; bool hasScreenSpaceReflections = false; bool hasAlphaMask = false; bool hasReflectionProbes = false; bool attachNothing = false; bool hasHeroProbes = false; bool isPBRTerrain = false; + bool hasTonemap = false; }; // ============= Structure for caching shader uniforms =============== @@ -169,14 +172,14 @@ public: static U32 sMaxGLTFNodes; static void initProfile(); - static void finishProfile(bool emit_report = true); + static void finishProfile(boost::json::value& stats=sDefaultStats); static void startProfile(); static void stopProfile(); void unload(); void clearStats(); - void dumpStats(); + void dumpStats(boost::json::object& stats); // place query objects for profiling if profiling is enabled // if for_runtime is true, will place timer query only whether or not profiling is enabled @@ -208,6 +211,7 @@ 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); @@ -219,10 +223,12 @@ public: void uniform1f(const LLStaticHashedString& uniform, GLfloat v); void uniform2f(const LLStaticHashedString& uniform, GLfloat x, GLfloat y); void uniform3f(const LLStaticHashedString& uniform, GLfloat x, GLfloat y, GLfloat z); + void uniform4f(const LLStaticHashedString& uniform, GLfloat x, GLfloat y, GLfloat z, GLfloat w); void uniform1fv(const LLStaticHashedString& uniform, U32 count, const GLfloat* v); 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 +245,10 @@ 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 +257,16 @@ 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); @@ -357,6 +367,11 @@ public: private: void unloadInternal(); + // This must be static because finishProfile() is called at least once + // within a __try block. If we default its stats parameter to a temporary + // json::value, that temporary must be destroyed when the stack is + // unwound, which __try forbids. + static boost::json::value sDefaultStats; }; //UI shader (declared here so llui_libtest will link properly) 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 fb02c72338..0146ed3119 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(); @@ -50,12 +51,16 @@ extern LL_COMMON_API bool on_main_thread(); //---------------------------------------------------------------------------- const F32 MIN_TEXTURE_LIFETIME = 10.f; +const F32 CONVERSION_SCRATCH_BUFFER_GL_VERSION = 3.29f; //which power of 2 is i? //assumes i is a power of 2 > 0 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 +68,19 @@ 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 +94,7 @@ 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 +119,7 @@ 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 +140,9 @@ 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 +159,10 @@ 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; +U32* LLImageGL::sManualScratch = nullptr; + //------------------------ //**************************************************************************************************** @@ -159,20 +172,6 @@ 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 { @@ -252,11 +251,32 @@ 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); - LLImageGLThread::sEnabledTextures = thread_texture_loads; - LLImageGLThread::sEnabledMedia = thread_media_updates; + LLImageGLThread::sEnabledTextures = gGLManager.mGLVersion > 3.95f ? thread_texture_loads : false; + LLImageGLThread::sEnabledMedia = gGLManager.mGLVersion > 3.95f ? thread_media_updates : false; + } +} + +void LLImageGL::allocateConversionBuffer() +{ + if (gGLManager.mGLVersion < CONVERSION_SCRATCH_BUFFER_GL_VERSION) + { + try + { + sManualScratch = new U32[MAX_IMAGE_AREA]; + } + catch (std::bad_alloc&) + { + LLError::LLUserWarningMsg::showOutOfMemory(); + LL_ERRS() << "Failed to allocate sManualScratch" << LL_ENDL; + } } } @@ -265,6 +285,14 @@ void LLImageGL::cleanupClass() { LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; LLImageGLThread::deleteSingleton(); + if (sScratchPBO != 0) + { + glDeleteBuffers(1, &sScratchPBO); + sScratchPBO = 0; + sScratchPBOSize = 0; + } + + delete[] sManualScratch; } @@ -273,6 +301,27 @@ S32 LLImageGL::dataFormatBits(S32 dataformat) { switch (dataformat) { +#if GL_VERSION_3_0 + case GL_COMPRESSED_RED: return 8; + case GL_COMPRESSED_RG: return 16; +#endif +#if GL_VERSION_1_3 + case GL_COMPRESSED_RGB: return 24; +#endif +#if GL_VERSION_2_1 + case GL_COMPRESSED_SRGB: return 32; +#endif +#if GL_VERSION_1_3 + case GL_COMPRESSED_RGBA: return 32; +#endif +#if GL_VERSION_2_1 + case GL_COMPRESSED_SRGB_ALPHA: return 32; +#endif +#if GL_VERSION_1_3 + case GL_COMPRESSED_LUMINANCE: return 8; + case GL_COMPRESSED_LUMINANCE_ALPHA: return 16; + case GL_COMPRESSED_ALPHA: return 8; +#endif #if GL_EXT_texture_compression_s3tc || GL_EXT_texture_compression_dxt1 case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT: return 4; case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT: return 8; @@ -286,23 +335,39 @@ S32 LLImageGL::dataFormatBits(S32 dataformat) case GL_LUMINANCE: return 8; case GL_ALPHA: return 8; case GL_RED: return 8; + case GL_R8: return 8; #if GL_VERSION_1_1 + case GL_LUMINANCE8: return 8; + case GL_ALPHA8: return 8; case GL_COLOR_INDEX: return 8; + case GL_LUMINANCE8_ALPHA8: return 16; #endif case GL_LUMINANCE_ALPHA: 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_R11F_G11F_B10F: return 32; case GL_RGBA: return 32; + case GL_RGBA8: return 32; + case GL_RGB10_A2: return 32; #if GL_VERSION_2_1 case GL_SRGB_ALPHA: return 32; case GL_BGRA: return 32; // Used for QuickTime media textures on the Mac #endif 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; } } @@ -354,6 +419,7 @@ S32 LLImageGL::dataFormatComponents(S32 dataformat) case GL_COLOR_INDEX: return 1; #endif 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; @@ -362,7 +428,7 @@ S32 LLImageGL::dataFormatComponents(S32 dataformat) case GL_BGRA: return 4; // Used for QuickTime media textures on the Mac #endif default: - LL_ERRS() << "LLImageGL::Unknown format: " << dataformat << LL_ENDL; + LL_ERRS() << "LLImageGL::Unknown format: " << std::hex << dataformat << std::dec << LL_ENDL; return 0; } } @@ -374,71 +440,26 @@ void LLImageGL::updateStats(F32 current_time) { LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; sLastFrameTime = current_time; + sBindCount = 0; + sUniqueCount = 0; } //---------------------------------------------------------------------------- //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(); } @@ -470,29 +491,29 @@ 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++; @@ -509,7 +530,7 @@ LLImageGL::LLImageGL( LLGLenum formatType, LLTexUnit::eTextureAddressMode addressMode) { - init(false); + init(false, true); mTexName = texName; mTarget = target; mComponents = components; @@ -531,7 +552,7 @@ LLImageGL::~LLImageGL() } } -void LLImageGL::init(bool usemipmaps) +void LLImageGL::init(bool usemipmaps, bool allow_compression) { #if LL_IMAGEGL_THREAD_CHECK mActiveThread = LLThread::currentID(); @@ -561,11 +582,7 @@ 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; @@ -641,9 +658,6 @@ 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; @@ -808,7 +822,7 @@ 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(); } @@ -1025,7 +1039,7 @@ 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(); } @@ -1062,106 +1076,6 @@ 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) -#if GL_VERSION_1_1 - mFormatInternal = GL_LUMINANCE8; -#endif - mFormatPrimary = GL_LUMINANCE; - mFormatType = GL_UNSIGNED_BYTE; - break; - case 2: - // Use luminance alpha (for fonts) -#if GL_VERSION_1_1 - mFormatInternal = GL_LUMINANCE8_ALPHA8; -#endif - 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 GL_VERSION_1_1 - if(mFormatSwapBytes) - { - glPixelStorei(GL_UNPACK_SWAP_BYTES, 1); - stop_glerror(); - } -#endif - - return true ; -} - -void LLImageGL::postAddToAtlas() -{ -#if GL_VERSION_1_1 - if(mFormatSwapBytes) - { - glPixelStorei(GL_UNPACK_SWAP_BYTES, 0); - stop_glerror(); - } -#endif - - glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); - gGL.getTexUnit(0)->setTextureFilteringOption(mFilterOption); - stop_glerror(); -} - U32 type_width_from_pixtype(U32 pixtype) { U32 type_width = 0; @@ -1192,11 +1106,11 @@ U32 type_width_from_pixtype(U32 pixtype) bool should_stagger_image_set(bool compressed) { #if LL_DARWIN - return false; + return !compressed && on_main_thread() && gGLManager.mIsAMD; #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 } @@ -1206,15 +1120,47 @@ void sub_image_lines(U32 target, S32 miplevel, S32 x_offset, S32 y_offset, S32 w { LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; + LL_PROFILE_ZONE_NUM(width); + LL_PROFILE_ZONE_NUM(height); + U32 components = LLImageGL::dataFormatComponents(pixformat); U32 type_width = type_width_from_pixtype(pixtype); const U32 line_width = data_width * components * type_width; const U32 y_offset_end = y_offset + height; - for (U32 y_pos = y_offset; y_pos < y_offset_end; ++y_pos) + + if (width == data_width && height % 32 == 0) { - glTexSubImage2D(target, miplevel, x_offset, y_pos, width, 1, pixformat, pixtype, src); - src += line_width; + LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE("subimage - batched lines"); + + // full width, batch multiple lines at a time + // set batch size based on width + U32 batch_size = 32; + + if (width > 1024) + { + batch_size = 8; + } + else if (width > 512) + { + batch_size = 16; + } + + // full width texture, do 32 lines at a time + for (U32 y_pos = y_offset; y_pos < y_offset_end; y_pos += batch_size) + { + glTexSubImage2D(target, miplevel, x_offset, y_pos, width, batch_size, pixformat, pixtype, src); + src += line_width * batch_size; + } + } + else + { + // partial width or strange height + for (U32 y_pos = y_offset; y_pos < y_offset_end; y_pos += 1) + { + glTexSubImage2D(target, miplevel, x_offset, y_pos, width, 1, pixformat, pixtype, src); + src += line_width; + } } } @@ -1382,13 +1328,37 @@ void LLImageGL::generateTextures(S32 numTextures, U32 *textures) } } +constexpr int DELETE_DELAY = 3; // number of frames to wait before deleting textures +static std::vector<U32> sFreeList[DELETE_DELAY+1]; + +// static +void LLImageGL::updateClass() +{ + sFrameCount++; + + // wait a few frames before actually deleting the textures to avoid + // synchronization issues with the GPU + U32 idx = (sFrameCount+DELETE_DELAY) % (DELETE_DELAY+1); + + 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); + } +} + // static void LLImageGL::deleteTextures(S32 numTextures, const U32 *textures) { if (gGLManager.mInited) { - free_tex_images(numTextures, textures); - glDeleteTextures(numTextures, textures); + LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; + U32 idx = sFrameCount % (DELETE_DELAY+1); + for (S32 i = 0; i < numTextures; ++i) + { + sFreeList[idx].push_back(textures[i]); + } } } @@ -1396,90 +1366,103 @@ 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; 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 >= CONVERSION_SCRATCH_BUFFER_GL_VERSION) + { + if (pixformat == GL_ALPHA) + { //GL_ALPHA is deprecated, convert to RGBA +#if GL_VERSION_3_3 + const GLint mask[] = { GL_ZERO, GL_ZERO, GL_ZERO, GL_RED }; + glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask); +#endif + 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 +#if GL_VERSION_3_3 + const GLint mask[] = { GL_RED, GL_RED, GL_RED, GL_ONE }; + glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask); +#endif + 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 +#if GL_VERSION_3_3 + const GLint mask[] = { GL_RED, GL_RED, GL_RED, GL_GREEN }; + glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, mask); +#endif + 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; + U32 pixel_count = (U32)(width * height); + for (U32 i = 0; i < pixel_count; i++) + { + U8* pix = (U8*)&sManualScratch[i]; + pix[0] = pix[1] = pix[2] = 0; + pix[3] = ((U8*)pixels)[i]; + } + + pixels = sManualScratch; } - U32 pixel_count = (U32)(width * height); - for (U32 i = 0; i < pixel_count; i++) + pixformat = GL_RGBA; + intformat = GL_RGBA8; + } + + if (pixformat == GL_LUMINANCE_ALPHA && pixtype == GL_UNSIGNED_BYTE) + { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA + if (pixels != nullptr) { - 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 lum = ((U8*)pixels)[i * 2 + 0]; + U8 alpha = ((U8*)pixels)[i * 2 + 1]; + + U8* pix = (U8*)&sManualScratch[i]; + pix[0] = pix[1] = pix[2] = lum; + pix[3] = alpha; + } - U8* pix = (U8*)&scratch[i]; - pix[0] = pix[1] = pix[2] = lum; - pix[3] = alpha; + pixels = sManualScratch; } - } - 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 && pixtype == GL_UNSIGNED_BYTE) + { //GL_LUMINANCE_ALPHA is deprecated, convert to RGB + 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; - } + U32 pixel_count = (U32)(width * height); + for (U32 i = 0; i < pixel_count; i++) + { + U8 lum = ((U8*)pixels)[i]; - U32 pixel_count = (U32)(width * height); - for (U32 i = 0; i < pixel_count; i++) - { - U8 lum = ((U8*)pixels)[i]; + U8* pix = (U8*)&sManualScratch[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 = sManualScratch; } + pixformat = GL_RGBA; + intformat = GL_RGB8; } - pixformat = GL_RGBA; - intformat = GL_RGB8; } } @@ -1488,6 +1471,18 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt { switch (intformat) { + case GL_RED: + case GL_R8: +#if GL_VERSION_3_0 + intformat = GL_COMPRESSED_RED; +#endif + break; + case GL_RG: + case GL_RG8: +#if GL_VERSION_3_0 + intformat = GL_COMPRESSED_RG; +#endif + break; case GL_RGB: case GL_RGB8: #if GL_VERSION_1_3 @@ -1536,14 +1531,8 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt intformat = GL_COMPRESSED_ALPHA; #endif break; - case GL_RED: - case GL_R8: -#if GL_VERSION_3_0 - intformat = GL_COMPRESSED_RED; -#endif - 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; } } @@ -1559,7 +1548,7 @@ 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 { @@ -1569,21 +1558,16 @@ 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 @@ -1833,7 +1817,6 @@ 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; @@ -1886,7 +1869,7 @@ void LLImageGL::syncToMainThread(LLGLuint new_tex_name) ref(); LL::WorkQueue::postMaybe( mMainQueue, - [=]() + [=, this]() { LL_PROFILE_ZONE_NAMED("cglt - delete callback"); syncTexName(new_tex_name); @@ -1911,8 +1894,7 @@ 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) { @@ -2295,6 +2277,8 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h) return ; } + LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; + U32 length = w * h; U32 alphatotal = 0; @@ -2308,13 +2292,13 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h) // suffer the worst from aliasing when used as alpha masks. if (w >= 2 && h >= 2) { - llassert(w%2 == 0); - llassert(h%2 == 0); + llassert(w % 2 == 0); + llassert(h % 2 == 0); const GLubyte* rowstart = ((const GLubyte*) data_in) + mAlphaOffset; - for (U32 y = 0; y < h; y+=2) + for (U32 y = 0; y < h; y += 2) { const GLubyte* current = rowstart; - for (U32 x = 0; x < w; x+=2) + for (U32 x = 0; x < w; x += 2) { const U32 s1 = current[0]; alphatotal += s1; @@ -2337,7 +2321,6 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h) sample[asum/(16*4)] += 4; } - rowstart += 2 * w * mAlphaStride; } length *= 2; // we sampled everything twice, essentially @@ -2392,6 +2375,8 @@ 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; @@ -2409,7 +2394,6 @@ U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight) //---------------------------------------------------------------------------- void LLImageGL::freePickMask() { - // pickmask validity depends on old image size, delete it if (mPickMask != NULL) { delete [] mPickMask; @@ -2450,8 +2434,6 @@ void LLImageGL::updatePickMask(S32 width, S32 height, const U8* data_in) return ; } - freePickMask(); - if (mFormatType != GL_UNSIGNED_BYTE || ((mFormatPrimary != GL_RGBA) #if GL_VERSION_2_1 @@ -2460,9 +2442,11 @@ void LLImageGL::updatePickMask(S32 width, S32 height, const U8* data_in) )) { //cannot generate a pick mask for this texture + freePickMask(); return; } + #ifdef SHOW_ASSERT const U32 pickSize = createPickMask(width, height); #else // SHOW_ASSERT @@ -2561,6 +2545,109 @@ void LLImageGL::resetCurTexSizebar() sCurTexSizeBar = -1 ; sCurTexPickSize = -1 ; } + +bool LLImageGL::scaleDown(S32 desired_discard) +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE; + + if (mTarget != GL_TEXTURE_2D + || mFormatInternal == -1 // not initialized + ) + { + 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 + glViewport(0, 0, desired_width, desired_height); + + // draw a full screen triangle + if (gGL.getTexUnit(0)->bind(this, true, true)) + { + glDrawArrays(GL_TRIANGLES, 0, 3); + + free_tex_image(mTexName); + glTexImage2D(mTarget, 0, mFormatInternal, desired_width, desired_height, 0, mFormatPrimary, mFormatType, nullptr); + glCopyTexSubImage2D(mTarget, 0, 0, 0, 0, 0, desired_width, desired_height); + alloc_tex_image(desired_width, desired_height, mFormatInternal, 1); + + mTexOptionsDirty = true; + + 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 + { + LL_WARNS_ONCE("LLImageGL") << "Failed to bind texture for downscaling." << LL_ENDL; + return false; + } + } + 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; + } + +#if GL_VERSION_1_3 + glGetTexImage(mTarget, mip, mFormatPrimary, mFormatType, nullptr); +#endif + + 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..6b4492c09e 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 @@ -296,11 +298,15 @@ public: public: static void initClass(LLWindow* window, S32 num_catagories, bool skip_analyze_alpha = false, bool thread_texture_loads = false, bool thread_media_updates = false); + static void allocateConversionBuffer(); static void cleanupClass() ; private: static S32 sMaxCategories; static bool sSkipAnalyzeAlpha; + static U32 sScratchPBO; + static U32 sScratchPBOSize; + static U32* sManualScratch; //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 29ec1408d5..2e70c6953d 100644 --- a/indra/llrender/llpostprocess.cpp +++ b/indra/llrender/llpostprocess.cpp @@ -349,7 +349,7 @@ 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 633a453ab0..cbb178b6f8 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -36,6 +36,7 @@ #include "lltexture.h" #include "llshadermgr.h" #include "hbxxh.h" +#include "glm/gtc/type_ptr.hpp" #if LL_WINDOWS extern void APIENTRY gl_debug_callback(GLenum source, @@ -50,8 +51,6 @@ extern void APIENTRY gl_debug_callback(GLenum source, thread_local LLRender gGL; -const U32 BATCH_SIZE = 16334; - // Handy copies of last good GL matrices F32 gGLModelView[16]; F32 gGLLastModelView[16]; @@ -59,8 +58,8 @@ F32 gGLLastProjection[16]; F32 gGLProjection[16]; // transform from last frame's camera space to this frame's camera space (and inverse) -F32 gGLDeltaModelView[16]; -F32 gGLInverseDeltaModelView[16]; +glm::mat4 gGLDeltaModelView; +glm::mat4 gGLInverseDeltaModelView; S32 gGLViewport[4]; @@ -68,9 +67,10 @@ S32 gGLViewport[4]; U32 LLRender::sUICalls = 0; U32 LLRender::sUIVerts = 0; U32 LLTexUnit::sWhiteTexture = 0; -bool LLRender::sGLCoreProfile = true; +bool LLRender::sGLCoreProfile = false; bool LLRender::sNsightDebugSupport = false; LLVector2 LLRender::sUIGLScaleFactor = LLVector2(1.f, 1.f); +bool LLRender::sClassicMode = false; struct LLVBCache { @@ -79,6 +79,7 @@ struct LLVBCache }; static std::unordered_map<U64, LLVBCache> sVBCache; +static thread_local std::list<LLVertexBufferData> *sBufferDataList = nullptr; static const GLenum sGLTextureType[] = { @@ -123,7 +124,7 @@ 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) { @@ -153,8 +154,6 @@ void LLTexUnit::refreshState(void) { glBindTexture(GL_TEXTURE_2D, 0); } - - setTextureColorSpace(mTexColorSpace); } void LLTexUnit::activate(void) @@ -249,7 +248,6 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind) setTextureAddressMode(gl_tex->mAddressMode); setTextureFilteringOption(gl_tex->mFilterOption); } - setTextureColorSpace(mTexColorSpace); } } else @@ -326,7 +324,6 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind, S32 setTextureFilteringOption(texture->mFilterOption); stop_glerror(); } - setTextureColorSpace(mTexColorSpace); } stop_glerror(); @@ -362,7 +359,6 @@ bool LLTexUnit::bind(LLCubeMap* cubeMap) setTextureAddressMode(cubeMap->mImages[0]->mAddressMode); setTextureFilteringOption(cubeMap->mImages[0]->mFilterOption); } - setTextureColorSpace(mTexColorSpace); return true; } else @@ -411,7 +407,6 @@ bool LLTexUnit::bindManual(eTextureType type, U32 texture, bool hasMips) mCurrTexture = texture; glBindTexture(sGLTextureType[type], texture); mHasMipMaps = hasMips; - setTextureColorSpace(mTexColorSpace); } return true; } @@ -432,8 +427,6 @@ 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); @@ -455,8 +448,6 @@ 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); @@ -526,7 +517,7 @@ void LLTexUnit::setTextureFilteringOption(LLTexUnit::eTextureFilterOptions optio } } - if (gGLManager.mGLVersion >= 4.59f) + if (gGLManager.mHasAnisotropic) { if (LLImageGL::sGlobalUseAnisotropic && option == TFO_ANISOTROPIC) { @@ -658,11 +649,6 @@ void LLTexUnit::debugTextureUnit(void) } } -void LLTexUnit::setTextureColorSpace(eTextureColorSpace space) -{ - mTexColorSpace = space; -} - LLLightState::LLLightState(S32 index) : mIndex(index), mEnabled(false), @@ -765,10 +751,9 @@ void LLLightState::setPosition(const LLVector4& position) ++gGL.mLightHash; mPosition = position; //transform position by current modelview matrix - glh::vec4f pos(position.mV); - const glh::matrix4f& mat = gGL.getModelviewMatrix(); - mat.mult_matrix_vec(pos); - mPosition.set(pos.v); + glm::vec4 pos(position); + pos = gGL.getModelviewMatrix() * pos; + mPosition.set(glm::value_ptr(pos)); } void LLLightState::setConstantAttenuation(const F32& atten) @@ -820,19 +805,18 @@ void LLLightState::setSpotDirection(const LLVector3& direction) { //always set direction because modelview matrix may have changed ++gGL.mLightHash; - mSpotDirection = direction; + //transform direction by current modelview matrix - glh::vec3f dir(direction.mV); - const glh::matrix4f& mat = gGL.getModelviewMatrix(); - mat.mult_matrix_dir(dir); + glm::vec3 dir(direction); + const glm::mat3 mat(gGL.getModelviewMatrix()); + dir = mat * dir; - mSpotDirection.set(dir.v); + mSpotDirection.set(glm::value_ptr(dir)); } LLRender::LLRender() : mDirty(false), mCount(0), - mQuadCycle(0), mMode(LLRender::TRIANGLES), mCurrTextureUnitIndex(0) { @@ -860,6 +844,10 @@ LLRender::LLRender() for (U32 i = 0; i < NUM_MATRIX_MODES; ++i) { + for (U32 j = 0; j < LL_MATRIX_STACK_DEPTH; ++j) + { + mMatrix[i][j] = glm::identity<glm::mat4>(); + } mMatIdx[i] = 0; mMatHash[i] = 0; mCurMatHash[i] = 0xFFFFFFFF; @@ -922,7 +910,7 @@ void LLRender::initVertexBuffer() llassert_always(mBuffer.isNull()); stop_glerror(); mBuffer = new LLVertexBuffer(immediate_mask); - mBuffer->allocateBuffer(BATCH_SIZE, 0); + mBuffer->allocateBuffer(4096, 0); mBuffer->getVertexStrider(mVerticesp); mBuffer->getTexCoord0Strider(mTexcoordsp); mBuffer->getColorStrider(mColorsp); @@ -1000,9 +988,13 @@ void LLRender::syncLightState() shader->uniform3fv(LLShaderMgr::LIGHT_DIFFUSE, LL_NUM_LIGHT_UNITS, diffuse[0].mV); shader->uniform3fv(LLShaderMgr::LIGHT_AMBIENT, 1, mAmbientLightColor.mV); shader->uniform1i(LLShaderMgr::SUN_UP_FACTOR, sun_primary[0] ? 1 : 0); - //shader->uniform3fv(LLShaderMgr::AMBIENT, 1, mAmbientLightColor.mV); - //shader->uniform3fv(LLShaderMgr::SUNLIGHT_COLOR, 1, diffuse[0].mV); - //shader->uniform3fv(LLShaderMgr::MOONLIGHT_COLOR, 1, diffuse_b[0].mV); + + if (sClassicMode) + { + shader->uniform3fv(LLShaderMgr::AMBIENT, 1, mAmbientLightColor.mV); + shader->uniform3fv(LLShaderMgr::SUNLIGHT_COLOR, 1, diffuse[0].mV); + shader->uniform3fv(LLShaderMgr::MOONLIGHT_COLOR, 1, diffuse_b[0].mV); + } } } @@ -1023,12 +1015,12 @@ void LLRender::syncMatrices() LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr; - static glh::matrix4f cached_mvp; - static glh::matrix4f cached_inv_mdv; + static glm::mat4 cached_mvp; + static glm::mat4 cached_inv_mdv; static U32 cached_mvp_mdv_hash = 0xFFFFFFFF; static U32 cached_mvp_proj_hash = 0xFFFFFFFF; - static glh::matrix4f cached_normal; + static glm::mat4 cached_normal; static U32 cached_normal_hash = 0xFFFFFFFF; if (shader) @@ -1038,15 +1030,15 @@ void LLRender::syncMatrices() U32 i = MM_MODELVIEW; if (mMatHash[MM_MODELVIEW] != shader->mMatHash[MM_MODELVIEW]) { //update modelview, normal, and MVP - glh::matrix4f& mat = mMatrix[MM_MODELVIEW][mMatIdx[MM_MODELVIEW]]; + const glm::mat4& mat = mMatrix[MM_MODELVIEW][mMatIdx[MM_MODELVIEW]]; // if MDV has changed, update the cached inverse as well if (cached_mvp_mdv_hash != mMatHash[MM_MODELVIEW]) { - cached_inv_mdv = mat.inverse(); + cached_inv_mdv = glm::inverse(mat); } - shader->uniformMatrix4fv(name[MM_MODELVIEW], 1, GL_FALSE, mat.m); + shader->uniformMatrix4fv(name[MM_MODELVIEW], 1, GL_FALSE, glm::value_ptr(mat)); shader->mMatHash[MM_MODELVIEW] = mMatHash[MM_MODELVIEW]; //update normal matrix @@ -1055,17 +1047,17 @@ void LLRender::syncMatrices() { if (cached_normal_hash != mMatHash[i]) { - cached_normal = cached_inv_mdv.transpose(); + cached_normal = glm::transpose(cached_inv_mdv); cached_normal_hash = mMatHash[i]; } - glh::matrix4f& norm = cached_normal; + auto norm = glm::value_ptr(cached_normal); F32 norm_mat[] = { - norm.m[0], norm.m[1], norm.m[2], - norm.m[4], norm.m[5], norm.m[6], - norm.m[8], norm.m[9], norm.m[10] + norm[0], norm[1], norm[2], + norm[4], norm[5], norm[6], + norm[8], norm[9], norm[10] }; shader->uniformMatrix3fv(LLShaderMgr::NORMAL_MATRIX, 1, GL_FALSE, norm_mat); @@ -1073,7 +1065,7 @@ void LLRender::syncMatrices() if (shader->getUniformLocation(LLShaderMgr::INVERSE_MODELVIEW_MATRIX)) { - shader->uniformMatrix4fv(LLShaderMgr::INVERSE_MODELVIEW_MATRIX, 1, GL_FALSE, cached_inv_mdv.m); + shader->uniformMatrix4fv(LLShaderMgr::INVERSE_MODELVIEW_MATRIX, 1, GL_FALSE, glm::value_ptr(cached_inv_mdv)); } //update MVP matrix @@ -1086,36 +1078,36 @@ void LLRender::syncMatrices() if (cached_mvp_mdv_hash != mMatHash[i] || cached_mvp_proj_hash != mMatHash[MM_PROJECTION]) { cached_mvp = mat; - cached_mvp.mult_left(mMatrix[proj][mMatIdx[proj]]); + cached_mvp = mMatrix[proj][mMatIdx[proj]] * cached_mvp; cached_mvp_mdv_hash = mMatHash[i]; cached_mvp_proj_hash = mMatHash[MM_PROJECTION]; } - shader->uniformMatrix4fv(LLShaderMgr::MODELVIEW_PROJECTION_MATRIX, 1, GL_FALSE, cached_mvp.m); + shader->uniformMatrix4fv(LLShaderMgr::MODELVIEW_PROJECTION_MATRIX, 1, GL_FALSE, glm::value_ptr(cached_mvp)); } } i = MM_PROJECTION; if (mMatHash[MM_PROJECTION] != shader->mMatHash[MM_PROJECTION]) { //update projection matrix, normal, and MVP - glh::matrix4f& mat = mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]]; + const glm::mat4& mat = mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]]; // GZ: This was previously disabled seemingly due to a bug involving the deferred renderer's regular pushing and popping of mats. // We're reenabling this and cleaning up the code around that - that would've been the appropriate course initially. // Anything beyond the standard proj and inv proj mats are special cases. Please setup special uniforms accordingly in the future. if (shader->getUniformLocation(LLShaderMgr::INVERSE_PROJECTION_MATRIX)) { - glh::matrix4f inv_proj = mat.inverse(); - shader->uniformMatrix4fv(LLShaderMgr::INVERSE_PROJECTION_MATRIX, 1, false, inv_proj.m); + glm::mat4 inv_proj = glm::inverse(mat); + shader->uniformMatrix4fv(LLShaderMgr::INVERSE_PROJECTION_MATRIX, 1, false, glm::value_ptr(inv_proj)); } // Used by some full screen effects - such as full screen lights, glow, etc. if (shader->getUniformLocation(LLShaderMgr::IDENTITY_MATRIX)) { - shader->uniformMatrix4fv(LLShaderMgr::IDENTITY_MATRIX, 1, GL_FALSE, glh::matrix4f::identity().m); + shader->uniformMatrix4fv(LLShaderMgr::IDENTITY_MATRIX, 1, GL_FALSE, glm::value_ptr(glm::identity<glm::mat4>())); } - shader->uniformMatrix4fv(name[MM_PROJECTION], 1, GL_FALSE, mat.m); + shader->uniformMatrix4fv(name[MM_PROJECTION], 1, GL_FALSE, glm::value_ptr(mat)); shader->mMatHash[MM_PROJECTION] = mMatHash[MM_PROJECTION]; if (!mvp_done) @@ -1128,12 +1120,12 @@ void LLRender::syncMatrices() { U32 mdv = MM_MODELVIEW; cached_mvp = mat; - cached_mvp.mult_right(mMatrix[mdv][mMatIdx[mdv]]); + cached_mvp *= mMatrix[mdv][mMatIdx[mdv]]; cached_mvp_mdv_hash = mMatHash[MM_MODELVIEW]; cached_mvp_proj_hash = mMatHash[MM_PROJECTION]; } - shader->uniformMatrix4fv(LLShaderMgr::MODELVIEW_PROJECTION_MATRIX, 1, GL_FALSE, cached_mvp.m); + shader->uniformMatrix4fv(LLShaderMgr::MODELVIEW_PROJECTION_MATRIX, 1, GL_FALSE, glm::value_ptr(cached_mvp)); } } } @@ -1142,7 +1134,7 @@ void LLRender::syncMatrices() { if (mMatHash[i] != shader->mMatHash[i]) { - shader->uniformMatrix4fv(name[i], 1, GL_FALSE, mMatrix[i][mMatIdx[i]].m); + shader->uniformMatrix4fv(name[i], 1, GL_FALSE, glm::value_ptr(mMatrix[i][mMatIdx[i]])); shader->mMatHash[i] = mMatHash[i]; } } @@ -1161,12 +1153,7 @@ void LLRender::translatef(const GLfloat& x, const GLfloat& y, const GLfloat& z) flush(); { - glh::matrix4f trans_mat(1,0,0,x, - 0,1,0,y, - 0,0,1,z, - 0,0,0,1); - - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].mult_right(trans_mat); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::translate(mMatrix[mMatrixMode][mMatIdx[mMatrixMode]], glm::vec3(x, y, z)); mMatHash[mMatrixMode]++; } } @@ -1176,12 +1163,7 @@ void LLRender::scalef(const GLfloat& x, const GLfloat& y, const GLfloat& z) flush(); { - glh::matrix4f scale_mat(x,0,0,0, - 0,y,0,0, - 0,0,z,0, - 0,0,0,1); - - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].mult_right(scale_mat); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::scale(mMatrix[mMatrixMode][mMatIdx[mMatrixMode]], glm::vec3(x, y, z)); mMatHash[mMatrixMode]++; } } @@ -1191,13 +1173,7 @@ void LLRender::ortho(F32 left, F32 right, F32 bottom, F32 top, F32 zNear, F32 zF flush(); { - - glh::matrix4f ortho_mat(2.f/(right-left),0,0, -(right+left)/(right-left), - 0,2.f/(top-bottom),0, -(top+bottom)/(top-bottom), - 0,0,-2.f/(zFar-zNear), -(zFar+zNear)/(zFar-zNear), - 0,0,0,1); - - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].mult_right(ortho_mat); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] *= glm::ortho(left, right, bottom, top, zNear, zFar); mMatHash[mMatrixMode]++; } } @@ -1207,19 +1183,7 @@ void LLRender::rotatef(const GLfloat& a, const GLfloat& x, const GLfloat& y, con flush(); { - F32 r = a * DEG_TO_RAD; - - F32 c = cosf(r); - F32 s = sinf(r); - - F32 ic = 1.f-c; - - glh::matrix4f rot_mat(x*x*ic+c, x*y*ic-z*s, x*z*ic+y*s, 0, - x*y*ic+z*s, y*y*ic+c, y*z*ic-x*s, 0, - x*z*ic-y*s, y*z*ic+x*s, z*z*ic+c, 0, - 0,0,0,1); - - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].mult_right(rot_mat); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::rotate(mMatrix[mMatrixMode][mMatIdx[mMatrixMode]], glm::radians(a), glm::vec3(x,y,z)); mMatHash[mMatrixMode]++; } } @@ -1261,7 +1225,7 @@ void LLRender::loadMatrix(const GLfloat* m) { flush(); { - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].set_value((GLfloat*) m); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::make_mat4((GLfloat*) m); mMatHash[mMatrixMode]++; } } @@ -1270,9 +1234,7 @@ void LLRender::multMatrix(const GLfloat* m) { flush(); { - glh::matrix4f mat((GLfloat*) m); - - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].mult_right(mat); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] *= glm::make_mat4(m); mMatHash[mMatrixMode]++; } } @@ -1315,17 +1277,17 @@ void LLRender::loadIdentity() { llassert_always(mMatrixMode < NUM_MATRIX_MODES) ; - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].make_identity(); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::identity<glm::mat4>(); mMatHash[mMatrixMode]++; } } -const glh::matrix4f& LLRender::getModelviewMatrix() +const glm::mat4& LLRender::getModelviewMatrix() { return mMatrix[MM_MODELVIEW][mMatIdx[MM_MODELVIEW]]; } -const glh::matrix4f& LLRender::getProjectionMatrix() +const glm::mat4& LLRender::getProjectionMatrix() { return mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]]; } @@ -1337,8 +1299,9 @@ void LLRender::translateUI(F32 x, F32 y, F32 z) LL_ERRS() << "Need to push a UI translation frame before offsetting" << LL_ENDL; } - LLVector4a add(x,y,z); - mUIOffset.back().add(add); + mUIOffset.back().mV[0] += x; + mUIOffset.back().mV[1] += y; + mUIOffset.back().mV[2] += z; } void LLRender::scaleUI(F32 x, F32 y, F32 z) @@ -1348,15 +1311,14 @@ void LLRender::scaleUI(F32 x, F32 y, F32 z) LL_ERRS() << "Need to push a UI transformation frame before scaling." << LL_ENDL; } - LLVector4a scale(x,y,z); - mUIScale.back().mul(scale); + mUIScale.back().scaleVec(LLVector3(x,y,z)); } void LLRender::pushUIMatrix() { if (mUIOffset.empty()) { - mUIOffset.emplace_back(LLVector4a::getZero()); + mUIOffset.emplace_back(0.f,0.f,0.f); } else { @@ -1365,7 +1327,7 @@ void LLRender::pushUIMatrix() if (mUIScale.empty()) { - mUIScale.emplace_back(LLVector4a(1.f)); + mUIScale.emplace_back(1.f,1.f,1.f); } else { @@ -1389,7 +1351,7 @@ LLVector3 LLRender::getUITranslation() { return LLVector3(0,0,0); } - return LLVector3(mUIOffset.back().getF32ptr()); + return mUIOffset.back(); } LLVector3 LLRender::getUIScale() @@ -1398,17 +1360,18 @@ LLVector3 LLRender::getUIScale() { return LLVector3(1,1,1); } - return LLVector3(mUIScale.back().getF32ptr()); + return mUIScale.back(); } + void LLRender::loadUIIdentity() { if (mUIOffset.empty()) { LL_ERRS() << "Need to push UI translation frame before clearing offset." << LL_ENDL; } - mUIOffset.back().splat(0.f); - mUIScale.back().splat(1.f); + mUIOffset.back().setVec(0,0,0); + mUIScale.back().setVec(1,1,1); } void LLRender::setColorMask(bool writeColor, bool writeAlpha) @@ -1531,7 +1494,7 @@ 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; @@ -1560,17 +1523,35 @@ 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) { - if (mode == LLRender::QUADS) - { - mQuadCycle = 1; - } - - if (mMode == LLRender::QUADS || - mMode == LLRender::LINES || + if (mMode == LLRender::LINES || mMode == LLRender::TRIANGLES || mMode == LLRender::POINTS) { @@ -1593,22 +1574,23 @@ void LLRender::end() //IMM_ERRS << "GL begin and end called with no vertices specified." << LL_ENDL; } - if ((mMode != LLRender::QUADS && - mMode != LLRender::LINES && + if ((mMode != LLRender::LINES && mMode != LLRender::TRIANGLES && mMode != LLRender::POINTS) || - mCount > (BATCH_SIZE / 2)) + mCount > 2048) { flush(); } } + void LLRender::flush() { STOP_GLERROR; if (mCount > 0) { LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE; - llassert(LLGLSLShader::sCurBoundShaderPtr != nullptr); + llassert_always(LLGLSLShader::sCurBoundShaderPtr != nullptr); + if (!mUIOffset.empty()) { sUICalls++; @@ -1618,21 +1600,12 @@ void LLRender::flush() //store mCount in a local variable to avoid re-entrance (drawArrays may call flush) U32 count = mCount; - if (mMode == LLRender::QUADS && !sGLCoreProfile) - { - if (mCount%4 != 0) - { - count -= (mCount % 4); - LL_WARNS() << "Incomplete quad requested." << LL_ENDL; - } - } - if (mMode == LLRender::TRIANGLES) { if (mCount%3 != 0) { - count -= (mCount % 3); - LL_WARNS() << "Incomplete triangle requested." << LL_ENDL; + count -= (mCount % 3); + LL_WARNS() << "Incomplete triangle requested." << LL_ENDL; } } @@ -1650,177 +1623,184 @@ void LLRender::flush() if (mBuffer) { - HBXXH64 hash; + LLVertexBuffer *vb; + U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; + if (sBufferDataList) { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash"); + 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 + { + 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(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); - if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) - { - vb->setTexCoord0Data(mTexcoordsp.get()); - } + 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_COLOR) + 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->setColorData(mColorsp.get()); + iter = sVBCache.erase(iter); } - - //LL_INFOS() << "LLVertexBuffer::sMappingMode " << LLVertexBuffer::sMappingMode << LL_ENDL; - if(LLVertexBuffer::sMappingMode == 3) + else { - vb->unmapBuffer(); + ++iter; } + } + } + } + return vb; +} - 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(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(); + vb->drawArrays(mode, 0, count); +} - mCount = 0; - } +void LLRender::resetStriders(S32 count) +{ + mVerticesp[0] = mVerticesp[count]; + mTexcoordsp[0] = mTexcoordsp[count]; + mColorsp[0] = mColorsp[count]; + + mCount = 0; } -void LLRender::vertex4a(const LLVector4a& vertex) +void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) { //the range of mVerticesp, mColorsp and mTexcoordsp is [0, 4095] - if (mCount > BATCH_SIZE / 2) + if (mCount > 2048) { //break when buffer gets reasonably full to keep GL command buffers happy and avoid overflow below switch (mMode) { case LLRender::POINTS: flush(); break; case LLRender::TRIANGLES: if (mCount%3==0) flush(); break; - case LLRender::QUADS: if(mCount%4 == 0) flush(); break; case LLRender::LINES: if (mCount%2 == 0) flush(); break; } } - if (mCount > BATCH_SIZE - 2) + if (mCount > 4094) { - LL_WARNS() << "GL immediate mode overflow. Some geometry not drawn." << LL_ENDL; + // LL_WARNS() << "GL immediate mode overflow. Some geometry not drawn." << LL_ENDL; return; } if (mUIOffset.empty()) { - mVerticesp[mCount] = vertex; + mVerticesp[mCount].set(x,y,z); } else { - mVerticesp[mCount].setAdd(vertex, mUIOffset.back()); - mVerticesp[mCount].mul(mUIScale.back()); - } - - if (mMode == LLRender::QUADS && LLRender::sGLCoreProfile) - { - mQuadCycle++; - if (mQuadCycle == 4) - { //copy two vertices so fourth quad element will add a triangle - mQuadCycle = 0; - - mCount++; - mVerticesp[mCount] = mVerticesp[mCount-3]; - mColorsp[mCount] = mColorsp[mCount-3]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-3]; - - mCount++; - mVerticesp[mCount] = mVerticesp[mCount-2]; - mColorsp[mCount] = mColorsp[mCount-2]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-2]; - } + LLVector3 vert = (LLVector3(x,y,z)+mUIOffset.back()).scaledVec(mUIScale.back()); + mVerticesp[mCount].set(vert.mV[VX], vert.mV[VY], vert.mV[VZ]); } mCount++; @@ -1831,56 +1811,19 @@ void LLRender::vertex4a(const LLVector4a& vertex) void LLRender::vertexBatchPreTransformed(LLVector4a* verts, S32 vert_count) { - if (mCount + vert_count > BATCH_SIZE - 2) + if (mCount + vert_count > 4094) { // LL_WARNS() << "GL immediate mode overflow. Some geometry not drawn." << LL_ENDL; return; } - if (sGLCoreProfile && mMode == LLRender::QUADS) - { //quads are deprecated, convert to triangle list - S32 i = 0; - - while (i < vert_count) - { - //read first three - mVerticesp[mCount++] = verts[i++]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount++] = verts[i++]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount++] = verts[i++]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - //copy two - mVerticesp[mCount++] = verts[i-3]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount++] = verts[i-1]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - //copy last one - mVerticesp[mCount++] = verts[i++]; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - } - } - else + for (S32 i = 0; i < vert_count; i++) { - for (S32 i = 0; i < vert_count; i++) - { - mVerticesp[mCount] = verts[i]; + mVerticesp[mCount] = verts[i]; - mCount++; - mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - } + mCount++; + mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; + mColorsp[mCount] = mColorsp[mCount-1]; } if( mCount > 0 ) // ND: Guard against crashes if mCount is zero, yes it can happen @@ -1889,56 +1832,19 @@ void LLRender::vertexBatchPreTransformed(LLVector4a* verts, S32 vert_count) void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, S32 vert_count) { - if (mCount + vert_count > BATCH_SIZE - 2) + if (mCount + vert_count > 4094) { // LL_WARNS() << "GL immediate mode overflow. Some geometry not drawn." << LL_ENDL; return; } - if (sGLCoreProfile && mMode == LLRender::QUADS) - { //quads are deprecated, convert to triangle list - S32 i = 0; - - while (i < vert_count) - { - //read first three - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount++] = uvs[i++]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount++] = uvs[i++]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount++] = uvs[i++]; - mColorsp[mCount] = mColorsp[mCount-1]; - - //copy last two - mVerticesp[mCount] = verts[i-3]; - mTexcoordsp[mCount++] = uvs[i-3]; - mColorsp[mCount] = mColorsp[mCount-1]; - - mVerticesp[mCount] = verts[i-1]; - mTexcoordsp[mCount++] = uvs[i-1]; - mColorsp[mCount] = mColorsp[mCount-1]; - - //copy last one - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount++] = uvs[i++]; - mColorsp[mCount] = mColorsp[mCount-1]; - } - } - else + for (S32 i = 0; i < vert_count; i++) { - for (S32 i = 0; i < vert_count; i++) - { - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; + mVerticesp[mCount] = verts[i]; + mTexcoordsp[mCount] = uvs[i]; - mCount++; - mColorsp[mCount] = mColorsp[mCount-1]; - } + mCount++; + mColorsp[mCount] = mColorsp[mCount-1]; } if (mCount > 0) @@ -1950,57 +1856,19 @@ void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, S32 void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, LLColor4U* colors, S32 vert_count) { - if (mCount + vert_count > BATCH_SIZE - 2) + if (mCount + vert_count > 4094) { // LL_WARNS() << "GL immediate mode overflow. Some geometry not drawn." << LL_ENDL; return; } - - if (sGLCoreProfile && mMode == LLRender::QUADS) - { //quads are deprecated, convert to triangle list - S32 i = 0; - - while (i < vert_count) - { - //read first three - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; - mColorsp[mCount++] = colors[i++]; - - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; - mColorsp[mCount++] = colors[i++]; - - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; - mColorsp[mCount++] = colors[i++]; - - //copy last two - mVerticesp[mCount] = verts[i-3]; - mTexcoordsp[mCount] = uvs[i-3]; - mColorsp[mCount++] = colors[i-3]; - - mVerticesp[mCount] = verts[i-1]; - mTexcoordsp[mCount] = uvs[i-1]; - mColorsp[mCount++] = colors[i-1]; - - //copy last one - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; - mColorsp[mCount++] = colors[i++]; - } - } - else + for (S32 i = 0; i < vert_count; i++) { - for (S32 i = 0; i < vert_count; i++) - { - mVerticesp[mCount] = verts[i]; - mTexcoordsp[mCount] = uvs[i]; - mColorsp[mCount] = colors[i]; + mVerticesp[mCount] = verts[i]; + mTexcoordsp[mCount] = uvs[i]; + mColorsp[mCount] = colors[i]; - mCount++; - } + mCount++; } if (mCount > 0) @@ -2011,6 +1879,25 @@ void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, LLCo } } +void LLRender::vertex2i(const GLint& x, const GLint& y) +{ + vertex3f((GLfloat) x, (GLfloat) y, 0); +} + +void LLRender::vertex2f(const GLfloat& x, const GLfloat& y) +{ + vertex3f(x,y,0); +} + +void LLRender::vertex2fv(const GLfloat* v) +{ + vertex3f(v[0], v[1], 0); +} + +void LLRender::vertex3fv(const GLfloat* v) +{ + vertex3f(v[0], v[1], v[2]); +} void LLRender::texCoord2f(const GLfloat& x, const GLfloat& y) { @@ -2149,9 +2036,11 @@ void LLRender::debugTexUnits(void) case LLTexUnit::TT_TEXTURE: LL_CONT << "Texture 2D"; break; +#if GL_VERSION_3_1 case LLTexUnit::TT_RECT_TEXTURE: LL_CONT << "Texture Rectangle"; break; +#endif case LLTexUnit::TT_CUBE_MAP: LL_CONT << "Cube Map"; break; @@ -2165,85 +2054,93 @@ void LLRender::debugTexUnits(void) LL_INFOS("TextureUnit") << "Active TexUnit Enabled : " << active_enabled << LL_ENDL; } - - -glh::matrix4f copy_matrix(F32* src) +glm::mat4 get_current_modelview() { - glh::matrix4f ret; - ret.set_value(src); - return ret; + return glm::make_mat4(gGLModelView); } -glh::matrix4f get_current_modelview() +glm::mat4 get_current_projection() { - return copy_matrix(gGLModelView); + return glm::make_mat4(gGLProjection); } -glh::matrix4f get_current_projection() +glm::mat4 get_last_modelview() { - return copy_matrix(gGLProjection); + return glm::make_mat4(gGLLastModelView); } -glh::matrix4f get_last_modelview() +glm::mat4 get_last_projection() { - return copy_matrix(gGLLastModelView); + return glm::make_mat4(gGLLastProjection); } -glh::matrix4f get_last_projection() -{ - return copy_matrix(gGLLastProjection); -} - -void copy_matrix(const glh::matrix4f& src, F32* dst) +void copy_matrix(const glm::mat4& src, F32* dst) { + auto matp = glm::value_ptr(src); for (U32 i = 0; i < 16; i++) { - dst[i] = src.m[i]; + dst[i] = matp[i]; } } -void set_current_modelview(const glh::matrix4f& mat) +void set_current_modelview(const glm::mat4& mat) { copy_matrix(mat, gGLModelView); } -void set_current_projection(glh::matrix4f& mat) +void set_current_projection(const glm::mat4& mat) { copy_matrix(mat, gGLProjection); } -glh::matrix4f gl_ortho(GLfloat left, GLfloat right, GLfloat bottom, GLfloat top, GLfloat znear, GLfloat zfar) +void set_last_modelview(const glm::mat4& mat) { - glh::matrix4f ret( - 2.f/(right-left), 0.f, 0.f, -(right+left)/(right-left), - 0.f, 2.f/(top-bottom), 0.f, -(top+bottom)/(top-bottom), - 0.f, 0.f, -2.f/(zfar-znear), -(zfar+znear)/(zfar-znear), - 0.f, 0.f, 0.f, 1.f); - - return ret; + copy_matrix(mat, gGLLastModelView); } -glh::matrix4f gl_perspective(GLfloat fovy, GLfloat aspect, GLfloat zNear, GLfloat zFar) +void set_last_projection(const glm::mat4& mat) { - GLfloat f = 1.f/tanf(DEG_TO_RAD*fovy/2.f); - - return glh::matrix4f(f/aspect, 0, 0, 0, - 0, f, 0, 0, - 0, 0, (zFar+zNear)/(zNear-zFar), (2.f*zFar*zNear)/(zNear-zFar), - 0, 0, -1.f, 0); + copy_matrix(mat, gGLLastProjection); } -glh::matrix4f gl_lookat(LLVector3 eye, LLVector3 center, LLVector3 up) +glm::vec3 mul_mat4_vec3(const glm::mat4& mat, const glm::vec3& vec) { - LLVector3 f = center-eye; - f.normVec(); - up.normVec(); - LLVector3 s = f % up; - LLVector3 u = s % f; +#if 1 // SIMD path results in strange crashes. Fall back to scalar for now. + const float w = vec[0] * mat[0][3] + vec[1] * mat[1][3] + vec[2] * mat[2][3] + mat[3][3]; + return glm::vec3( + (vec[0] * mat[0][0] + vec[1] * mat[1][0] + vec[2] * mat[2][0] + mat[3][0]) / w, + (vec[0] * mat[0][1] + vec[1] * mat[1][1] + vec[2] * mat[2][1] + mat[3][1]) / w, + (vec[0] * mat[0][2] + vec[1] * mat[1][2] + vec[2] * mat[2][2] + mat[3][2]) / w + ); +#else + LLVector4a x, y, z, s, t, p, q; + + x.splat(vec.x); + y.splat(vec.y); + z.splat(vec.z); + + s.splat<3>(mat[0].data); + t.splat<3>(mat[1].data); + p.splat<3>(mat[2].data); + q.splat<3>(mat[3].data); - return glh::matrix4f(s[0], s[1], s[2], 0, - u[0], u[1], u[2], 0, - -f[0], -f[1], -f[2], 0, - 0, 0, 0, 1); + s.mul(x); + t.mul(y); + p.mul(z); + q.add(s); + t.add(p); + q.add(t); + x.mul(mat[0].data); + y.mul(mat[1].data); + z.mul(mat[2].data); + + x.add(y); + z.add(mat[3].data); + LLVector4a res; + res.load3(glm::value_ptr(vec)); + res.setAdd(x, z); + res.div(q); + return glm::make_vec3(res.getF32ptr()); +#endif } diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h index 5852201c94..97c47bcae9 100644 --- a/indra/llrender/llrender.h +++ b/indra/llrender/llrender.h @@ -43,15 +43,17 @@ #include "llpointer.h" #include "llglheaders.h" #include "llmatrix4a.h" -#include "glh/glh_linear.h" +#include "glm/mat4x4.hpp" #include <array> +#include <list> class LLVertexBuffer; class LLCubeMap; class LLImageGL; class LLRenderTarget; -class LLTexture ; +class LLTexture; +class LLVertexBufferData; #define LL_MATRIX_STACK_DEPTH 32 @@ -67,10 +69,16 @@ public: typedef enum { TT_TEXTURE = 0, // Standard 2D Texture +#if GL_VERSION_3_1 TT_RECT_TEXTURE, // Non power of 2 texture +#endif TT_CUBE_MAP, // 6-sided cube map texture +#if GL_VERSION_4_0 TT_CUBE_MAP_ARRAY, // Array of cube maps +#endif +#if GL_VERSION_3_2 TT_MULTISAMPLE_TEXTURE, // see GL_ARB_texture_multisample +#endif TT_TEXTURE_3D, // standard 3D Texture TT_NONE, // No texture type is currently enabled } eTextureType; @@ -220,17 +228,12 @@ 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; @@ -294,11 +297,18 @@ 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 @@ -316,7 +326,6 @@ public: POINTS, LINES, LINE_STRIP, - QUADS, LINE_LOOP, NUM_MODES }; @@ -398,8 +407,8 @@ public: void matrixMode(eMatrixMode mode); eMatrixMode getMatrixMode(); - const glh::matrix4f& getModelviewMatrix(); - const glh::matrix4f& getProjectionMatrix(); + const glm::mat4& getModelviewMatrix(); + const glm::mat4& getProjectionMatrix(); void syncMatrices(); void syncLightState(); @@ -414,17 +423,20 @@ 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(); - LL_FORCE_INLINE void vertex2i(const GLint& x, const GLint& y) { vertex4a(LLVector4a((GLfloat)x,(GLfloat)y,0.f)); } - LL_FORCE_INLINE void vertex2f(const GLfloat& x, const GLfloat& y) { vertex4a(LLVector4a(x,y,0.f)); } - LL_FORCE_INLINE void vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) { vertex4a(LLVector4a(x,y,z)); } - LL_FORCE_INLINE void vertex2fv(const GLfloat* v) { vertex4a(LLVector4a(v[0],v[1],0.f)); } - LL_FORCE_INLINE void vertex3fv(const GLfloat* v) { vertex4a(LLVector4a(v[0],v[1],v[2])); } - - void vertex4a(const LLVector4a& v); + 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); + void vertex2fv(const GLfloat* v); + void vertex3fv(const GLfloat* v); void texCoord2i(const GLint& x, const GLint& y); void texCoord2f(const GLfloat& x, const GLfloat& y); @@ -484,29 +496,32 @@ public: static bool sGLCoreProfile; static bool sNsightDebugSupport; static LLVector2 sUIGLScaleFactor; - - //static U32 sMappingMode; + static bool sClassicMode; // classic sky mode active 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]; - glh::matrix4f mMatrix[NUM_MATRIX_MODES][LL_MATRIX_STACK_DEPTH]; + glm::mat4 mMatrix[NUM_MATRIX_MODES][LL_MATRIX_STACK_DEPTH]; U32 mCurMatHash[NUM_MATRIX_MODES]; U32 mLightHash; LLColor4 mAmbientLightColor; bool mDirty; - U32 mQuadCycle; U32 mCount; U32 mMode; U32 mCurrTextureUnitIndex; bool mCurrColorMask[4]; LLPointer<LLVertexBuffer> mBuffer; - LLStrider<LLVector4a> mVerticesp; + LLStrider<LLVector4a> mVerticesp; LLStrider<LLVector2> mTexcoordsp; LLStrider<LLColor4U> mColorsp; std::array<LLTexUnit, LL_NUM_TEXTURE_LAYERS> mTexUnits; @@ -518,9 +533,8 @@ private: eBlendFactor mCurrBlendAlphaSFactor; eBlendFactor mCurrBlendAlphaDFactor; - std::vector<LLVector4a> mUIOffset; - std::vector<LLVector4a> mUIScale; - + std::vector<LLVector3> mUIOffset; + std::vector<LLVector3> mUIScale; }; extern F32 gGLModelView[16]; @@ -528,8 +542,8 @@ extern F32 gGLLastModelView[16]; extern F32 gGLLastProjection[16]; extern F32 gGLProjection[16]; extern S32 gGLViewport[4]; -extern F32 gGLDeltaModelView[16]; -extern F32 gGLInverseDeltaModelView[16]; +extern glm::mat4 gGLDeltaModelView; +extern glm::mat4 gGLInverseDeltaModelView; extern thread_local LLRender gGL; @@ -540,19 +554,20 @@ const F32 OGL_TO_CFR_ROTATION[16] = { 0.f, 0.f, -1.f, 0.f, // -Z becomes X 0.f, 1.f, 0.f, 0.f, // Y becomes Z 0.f, 0.f, 0.f, 1.f }; -glh::matrix4f copy_matrix(F32* src); -glh::matrix4f get_current_modelview(); -glh::matrix4f get_current_projection(); -glh::matrix4f get_last_modelview(); -glh::matrix4f get_last_projection(); +glm::mat4 copy_matrix(F32* src); +glm::mat4 get_current_modelview(); +glm::mat4 get_current_projection(); +glm::mat4 get_last_modelview(); +glm::mat4 get_last_projection(); -void copy_matrix(const glh::matrix4f& src, F32* dst); -void set_current_modelview(const glh::matrix4f& mat); -void set_current_projection(glh::matrix4f& mat); +void copy_matrix(const glm::mat4& src, F32* dst); +void set_current_modelview(const glm::mat4& mat); +void set_current_projection(const glm::mat4& mat); +void set_last_modelview(const glm::mat4& mat); +void set_last_projection(const glm::mat4& mat); -glh::matrix4f gl_ortho(GLfloat left, GLfloat right, GLfloat bottom, GLfloat top, GLfloat znear, GLfloat zfar); -glh::matrix4f gl_perspective(GLfloat fovy, GLfloat aspect, GLfloat zNear, GLfloat zFar); -glh::matrix4f gl_lookat(LLVector3 eye, LLVector3 center, LLVector3 up); +// glh compat +glm::vec3 mul_mat4_vec3(const glm::mat4& mat, const glm::vec3& vec); #define LL_SHADER_LOADING_WARNS(...) LL_WARNS() diff --git a/indra/llrender/llrender2dutils.cpp b/indra/llrender/llrender2dutils.cpp index 97e9875afb..9144ce6d62 100644 --- a/indra/llrender/llrender2dutils.cpp +++ b/indra/llrender/llrender2dutils.cpp @@ -119,10 +119,10 @@ void gl_rect_2d(S32 left, S32 top, S32 right, S32 bottom, bool filled ) { gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE); + // Counterclockwise quad will face the viewer if( filled ) { gGL.begin( LLRender::TRIANGLES ); - { gGL.vertex2i(left, top); gGL.vertex2i(left, bottom); gGL.vertex2i(right, bottom); @@ -130,7 +130,6 @@ void gl_rect_2d(S32 left, S32 top, S32 right, S32 bottom, bool filled ) gGL.vertex2i(left, top); gGL.vertex2i(right, bottom); gGL.vertex2i(right, top); - } gGL.end(); } else @@ -175,73 +174,71 @@ void gl_drop_shadow(S32 left, S32 top, S32 right, S32 bottom, const LLColor4 &st LLColor4 end_color = start_color; end_color.mV[VALPHA] = 0.f; + gGL.begin(LLRender::TRIANGLES); - // Right edge + // Right edge, CCW faces screen gGL.color4fv(start_color.mV); - gGL.vertex2i(right, top-lines); + gGL.vertex2i(right, top - lines); gGL.vertex2i(right, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(right+lines, bottom); + gGL.vertex2i(right + lines, bottom); gGL.color4fv(start_color.mV); - gGL.vertex2i(right, top-lines); + gGL.vertex2i(right, top - lines); gGL.color4fv(end_color.mV); - gGL.vertex2i(right+lines, bottom); - gGL.vertex2i(right+lines, top-lines); + gGL.vertex2i(right + lines, bottom); + gGL.vertex2i(right + lines, top - lines); - // Bottom edge + // Bottom edge, CCW faces screen gGL.color4fv(start_color.mV); gGL.vertex2i(right, bottom); - gGL.vertex2i(left+lines, bottom); + gGL.vertex2i(left + lines, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(left+lines, bottom-lines); + gGL.vertex2i(left + lines, bottom - lines); gGL.color4fv(start_color.mV); gGL.vertex2i(right, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(left+lines, bottom-lines); - gGL.vertex2i(right, bottom-lines); + gGL.vertex2i(left + lines, bottom - lines); + gGL.vertex2i(right, bottom - lines); // bottom left Corner gGL.color4fv(start_color.mV); - gGL.vertex2i(left+lines, bottom); + gGL.vertex2i(left + lines, bottom); gGL.color4fv(end_color.mV); gGL.vertex2i(left, bottom); - - // bottom left corner - gGL.vertex2i(left+1, bottom-lines+1); + // make the bottom left corner not sharp + gGL.vertex2i(left + 1, bottom - lines + 1); gGL.color4fv(start_color.mV); - gGL.vertex2i(left+lines, bottom); + gGL.vertex2i(left + lines, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(left+1, bottom-lines+1); - gGL.vertex2i(left+lines, bottom-lines); + gGL.vertex2i(left + 1, bottom - lines + 1); + gGL.vertex2i(left + lines, bottom - lines); // bottom right corner gGL.color4fv(start_color.mV); gGL.vertex2i(right, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(right, bottom-lines); - + gGL.vertex2i(right, bottom - lines); // make the rightmost corner not sharp - gGL.vertex2i(right+lines-1, bottom-lines+1); + gGL.vertex2i(right + lines - 1, bottom - lines + 1); gGL.color4fv(start_color.mV); gGL.vertex2i(right, bottom); gGL.color4fv(end_color.mV); - gGL.vertex2i(right+lines-1, bottom-lines+1); - gGL.vertex2i(right+lines, bottom); + gGL.vertex2i(right + lines - 1, bottom - lines + 1); + gGL.vertex2i(right + lines, bottom); // top right corner gGL.color4fv(start_color.mV); - gGL.vertex2i( right, top-lines ); + gGL.vertex2i(right, top - lines); gGL.color4fv(end_color.mV); - gGL.vertex2i( right+lines, top-lines ); - + gGL.vertex2i(right + lines, top - lines); // make the corner not sharp - gGL.vertex2i( right+lines-1, top-1 ); + gGL.vertex2i(right + lines - 1, top - 1); gGL.color4fv(start_color.mV); - gGL.vertex2i( right, top-lines ); + gGL.vertex2i(right, top - lines); gGL.color4fv(end_color.mV); - gGL.vertex2i( right+lines-1, top-1 ); - gGL.vertex2i( right, top ); + gGL.vertex2i(right + lines - 1, top - 1); + gGL.vertex2i(right, top); gGL.end(); stop_glerror(); @@ -390,7 +387,7 @@ 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(); @@ -401,8 +398,8 @@ 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); @@ -439,253 +436,261 @@ 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], ui_translation.mV[VX] + width * ui_scale.mV[VX], ui_translation.mV[VY]); - LLGLSUIDefault gls_ui; - gGL.getTexUnit(0)->bind(image, true); gGL.color4fv(color.mV); - const S32 NUM_VERTICES = 9 * 6; - LLVector2 uv[NUM_VERTICES]; - LLVector4a pos[NUM_VERTICES]; + constexpr S32 NUM_VERTICES = 9 * 2 * 3; // 9 quads, 2 triangles per quad, 3 vertices per triangle + static thread_local LLVector2 uv[NUM_VERTICES]; + static thread_local LLVector4a pos[NUM_VERTICES]; S32 index = 0; gGL.begin(LLRender::TRIANGLES); { - // draw bottom left - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_outer_rect.mBottom, 0.f); + // draw bottom left triangles + // 1 + uv[index].set(uv_outer_rect.mLeft, uv_outer_rect.mBottom); + pos[index].set(draw_outer_rect.mLeft, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_outer_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_outer_rect.mLeft, uv_outer_rect.mBottom); + pos[index].set(draw_outer_rect.mLeft, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - // draw bottom middle - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); + // draw bottom middle triangles + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - // draw bottom right - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); + // draw bottom right triangles + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_outer_rect.mBottom, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_outer_rect.mBottom); + pos[index].set(draw_outer_rect.mRight, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - // draw left - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); + // draw left triangles + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - // draw middle - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + // draw middle triangles + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - // draw right - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + // draw right triangles + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mBottom); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); + // 2 + uv[index].set(uv_center_rect.mRight, uv_center_rect.mBottom); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mBottom, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - // draw top left - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); + // draw top left triangles + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); + // 2 + uv[index].set(uv_outer_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mLeft, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mLeft, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mLeft, uv_outer_rect.mTop); + pos[index].set(draw_outer_rect.mLeft, draw_outer_rect.mTop, 0.f); index++; - // draw top middle - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + // draw top middle triangles + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); + // 2 + uv[index].set(uv_center_rect.mLeft, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mLeft, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mLeft, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mLeft, draw_outer_rect.mTop, 0.f); index++; - // draw top right - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + // draw top right triangles + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_outer_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_outer_rect.mTop); + pos[index].set(draw_outer_rect.mRight, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_center_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); + // 2 + uv[index].set(uv_center_rect.mRight, uv_center_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_center_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_outer_rect.mRight, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_outer_rect.mRight, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_outer_rect.mRight, uv_outer_rect.mTop); + pos[index].set(draw_outer_rect.mRight, draw_outer_rect.mTop, 0.f); index++; - uv[index] = LLVector2(uv_center_rect.mRight, uv_outer_rect.mTop); - pos[index] = LLVector4a(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); + uv[index].set(uv_center_rect.mRight, uv_outer_rect.mTop); + pos[index].set(draw_center_rect.mRight, draw_outer_rect.mTop, 0.f); index++; gGL.vertexBatchPreTransformed(pos, uv, NUM_VERTICES); @@ -712,8 +717,6 @@ void gl_draw_scaled_rotated_image(S32 x, S32 y, S32 width, S32 height, F32 degre return; } - LLGLSUIDefault gls_ui; - if(image != NULL) { gGL.getTexUnit(0)->bind(image, true); @@ -727,9 +730,9 @@ void gl_draw_scaled_rotated_image(S32 x, S32 y, S32 width, S32 height, F32 degre if (degrees == 0.f) { - const S32 NUM_VERTICES = 6; - LLVector2 uv[NUM_VERTICES]; - LLVector4a pos[NUM_VERTICES]; + constexpr S32 NUM_VERTICES = 2 * 3; + static thread_local LLVector2 uv[NUM_VERTICES +1]; + static thread_local LLVector4a pos[NUM_VERTICES +1]; gGL.begin(LLRender::TRIANGLES); { @@ -742,28 +745,28 @@ void gl_draw_scaled_rotated_image(S32 x, S32 y, S32 width, S32 height, F32 degre S32 scaled_width = ll_round(width * ui_scale.mV[VX]); S32 scaled_height = ll_round(height * ui_scale.mV[VY]); - uv[index] = LLVector2(uv_rect.mRight, uv_rect.mTop); - pos[index] = LLVector4a(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY] + scaled_height, 0.f); + uv[index].set(uv_rect.mRight, uv_rect.mTop); + pos[index].set(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY] + scaled_height, 0.f); index++; - uv[index] = LLVector2(uv_rect.mLeft, uv_rect.mTop); - pos[index] = LLVector4a(ui_translation.mV[VX], ui_translation.mV[VY] + scaled_height, 0.f); + uv[index].set(uv_rect.mLeft, uv_rect.mTop); + pos[index].set(ui_translation.mV[VX], ui_translation.mV[VY] + scaled_height, 0.f); index++; - uv[index] = LLVector2(uv_rect.mLeft, uv_rect.mBottom); - pos[index] = LLVector4a(ui_translation.mV[VX], ui_translation.mV[VY], 0.f); + uv[index].set(uv_rect.mLeft, uv_rect.mBottom); + pos[index].set(ui_translation.mV[VX], ui_translation.mV[VY], 0.f); index++; - uv[index] = LLVector2(uv_rect.mRight, uv_rect.mTop); - pos[index] = LLVector4a(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY] + scaled_height, 0.f); + uv[index].set(uv_rect.mRight, uv_rect.mTop); + pos[index].set(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY] + scaled_height, 0.f); index++; - uv[index] = LLVector2(uv_rect.mLeft, uv_rect.mBottom); - pos[index] = LLVector4a(ui_translation.mV[VX], ui_translation.mV[VY], 0.f); + uv[index].set(uv_rect.mLeft, uv_rect.mBottom); + pos[index].set(ui_translation.mV[VX], ui_translation.mV[VY], 0.f); index++; - uv[index] = LLVector2(uv_rect.mRight, uv_rect.mBottom); - pos[index] = LLVector4a(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY], 0.f); + uv[index].set(uv_rect.mRight, uv_rect.mBottom); + pos[index].set(ui_translation.mV[VX] + scaled_width, ui_translation.mV[VY], 0.f); index++; gGL.vertexBatchPreTransformed(pos, uv, NUM_VERTICES); @@ -795,30 +798,32 @@ void gl_draw_scaled_rotated_image(S32 x, S32 y, S32 width, S32 height, F32 degre gGL.begin(LLRender::TRIANGLES); { - LLVector3 v1 = LLVector3(offset_x, offset_y, 0.f) * quat; - LLVector3 v2 = LLVector3(-offset_x, offset_y, 0.f) * quat; - LLVector3 v3 = LLVector3(-offset_x, -offset_y, 0.f) * quat; - LLVector3 v4 = LLVector3(offset_x, -offset_y, 0.f) * quat; + LLVector3 v; + v = LLVector3(offset_x, offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mRight, uv_rect.mTop); - gGL.vertex2f(v1.mV[0], v1.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1] ); + v = LLVector3(-offset_x, offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mLeft, uv_rect.mTop); - gGL.vertex2f(v2.mV[0], v2.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1] ); + v = LLVector3(-offset_x, -offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mLeft, uv_rect.mBottom); - gGL.vertex2f(v3.mV[0], v3.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1] ); + v = LLVector3(offset_x, offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mRight, uv_rect.mTop); - gGL.vertex2f(v1.mV[0], v1.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1]); + v = LLVector3(-offset_x, -offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mLeft, uv_rect.mBottom); - gGL.vertex2f(v3.mV[0], v3.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1]); + v = LLVector3(offset_x, -offset_y, 0.f) * quat; gGL.texCoord2f(uv_rect.mRight, uv_rect.mBottom); - gGL.vertex2f(v4.mV[0], v4.mV[1] ); + gGL.vertex2f(v.mV[0], v.mV[1] ); } - gGL.end(); gGL.popUIMatrix(); } @@ -838,7 +843,7 @@ 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) @@ -1061,7 +1066,7 @@ void gl_washer_segment_2d(F32 outer_radius, F32 inner_radius, F32 start_radians, void gl_rect_2d_simple_tex( S32 width, S32 height ) { gGL.begin( LLRender::TRIANGLES ); - { + gGL.texCoord2f(1.f, 1.f); gGL.vertex2i(width, height); @@ -1079,14 +1084,13 @@ void gl_rect_2d_simple_tex( S32 width, S32 height ) gGL.texCoord2f(1.f, 0.f); gGL.vertex2i(width, 0); - } + gGL.end(); } void gl_rect_2d_simple( S32 width, S32 height ) { gGL.begin( LLRender::TRIANGLES ); - { gGL.vertex2i(width, height); gGL.vertex2i(0, height); gGL.vertex2i(0, 0); @@ -1094,7 +1098,6 @@ void gl_rect_2d_simple( S32 width, S32 height ) gGL.vertex2i(width, height); gGL.vertex2i(0, 0); gGL.vertex2i(width, 0); - } gGL.end(); } @@ -1563,7 +1566,7 @@ void gl_segmented_rect_3d_tex(const LLRectf& clip_rect, const LLRectf& center_uv { LL_PROFILE_ZONE_SCOPED_CATEGORY_UI; - gGL.begin(LLRender::TRIANGLES); + gGL.begin(LLRender::TRIANGLES); { // draw bottom left gGL.texCoord2f(clip_rect.mLeft, clip_rect.mBottom); @@ -1671,7 +1674,7 @@ void gl_segmented_rect_3d_tex(const LLRectf& clip_rect, const LLRectf& center_uv gGL.vertex3fv((width_vec + center_draw_rect.mTop * height_vec).mV); gGL.texCoord2f(center_uv_rect.mRight, center_uv_rect.mBottom); - gGL.vertex3fv((center_draw_rect.mRight * width_vec + center_draw_rect.mBottom * height_vec).mV); + gGL.vertex3fv((center_draw_rect.mRight* width_vec + center_draw_rect.mBottom * height_vec).mV); gGL.texCoord2f(clip_rect.mRight, center_uv_rect.mTop); gGL.vertex3fv((width_vec + center_draw_rect.mTop * height_vec).mV); @@ -1712,7 +1715,7 @@ void gl_segmented_rect_3d_tex(const LLRectf& clip_rect, const LLRectf& center_uv gGL.vertex3fv((center_draw_rect.mLeft * width_vec + center_draw_rect.mTop * height_vec).mV); gGL.texCoord2f(center_uv_rect.mRight, clip_rect.mTop); - gGL.vertex3fv((center_draw_rect.mRight * width_vec + height_vec).mV); + gGL.vertex3fv((center_draw_rect.mRight* width_vec + height_vec).mV); gGL.texCoord2f(center_uv_rect.mLeft, clip_rect.mTop); gGL.vertex3fv((center_draw_rect.mLeft * width_vec + height_vec).mV); @@ -1728,7 +1731,7 @@ void gl_segmented_rect_3d_tex(const LLRectf& clip_rect, const LLRectf& center_uv gGL.vertex3fv((width_vec + height_vec).mV); gGL.texCoord2f(center_uv_rect.mRight, center_uv_rect.mTop); - gGL.vertex3fv((center_draw_rect.mRight * width_vec + center_draw_rect.mTop * height_vec).mV); + gGL.vertex3fv((center_draw_rect.mRight* width_vec + center_draw_rect.mTop * height_vec).mV); gGL.texCoord2f(clip_rect.mRight, clip_rect.mTop); gGL.vertex3fv((width_vec + height_vec).mV); 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/llrendersphere.cpp b/indra/llrender/llrendersphere.cpp index 9570180554..cd8ef7d68e 100644 --- a/indra/llrender/llrendersphere.cpp +++ b/indra/llrender/llrendersphere.cpp @@ -34,6 +34,8 @@ #include "llerror.h" #include "llglheaders.h" +#include "llvertexbuffer.h" +#include "llglslshader.h" LLRenderSphere gSphere; @@ -53,12 +55,20 @@ inline LLVector3 polar_to_cart(F32 latitude, F32 longitude) void LLRenderSphere::renderGGL() { + LL_PROFILE_ZONE_SCOPED; S32 const LATITUDE_SLICES = 20; S32 const LONGITUDE_SLICES = 30; - if (mSpherePoints.empty()) + if (mVertexBuffer.isNull()) { mSpherePoints.resize(LATITUDE_SLICES + 1); + mVertexBuffer = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX); + + mVertexBuffer->allocateBuffer((U32)(LATITUDE_SLICES + 1) * (LONGITUDE_SLICES + 1), LATITUDE_SLICES * LONGITUDE_SLICES * 6); + + LLStrider<LLVector3> v; + mVertexBuffer->getVertexStrider(v); + for (S32 lat_i = 0; lat_i < LATITUDE_SLICES + 1; lat_i++) { mSpherePoints[lat_i].resize(LONGITUDE_SLICES + 1); @@ -68,24 +78,52 @@ void LLRenderSphere::renderGGL() F32 lon = (F32)lon_i / LONGITUDE_SLICES; mSpherePoints[lat_i][lon_i] = polar_to_cart(lat, lon); + v[lat_i * (LONGITUDE_SLICES + 1) + lon_i] = mSpherePoints[lat_i][lon_i]; } } + + LLStrider<U16> i; + mVertexBuffer->getIndexStrider(i); + + for (S32 lat_i = 0; lat_i < LATITUDE_SLICES; lat_i++) + { + for (S32 lon_i = 0; lon_i < LONGITUDE_SLICES; lon_i++) + { + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 0] = lat_i * (LONGITUDE_SLICES + 1) + lon_i; + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 1] = lat_i * (LONGITUDE_SLICES + 1) + lon_i + 1; + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 2] = (lat_i + 1) * (LONGITUDE_SLICES + 1) + lon_i; + + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 3] = (lat_i + 1) * (LONGITUDE_SLICES + 1) + lon_i; + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 4] = lat_i * (LONGITUDE_SLICES + 1) + lon_i + 1; + i[(lat_i * LONGITUDE_SLICES + lon_i) * 6 + 5] = (lat_i + 1) * (LONGITUDE_SLICES + 1) + lon_i + 1; + } + } + + mVertexBuffer->unmapBuffer(); } - gGL.begin(LLRender::TRIANGLES); - for (S32 lat_i = 0; lat_i < LATITUDE_SLICES; lat_i++) - { - for (S32 lon_i = 0; lon_i < LONGITUDE_SLICES; lon_i++) + if (LLGLSLShader::sCurBoundShaderPtr->mAttributeMask == LLVertexBuffer::MAP_VERTEX) + { // shader expects only vertex positions in vertex buffer, use fast path + mVertexBuffer->setBuffer(); + mVertexBuffer->drawRange(LLRender::TRIANGLES, 0, mVertexBuffer->getNumVerts(), mVertexBuffer->getNumIndices(), 0); + } + else + { //shader wants colors in the vertex stream, use slow path + gGL.begin(LLRender::TRIANGLES); + for (S32 lat_i = 0; lat_i < LATITUDE_SLICES; lat_i++) { - gGL.vertex3fv(mSpherePoints[lat_i][lon_i].mV); - gGL.vertex3fv(mSpherePoints[lat_i][lon_i+1].mV); - gGL.vertex3fv(mSpherePoints[lat_i+1][lon_i].mV); + for (S32 lon_i = 0; lon_i < LONGITUDE_SLICES; lon_i++) + { + gGL.vertex3fv(mSpherePoints[lat_i][lon_i].mV); + gGL.vertex3fv(mSpherePoints[lat_i][lon_i + 1].mV); + gGL.vertex3fv(mSpherePoints[lat_i + 1][lon_i].mV); - gGL.vertex3fv(mSpherePoints[lat_i+1][lon_i].mV); - gGL.vertex3fv(mSpherePoints[lat_i][lon_i+1].mV); - gGL.vertex3fv(mSpherePoints[lat_i+1][lon_i+1].mV); + gGL.vertex3fv(mSpherePoints[lat_i + 1][lon_i].mV); + gGL.vertex3fv(mSpherePoints[lat_i][lon_i + 1].mV); + gGL.vertex3fv(mSpherePoints[lat_i + 1][lon_i + 1].mV); + } } + gGL.end(); } - gGL.end(); } diff --git a/indra/llrender/llrendersphere.h b/indra/llrender/llrendersphere.h index e2e886fa06..5b6eabecb8 100644 --- a/indra/llrender/llrendersphere.h +++ b/indra/llrender/llrendersphere.h @@ -45,6 +45,7 @@ public: private: std::vector< std::vector<LLVector3> > mSpherePoints; + LLPointer<LLVertexBuffer> mVertexBuffer; }; extern LLRenderSphere gSphere; diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp index dcc81a8874..0e4aa2ee7a 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) @@ -255,12 +255,14 @@ bool LLRenderTarget::addColorAttachment(U32 color_fmt) stop_glerror(); } +#if GL_VERSION_3_1 if (mUsage != LLTexUnit::TT_RECT_TEXTURE) { gGL.getTexUnit(0)->setTextureAddressMode(LLTexUnit::TAM_MIRROR); stop_glerror(); } else +#endif { // ATI doesn't support mirrored repeat for rectangular textures. gGL.getTexUnit(0)->setTextureAddressMode(LLTexUnit::TAM_CLAMP); @@ -426,15 +428,18 @@ 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 - GLenum drawbuffers[] = {GL_NONE}; - glDrawBuffers(0, drawbuffers); + GLenum buffers[] = {GL_NONE}; + glDrawBuffers(0, buffers); glReadBuffer(GL_NONE); } - + else + { + glDrawBuffers(static_cast<GLsizei>(mTex.size()), drawbuffers); + glReadBuffer(GL_COLOR_ATTACHMENT0); + } check_framebuffer_status(); glViewport(0, 0, mResX, mResY); @@ -473,12 +478,10 @@ 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]; @@ -511,7 +514,6 @@ 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() @@ -522,7 +524,8 @@ 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); @@ -543,6 +546,9 @@ void LLRenderTarget::flush() glViewport(gGLViewport[0], gGLViewport[1], gGLViewport[2], gGLViewport[3]); sCurResX = gGLViewport[2]; sCurResY = gGLViewport[3]; + glReadBuffer(GL_BACK); + GLenum drawbuffers[] = {GL_BACK}; + glDrawBuffers(1, drawbuffers); } } 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..4807c12226 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() { @@ -80,7 +79,7 @@ bool LLShaderMgr::attachShaderFeatures(LLGLSLShader * shader) ////////////////////////////////////// // NOTE order of shader object attaching is VERY IMPORTANT!!! - if (features->calculatesAtmospherics) + if (features->calculatesAtmospherics || features->hasGamma || features->isDeferred) { if (!shader->attachVertexObject("windlight/atmosphericsVarsV.glsl")) { @@ -224,6 +223,14 @@ bool LLShaderMgr::attachShaderFeatures(LLGLSLShader * shader) } } + if (features->hasFullGBuffer) + { + if (!shader->attachFragmentObject("deferred/gbufferUtil.glsl")) + { + return false; + } + } + if (features->hasScreenSpaceReflections || features->hasReflectionProbes) { if (!shader->attachFragmentObject("deferred/screenSpaceReflUtil.glsl")) @@ -284,6 +291,14 @@ bool LLShaderMgr::attachShaderFeatures(LLGLSLShader * shader) } } + if (features->hasTonemap) + { + if (!shader->attachFragmentObject("deferred/tonemapUtilF.glsl")) + { + return false; + } + } + // NOTE order of shader object attaching is VERY IMPORTANT!!! if (features->hasAtmospherics) { @@ -459,6 +474,7 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev if (filename.empty()) { + LL_WARNS("ShaderLoading") << "tried loading empty filename" << LL_ENDL; return 0; } @@ -560,17 +576,11 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev } else if (major_version == 3) { - if (minor_version < 10) + if (minor_version <= 29) { - shader_code_text[shader_code_count++] = strdup("#version 300\n"); - } - else if (minor_version <= 19) - { - shader_code_text[shader_code_count++] = strdup("#version 310\n"); - } - else if (minor_version <= 29) - { - shader_code_text[shader_code_count++] = strdup("#version 320\n"); + // OpenGL 3.2 had GLSL version 1.50. anything after that the version numbers match. + // https://www.khronos.org/opengl/wiki/Core_Language_(GLSL)#OpenGL_and_GLSL_versions + shader_code_text[shader_code_count++] = strdup("#version 150\n"); } else { @@ -579,7 +589,8 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev } else { - if (type == GL_GEOMETRY_SHADER) + // OpenGL 3.2 had GLSL version 1.50. anything after that the version numbers match. + if (type == GL_GEOMETRY_SHADER || minor_version >= 50) { //set version to 1.50 shader_code_text[shader_code_count++] = strdup("#version 150\n"); @@ -596,13 +607,15 @@ 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 HERO_PROBES 1\n"); + extra_code_text[extra_code_count++] = strdup("#define FRAGMENT_SHADER 1\n"); + } + else + { + extra_code_text[extra_code_count++] = strdup("#define VERTEX_SHADER 1\n"); } // Use alpha float to store bit flags @@ -611,7 +624,7 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev extra_code_text[extra_code_count++] = strdup("#define GBUFFER_FLAG_HAS_ATMOS 0.34\n"); // bit 0 extra_code_text[extra_code_count++] = strdup("#define GBUFFER_FLAG_HAS_PBR 0.67\n"); // bit 1 extra_code_text[extra_code_count++] = strdup("#define GBUFFER_FLAG_HAS_HDRI 1.0\n"); // bit 2 - extra_code_text[extra_code_count++] = strdup("#define GET_GBUFFER_FLAG(flag) (abs(norm.w-flag)< 0.1)\n"); + extra_code_text[extra_code_count++] = strdup("#define GET_GBUFFER_FLAG(data, flag) (abs(data-flag)< 0.1)\n"); if (defines) { @@ -722,6 +735,9 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev } } + // Master definition can be found in deferredUtil.glsl + extra_code_text[extra_code_count++] = strdup("struct GBufferInfo { vec4 albedo; vec4 specular; vec3 normal; vec4 emissive; float gbufferFlag; float envIntensity; };\n"); + //copy file into memory enum { flag_write_to_out_of_extra_block_area = 0x01 @@ -916,6 +932,8 @@ GLuint LLShaderMgr::loadShaderFile(const std::string& filename, S32 & shader_lev } LL_WARNS("ShaderLoading") << "Failed to load " << filename << LL_ENDL; } + + LL_DEBUGS("ShaderLoading") << "loadShaderFile() completed, ret: " << U32(ret) << LL_ENDL; return ret; } @@ -1003,7 +1021,7 @@ 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 +1052,7 @@ 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 +1111,7 @@ 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 +1149,7 @@ 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; @@ -1254,6 +1272,7 @@ void LLShaderMgr::initAttribsAndUniforms() mReservedUniforms.push_back("sky_hdr_scale"); mReservedUniforms.push_back("sky_sunlight_scale"); mReservedUniforms.push_back("sky_ambient_scale"); + mReservedUniforms.push_back("classic_mode"); mReservedUniforms.push_back("blue_horizon"); mReservedUniforms.push_back("blue_density"); mReservedUniforms.push_back("haze_horizon"); @@ -1381,6 +1400,7 @@ void LLShaderMgr::initAttribsAndUniforms() mReservedUniforms.push_back("screenTex"); mReservedUniforms.push_back("screenDepth"); mReservedUniforms.push_back("refTex"); + mReservedUniforms.push_back("exclusionTex"); mReservedUniforms.push_back("eyeVec"); mReservedUniforms.push_back("time"); mReservedUniforms.push_back("waveDir1"); @@ -1415,6 +1435,7 @@ 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 +1460,8 @@ 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 +1496,11 @@ 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..46788841a5 100644 --- a/indra/llrender/llshadermgr.h +++ b/indra/llrender/llshadermgr.h @@ -36,6 +36,8 @@ public: LLShaderMgr(); virtual ~LLShaderMgr(); + // Note: although you can use statically hashed strings to just bind a random uniform, it's generally preferably that you use this. + // Always document what the actual shader uniform is next to the shader uniform in this struct. // clang-format off typedef enum { // Shader uniform name, set in LLShaderMgr::initAttribsAndUniforms() @@ -121,6 +123,7 @@ public: SKY_HDR_SCALE, // "sky_hdr_scale" SKY_SUNLIGHT_SCALE, // "sky_sunlight_scale" SKY_AMBIENT_SCALE, // "sky_ambient_scale" + CLASSIC_MODE, // "classic_mode" BLUE_HORIZON, // "blue_horizon" BLUE_DENSITY, // "blue_density" HAZE_HORIZON, // "haze_horizon" @@ -233,6 +236,7 @@ public: WATER_SCREENTEX, // "screenTex" WATER_SCREENDEPTH, // "screenDepth" WATER_REFTEX, // "refTex" + WATER_EXCLUSIONTEX, // "exclusionTex" WATER_EYEVEC, // "eyeVec" WATER_TIME, // "time" WATER_WAVE_DIR1, // "waveDir1" @@ -267,6 +271,7 @@ 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 +296,8 @@ 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 +336,11 @@ 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 +390,6 @@ 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 6cded7f67a..ac6db0b34f 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -36,10 +36,7 @@ #include "llshadermgr.h" #include "llglslshader.h" #include "llmemory.h" - -#include "llcontrol.h" - -extern LLControlGroup gSavedSettings; +#include <glm/gtc/type_ptr.hpp> //Next Highest Power Of Two //helper function, returns first number > v that is a power of 2, or v if v is already a power of 2 @@ -275,11 +272,13 @@ static GLuint gen_buffer() { LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("gen buffer"); sIndex = pool_size; +#if !LL_DARWIN if (!gGLManager.mIsAMD) { glGenBuffers(pool_size, sNamePool); } else +#endif { // work around for AMD driver bug for (U32 i = 0; i < pool_size; ++i) { @@ -292,22 +291,58 @@ static GLuint gen_buffer() return ret; } -#define ANALYZE_VBO_POOL 0 +static void delete_buffers(S32 count, GLuint* buffers) +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + // wait a few frames before actually deleting the buffers to avoid + // synchronization issues with the GPU + static std::vector<GLuint> sFreeList[4]; -#if 0 // LL_DARWIN + if (gGLManager.mInited) + { + U32 idx = LLImageGL::sFrameCount % 4; + + for (S32 i = 0; i < count; ++i) + { + sFreeList[idx].push_back(buffers[i]); + } + + idx = (LLImageGL::sFrameCount + 3) % 4; -// experimental -- disable VBO pooling on OS X and use glMapBuffer + if (!sFreeList[idx].empty()) + { + glDeleteBuffers((GLsizei)sFreeList[idx].size(), sFreeList[idx].data()); + sFreeList[idx].resize(0); + } + } +} + + +#define ANALYZE_VBO_POOL 0 + +// VBO Pool interface class LLVBOPool { + public: + virtual ~LLVBOPool() = default; + virtual void allocate(GLenum type, U32 size, GLuint& name, U8*& data) = 0; + virtual void free(GLenum type, U32 size, GLuint name, U8* data) = 0; + virtual U64 getVramBytesUsed() = 0; +}; + +// VBO Pool for Apple GPUs (as in M1/M2 etc, not Intel macs) +// Effectively disables VBO pooling +class LLAppleVBOPool final: public LLVBOPool +{ public: U64 mAllocated = 0; - U64 getVramBytesUsed() + U64 getVramBytesUsed() override { return mAllocated; } - void allocate(GLenum type, U32 size, GLuint& name, U8*& data) + void allocate(GLenum type, U32 size, GLuint& name, U8*& data) override { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; STOP_GLERROR; @@ -327,7 +362,7 @@ public: } } - void free(GLenum type, U32 size, GLuint name, U8* data) + void free(GLenum type, U32 size, GLuint name, U8* data) override { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER); @@ -342,21 +377,17 @@ public: STOP_GLERROR; if (name) { - glDeleteBuffers(1, &name); + delete_buffers(1, &name); } STOP_GLERROR; } }; -#else - -class LLVBOPool +// VBO Pool for GPUs that benefit from VBO pooling +class LLDefaultVBOPool final : public LLVBOPool { public: typedef std::chrono::steady_clock::time_point Time; - - U32 mMappingMode; - struct Entry { U8* mData; @@ -364,16 +395,8 @@ public: Time mAge; }; - /* - LLVBOPool() - { - - } - */ - - ~LLVBOPool() + ~LLDefaultVBOPool() override { - if(mMappingMode == 3) return; clear(); } @@ -390,10 +413,9 @@ public: U32 mMisses = 0; U32 mHits = 0; - U64 getVramBytesUsed() + U64 getVramBytesUsed() override { - if(mMappingMode == 3) return mAllocated; - else return mAllocated + mReserved; + return mAllocated + mReserved; } // increase the size to some common value (e.g. a power of two) to increase hit rate @@ -407,7 +429,7 @@ public: size += block_size - (size % block_size); } - void allocate(GLenum type, U32 size, GLuint& name, U8*& data) + void allocate(GLenum type, U32 size, GLuint& name, U8*& data) override { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER); @@ -415,20 +437,6 @@ public: llassert(data == nullptr); // non null data indicates a buffer that wasn't freed llassert(size >= 2); // any buffer size smaller than a single index is nonsensical - if(mMappingMode == 3) - { - mAllocated += size; - - { //allocate a new buffer - LL_PROFILE_GPU_ZONE("vbo alloc"); - // ON OS X, we don't allocate a VBO until the last possible moment - // in unmapBuffer - data = (U8*) ll_aligned_malloc_16(size); - //STOP_GLERROR; - } - return; - } - mDistributed += size; adjustSize(size); mAllocated += size; @@ -477,30 +485,11 @@ public: clean(); } - void free(GLenum type, U32 size, GLuint name, U8* data) + void free(GLenum type, U32 size, GLuint name, U8* data) override { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER); llassert(size >= 2); - - if(mMappingMode == 3) - { - if (data) - { - ll_aligned_free_16(data); - } - - mAllocated -= size; - //STOP_GLERROR; - if (name) - { - glDeleteBuffers(1, &name); - } - //STOP_GLERROR; - - return; - } - llassert(name != 0); llassert(data != nullptr); @@ -559,7 +548,7 @@ public: LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vbo cache timeout"); auto& entry = entries.back(); ll_aligned_free_16(entry.mData); - glDeleteBuffers(1, &entry.mGLName); + delete_buffers(1, &entry.mGLName); llassert(mReserved >= iter->first); mReserved -= iter->first; entries.pop_back(); @@ -595,7 +584,7 @@ public: for (auto& entry : entries.second) { ll_aligned_free_16(entry.mData); - glDeleteBuffers(1, &entry.mGLName); + delete_buffers(1, &entry.mGLName); } } @@ -604,7 +593,7 @@ public: for (auto& entry : entries.second) { ll_aligned_free_16(entry.mData); - glDeleteBuffers(1, &entry.mGLName); + delete_buffers(1, &entry.mGLName); } } @@ -614,10 +603,71 @@ public: mVBOPool.clear(); } }; -#endif static LLVBOPool* sVBOPool = nullptr; +void LLVertexBufferData::drawWithMatrix() +{ + 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(glm::value_ptr(mModelView)); + gGL.matrixMode(LLRender::MM_PROJECTION); + gGL.pushMatrix(); + gGL.loadMatrix(glm::value_ptr(mProjection)); + gGL.matrixMode(LLRender::MM_TEXTURE0); + gGL.pushMatrix(); + gGL.loadMatrix(glm::value_ptr(mTexture0)); + + mVB->setBuffer(); + mVB->drawArrays(mMode, 0, mCount); + + gGL.popMatrix(); + gGL.matrixMode(LLRender::MM_PROJECTION); + gGL.popMatrix(); + gGL.matrixMode(LLRender::MM_MODELVIEW); + gGL.popMatrix(); +} + +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); + } + + mVB->setBuffer(); + mVB->drawArrays(mMode, 0, mCount); +} + +//============================================================================ + //static U64 LLVertexBuffer::getBytesAllocated() { @@ -632,7 +682,6 @@ U32 LLVertexBuffer::sGLRenderIndices = 0; U32 LLVertexBuffer::sLastMask = 0; U32 LLVertexBuffer::sVertexCount = 0; -U32 LLVertexBuffer::sMappingMode = 0; //NOTE: each component must be AT LEAST 4 bytes in size to avoid a performance penalty on AMD hardware const U32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] = @@ -681,7 +730,6 @@ const U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] = GL_POINTS, GL_LINES, GL_LINE_STRIP, - GL_QUADS, GL_LINE_LOOP, }; @@ -843,6 +891,18 @@ 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)); @@ -884,10 +944,17 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const void LLVertexBuffer::initClass(LLWindow* window) { llassert(sVBOPool == nullptr); - sVBOPool = new LLVBOPool(); - sVBOPool->mMappingMode = sMappingMode; - //LL_INFOS() << "sVBOPool intialized with mapping mode: " << sMappingMode << LL_ENDL; + if (gGLManager.mIsApple) + { + LL_INFOS() << "VBO Pooling Disabled" << LL_ENDL; + sVBOPool = new LLAppleVBOPool(); + } + else + { + LL_INFOS() << "VBO Pooling Enabled" << LL_ENDL; + sVBOPool = new LLDefaultVBOPool(); + } #if ENABLE_GL_WORK_QUEUE sQueue = new GLWorkQueue(); @@ -946,6 +1013,24 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask) } } +// list of mapped buffers +// NOTE: must not be LLPointer<LLVertexBuffer> to avoid breaking non-ref-counted LLVertexBuffer instances +static std::vector<LLVertexBuffer*> sMappedBuffers; + +//static +void LLVertexBuffer::flushBuffers() +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + // must only be called from main thread + for (auto& buffer : sMappedBuffers) + { + buffer->_unmapBuffer(); + buffer->mMapped = false; + } + + sMappedBuffers.resize(0); +} + //static U32 LLVertexBuffer::calcOffsets(const U32& typemask, U32* offsets, U32 num_vertices) { @@ -989,6 +1074,12 @@ U32 LLVertexBuffer::calcVertexSize(const U32& typemask) //virtual LLVertexBuffer::~LLVertexBuffer() { + if (mMapped) + { // is on the mapped buffer list but doesn't need to be flushed + mMapped = false; + unmapBuffer(); + } + destroyGLBuffer(); destroyGLIndices(); @@ -1190,13 +1281,14 @@ bool expand_region(LLVertexBuffer::MappedRegion& region, U32 start, U32 end) U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 index, S32 count) { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + _mapBuffer(); if (count == -1) { count = mNumVerts - index; } - if(sMappingMode != 3) + if (!gGLManager.mIsApple) { U32 start = mOffsets[type] + sTypeSize[type] * index; U32 end = start + sTypeSize[type] * count-1; @@ -1219,7 +1311,6 @@ U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 inde mMappedVertexRegions.push_back({ start, end }); } } - return mMappedData+mOffsets[type]+sTypeSize[type]*index; } @@ -1227,13 +1318,14 @@ U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 inde U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count) { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + _mapBuffer(); if (count == -1) { count = mNumIndices-index; } - if(sMappingMode != 3) + if (!gGLManager.mIsApple) { U32 start = sizeof(U16) * index; U32 end = start + sizeof(U16) * count-1; @@ -1268,57 +1360,64 @@ U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count) // dst -- mMappedData or mMappedIndexData void LLVertexBuffer::flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst) { - if(sMappingMode == 2) - { - //LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb glMapBufferRange"); - if (end == 0) return; - U32 buffer_size = end-start+1; - U8 * mptr = (U8*) glMapBufferRange( target, start, end-start+1, GL_MAP_WRITE_BIT); - - if (mptr) - { - std::memcpy(mptr, (U8*) data, buffer_size); - if(!glUnmapBuffer(target)) LL_WARNS() << "glUnmapBuffer() failed" << LL_ENDL; - } - else LL_WARNS() << "glMapBufferRange() returned NULL" << LL_ENDL; - return; - } - - if(sMappingMode == 3) + if (gGLManager.mIsApple) { + // on OS X, flush_vbo doesn't actually write to the GL buffer, so be sure to call + // _mapBuffer to tag the buffer for flushing to GL + _mapBuffer(); LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb memcpy"); - //STOP_GLERROR; + STOP_GLERROR; // copy into mapped buffer memcpy(dst+start, data, end-start+1); - return; } - - llassert(target == GL_ARRAY_BUFFER ? sGLRenderBuffer == mGLBuffer : sGLRenderIndices == mGLIndices); - - // skip mapped data and stream to GPU via glBufferSubData - if (end != 0) + else { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData"); - LL_PROFILE_ZONE_NUM(start); - LL_PROFILE_ZONE_NUM(end); - LL_PROFILE_ZONE_NUM(end-start); - - constexpr U32 block_size = 8192; + llassert(target == GL_ARRAY_BUFFER ? sGLRenderBuffer == mGLBuffer : sGLRenderIndices == mGLIndices); - for (U32 i = start; i <= end; i += block_size) + // skip mapped data and stream to GPU via glBufferSubData + if (end != 0) { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData block"); - //LL_PROFILE_GPU_ZONE("glBufferSubData"); - U32 tend = llmin(i + block_size, end); - U32 size = tend - i + 1; - glBufferSubData(target, i, size, (U8*) data + (i-start)); + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData"); + LL_PROFILE_ZONE_NUM(start); + LL_PROFILE_ZONE_NUM(end); + LL_PROFILE_ZONE_NUM(end-start); + + constexpr U32 block_size = 65536; + + for (U32 i = start; i <= end; i += block_size) + { + //LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData block"); + //LL_PROFILE_GPU_ZONE("glBufferSubData"); + U32 tend = llmin(i + block_size, end); + U32 size = tend - i + 1; + glBufferSubData(target, i, size, (U8*) data + (i-start)); + } } } } void LLVertexBuffer::unmapBuffer() { + flushBuffers(); +} + +void LLVertexBuffer::_mapBuffer() +{ + if (!mMapped) + { + mMapped = true; + sMappedBuffers.push_back(this); + } +} + +void LLVertexBuffer::_unmapBuffer() +{ STOP_GLERROR; + if (!mMapped) + { + return; + } + struct SortMappedRegion { bool operator()(const MappedRegion& lhs, const MappedRegion& rhs) @@ -1327,114 +1426,115 @@ void LLVertexBuffer::unmapBuffer() } }; - if(sMappingMode == 3) + if (gGLManager.mIsApple) { - //STOP_GLERROR; + STOP_GLERROR; if (mMappedData) { if (mGLBuffer) { - glDeleteBuffers(1, &mGLBuffer); + delete_buffers(1, &mGLBuffer); } mGLBuffer = gen_buffer(); glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); sGLRenderBuffer = mGLBuffer; - glBufferData(GL_ARRAY_BUFFER, mSize, mMappedData, GL_DYNAMIC_DRAW); + glBufferData(GL_ARRAY_BUFFER, mSize, mMappedData, GL_STATIC_DRAW); } else if (mGLBuffer != sGLRenderBuffer) { glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); sGLRenderBuffer = mGLBuffer; } - //STOP_GLERROR; + STOP_GLERROR; if (mMappedIndexData) { if (mGLIndices) { - glDeleteBuffers(1, &mGLIndices); + delete_buffers(1, &mGLIndices); } mGLIndices = gen_buffer(); glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); sGLRenderIndices = mGLIndices; - glBufferData(GL_ELEMENT_ARRAY_BUFFER, mIndicesSize, mMappedIndexData, GL_DYNAMIC_DRAW); + glBufferData(GL_ELEMENT_ARRAY_BUFFER, mIndicesSize, mMappedIndexData, GL_STATIC_DRAW); } else if (mGLIndices != sGLRenderIndices) { glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); sGLRenderIndices = mGLIndices; } - //STOP_GLERROR; - return; + STOP_GLERROR; } - - if (!mMappedVertexRegions.empty()) + else { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - vertex"); - - if (sGLRenderBuffer != mGLBuffer) + if (!mMappedVertexRegions.empty()) { - glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); - sGLRenderBuffer = mGLBuffer; - } - - U32 start = 0; - U32 end = 0; - - std::sort(mMappedVertexRegions.begin(), mMappedVertexRegions.end(), SortMappedRegion()); + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - vertex"); - for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) - { - const MappedRegion& region = mMappedVertexRegions[i]; - if (region.mStart == end + 1) - { - end = region.mEnd; - } - else + if (sGLRenderBuffer != mGLBuffer) { - flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData); - start = region.mStart; - end = region.mEnd; + glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); + sGLRenderBuffer = mGLBuffer; } - } - flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData); - mMappedVertexRegions.clear(); - } + U32 start = 0; + U32 end = 0; - if (!mMappedIndexRegions.empty()) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - index"); + std::sort(mMappedVertexRegions.begin(), mMappedVertexRegions.end(), SortMappedRegion()); - if (mGLIndices != sGLRenderIndices) - { - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); - sGLRenderIndices = mGLIndices; - } - U32 start = 0; - U32 end = 0; + for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) + { + const MappedRegion& region = mMappedVertexRegions[i]; + if (region.mStart == end + 1) + { + end = region.mEnd; + } + else + { + flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData); + start = region.mStart; + end = region.mEnd; + } + } - std::sort(mMappedIndexRegions.begin(), mMappedIndexRegions.end(), SortMappedRegion()); + flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData); + mMappedVertexRegions.clear(); + } - for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) + if (!mMappedIndexRegions.empty()) { - const MappedRegion& region = mMappedIndexRegions[i]; - if (region.mStart == end + 1) + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - index"); + + if (mGLIndices != sGLRenderIndices) { - end = region.mEnd; + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); + sGLRenderIndices = mGLIndices; } - else + U32 start = 0; + U32 end = 0; + + std::sort(mMappedIndexRegions.begin(), mMappedIndexRegions.end(), SortMappedRegion()); + + for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) { - flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start, mMappedIndexData); - start = region.mStart; - end = region.mEnd; + const MappedRegion& region = mMappedIndexRegions[i]; + if (region.mStart == end + 1) + { + end = region.mEnd; + } + else + { + flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start, mMappedIndexData); + start = region.mStart; + end = region.mEnd; + } } - } - flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start, mMappedIndexData); - mMappedIndexRegions.clear(); + flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start, mMappedIndexData); + mMappedIndexRegions.clear(); + } } } @@ -1556,12 +1656,12 @@ bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, U32 in // Set for rendering void LLVertexBuffer::setBuffer() { - if(sMappingMode == 3) + STOP_GLERROR; + + if (mMapped) { - if (!mGLBuffer) - { - return; - } + LL_WARNS_ONCE() << "Missing call to unmapBuffer or flushBuffers" << LL_ENDL; + _unmapBuffer(); } // no data may be pending @@ -1607,55 +1707,58 @@ void LLVertexBuffer::setupVertexBuffer() STOP_GLERROR; U8* base = nullptr; + AttributeType loc; + void* ptr = nullptr; + U32 data_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; if (data_mask & MAP_NORMAL) { - AttributeType loc = TYPE_NORMAL; - void* ptr = (void*)(base + mOffsets[TYPE_NORMAL]); + loc = TYPE_NORMAL; + ptr = (void*)(base + mOffsets[TYPE_NORMAL]); glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_NORMAL], ptr); } if (data_mask & MAP_TEXCOORD3) { - AttributeType loc = TYPE_TEXCOORD3; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]); + loc = TYPE_TEXCOORD3; + ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD3], ptr); } if (data_mask & MAP_TEXCOORD2) { - AttributeType loc = TYPE_TEXCOORD2; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]); + loc = TYPE_TEXCOORD2; + ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD2], ptr); } if (data_mask & MAP_TEXCOORD1) { - AttributeType loc = TYPE_TEXCOORD1; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]); + loc = TYPE_TEXCOORD1; + ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr); } if (data_mask & MAP_TANGENT) { - AttributeType loc = TYPE_TANGENT; - void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]); + loc = TYPE_TANGENT; + ptr = (void*)(base + mOffsets[TYPE_TANGENT]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr); } if (data_mask & MAP_TEXCOORD0) { - AttributeType loc = TYPE_TEXCOORD0; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]); + loc = TYPE_TEXCOORD0; + ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD0], ptr); } if (data_mask & MAP_COLOR) { - AttributeType loc = TYPE_COLOR; + loc = TYPE_COLOR; //bind emissive instead of color pointer if emissive is present - void* ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]); + ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]); glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_COLOR], ptr); } if (data_mask & MAP_EMISSIVE) { - AttributeType loc = TYPE_EMISSIVE; - void* ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]); + loc = TYPE_EMISSIVE; + ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]); glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); if (!(data_mask & MAP_COLOR)) @@ -1666,38 +1769,38 @@ void LLVertexBuffer::setupVertexBuffer() } if (data_mask & MAP_WEIGHT) { - AttributeType loc = TYPE_WEIGHT; - void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT]); + loc = TYPE_WEIGHT; + ptr = (void*)(base + mOffsets[TYPE_WEIGHT]); glVertexAttribPointer(loc, 1, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT], ptr); } if (data_mask & MAP_WEIGHT4) { - AttributeType loc = TYPE_WEIGHT4; - void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]); + loc = TYPE_WEIGHT4; + ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr); } if (data_mask & MAP_JOINT) { - AttributeType loc = TYPE_JOINT; - void* ptr = (void*)(base + mOffsets[TYPE_JOINT]); + loc = TYPE_JOINT; + ptr = (void*)(base + mOffsets[TYPE_JOINT]); glVertexAttribIPointer(loc, 4, GL_UNSIGNED_SHORT, LLVertexBuffer::sTypeSize[TYPE_JOINT], ptr); } if (data_mask & MAP_CLOTHWEIGHT) { - AttributeType loc = TYPE_CLOTHWEIGHT; - void* ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]); + loc = TYPE_CLOTHWEIGHT; + ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_CLOTHWEIGHT], ptr); } if (data_mask & MAP_TEXTURE_INDEX) { - AttributeType loc = TYPE_TEXTURE_INDEX; - void* ptr = (void*)(base + mOffsets[TYPE_VERTEX] + 12); + loc = TYPE_TEXTURE_INDEX; + ptr = (void*)(base + mOffsets[TYPE_VERTEX] + 12); glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); } if (data_mask & MAP_VERTEX) { - AttributeType loc = TYPE_VERTEX; - void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]); + loc = TYPE_VERTEX; + ptr = (void*)(base + mOffsets[TYPE_VERTEX]); glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); } STOP_GLERROR; @@ -1814,7 +1917,3 @@ void LLVertexBuffer::setIndexData(const U32* data, U32 offset, U32 count) } flush_vbo(GL_ELEMENT_ARRAY_BUFFER, offset * sizeof(U32), (offset + count) * sizeof(U32) - 1, (U8*)data, mMappedIndexData); } - - - - diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index 4dd375a4af..375ad76fb8 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -38,6 +38,7 @@ #include <set> #include <vector> #include <list> +#include <glm/gtc/matrix_transform.hpp> #define LL_MAX_VERTEX_ATTRIB_LOCATION 64 @@ -53,6 +54,41 @@ //============================================================================ // base class class LLPrivateMemoryPool; +class LLVertexBuffer; + +class LLVertexBufferData +{ +public: + LLVertexBufferData() + : mVB(nullptr) + , mMode(0) + , mCount(0) + , mTexName(0) + , mProjection(glm::identity<glm::mat4>()) + , mModelView(glm::identity<glm::mat4>()) + , mTexture0(glm::identity<glm::mat4>()) + {} + LLVertexBufferData(LLVertexBuffer* buffer, U8 mode, U32 count, U32 tex_name, const glm::mat4& model_view, const glm::mat4& projection, const glm::mat4& texture0) + : mVB(buffer) + , mMode(mode) + , mCount(count) + , mTexName(tex_name) + , mProjection(model_view) + , mModelView(projection) + , mTexture0(texture0) + {} + void drawWithMatrix(); + void draw(); + LLPointer<LLVertexBuffer> mVB; + U8 mMode; + U32 mCount; + U32 mTexName; + glm::mat4 mProjection; + glm::mat4 mModelView; + glm::mat4 mTexture0; +}; +typedef std::list<LLVertexBufferData> buffer_data_list_t; + class LLVertexBuffer final : public LLRefCount { public: @@ -89,6 +125,9 @@ public: // indexed by the following enum static U32 calcOffsets(const U32& typemask, U32* offsets, U32 num_vertices); + // flush any pending mapped buffers + static void flushBuffers(); + //WARNING -- when updating these enums you MUST // 1 - update LLVertexBuffer::sTypeSize // 2 - update LLVertexBuffer::vb_type_name @@ -159,6 +198,8 @@ public: // map for data access (see also getFooStrider below) U8* mapVertexBuffer(AttributeType type, U32 index, S32 count = -1); U8* mapIndexBuffer(U32 index, S32 count = -1); + + // synonym for flushBuffers void unmapBuffer(); // set for rendering @@ -169,7 +210,7 @@ public: 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 +259,12 @@ 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 +283,7 @@ public: void setLabel(const char* label); #endif + void clone(LLVertexBuffer& target) const; protected: U32 mGLBuffer = 0; // GL VBO handle @@ -280,6 +322,13 @@ private: bool allocateBuffer(S32 nverts, S32 nindices, bool create) { return allocateBuffer(nverts, nindices); } + // actually unmap buffer + void _unmapBuffer(); + + // add to set of mapped buffers + void _mapBuffer(); + bool mMapped = false; + public: static U64 getBytesAllocated(); @@ -289,8 +338,6 @@ public: static U32 sGLRenderIndices; static U32 sLastMask; static U32 sVertexCount; - - static U32 sMappingMode; }; #ifdef LL_PROFILER_ENABLE_RENDER_DOC |