From 3a7374892945a233a9eaa3a88c94a984873373d3 Mon Sep 17 00:00:00 2001 From: Rye Mutt Date: Thu, 29 Aug 2024 15:56:28 -0400 Subject: Remove dead texunit colorspace conversion code --- indra/llrender/llrender.cpp | 18 ++---------------- 1 file changed, 2 insertions(+), 16 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 1301d325a0..3572419a7a 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -115,7 +115,7 @@ static const GLenum sGLBlendFactor[] = LLTexUnit::LLTexUnit(S32 index) : mCurrTexType(TT_NONE), - mCurrColorScale(1), mCurrAlphaScale(1), mCurrTexture(0), mTexColorSpace(TCS_LINEAR), + mCurrColorScale(1), mCurrAlphaScale(1), mCurrTexture(0), mHasMipMaps(false), mIndex(index) { @@ -145,8 +145,6 @@ void LLTexUnit::refreshState(void) { glBindTexture(GL_TEXTURE_2D, 0); } - - setTextureColorSpace(mTexColorSpace); } void LLTexUnit::activate(void) @@ -241,7 +239,6 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind) setTextureAddressMode(gl_tex->mAddressMode); setTextureFilteringOption(gl_tex->mFilterOption); } - setTextureColorSpace(mTexColorSpace); } } else @@ -318,7 +315,6 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind, S32 setTextureFilteringOption(texture->mFilterOption); stop_glerror(); } - setTextureColorSpace(mTexColorSpace); } stop_glerror(); @@ -354,7 +350,6 @@ bool LLTexUnit::bind(LLCubeMap* cubeMap) setTextureAddressMode(cubeMap->mImages[0]->mAddressMode); setTextureFilteringOption(cubeMap->mImages[0]->mFilterOption); } - setTextureColorSpace(mTexColorSpace); return true; } else @@ -403,7 +398,6 @@ bool LLTexUnit::bindManual(eTextureType type, U32 texture, bool hasMips) mCurrTexture = texture; glBindTexture(sGLTextureType[type], texture); mHasMipMaps = hasMips; - setTextureColorSpace(mTexColorSpace); } return true; } @@ -424,8 +418,6 @@ void LLTexUnit::unbind(eTextureType type) { mCurrTexture = 0; - // Always make sure our texture color space is reset to linear. SRGB sampling should be opt-in in the vast majority of cases. Also prevents color space "popping". - mTexColorSpace = TCS_LINEAR; if (type == LLTexUnit::TT_TEXTURE) { glBindTexture(sGLTextureType[type], sWhiteTexture); @@ -447,8 +439,6 @@ void LLTexUnit::unbindFast(eTextureType type) { mCurrTexture = 0; - // Always make sure our texture color space is reset to linear. SRGB sampling should be opt-in in the vast majority of cases. Also prevents color space "popping". - mTexColorSpace = TCS_LINEAR; if (type == LLTexUnit::TT_TEXTURE) { glBindTexture(sGLTextureType[type], sWhiteTexture); @@ -642,11 +632,6 @@ void LLTexUnit::debugTextureUnit(void) } } -void LLTexUnit::setTextureColorSpace(eTextureColorSpace space) -{ - mTexColorSpace = space; -} - LLLightState::LLLightState(S32 index) : mIndex(index), mEnabled(false), @@ -1634,6 +1619,7 @@ void LLRender::flush() { HBXXH64 hash; + U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; { -- cgit v1.2.3 From 958afaa7a7bfb1217bbffee3fe562f21901277ac Mon Sep 17 00:00:00 2001 From: Rye Mutt Date: Tue, 27 Aug 2024 22:47:32 -0400 Subject: Integrate SMAA and rework post process chain for better visual quality Add SMAA buffer generation passes Add quality levels for both FXAA and SMAA Separate gamma correction and tonemapping for effects that require linear-but-tonemapped inputs Move application of noise to final render pass to screen to avoid damaging other post process effects --- indra/llrender/llrender.cpp | 1 - 1 file changed, 1 deletion(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 3572419a7a..7a52f9cfb5 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1619,7 +1619,6 @@ void LLRender::flush() { HBXXH64 hash; - U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; { -- cgit v1.2.3 From 4ae1de1f8a78d795958d67afab8356f9a13f707d Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Wed, 28 Aug 2024 23:05:58 +0300 Subject: viewer#2411 LLFontGL::render optimizations --- indra/llrender/llrender.cpp | 197 ++++++++++++++++++++++++++++++++++++-------- 1 file changed, 164 insertions(+), 33 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 7a52f9cfb5..8cd2d9cc15 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1570,6 +1570,51 @@ void LLRender::end() flush(); } } + +LLVertexBuffer* LLRender::beginNoCache(const GLuint& mode, S32& count) +{ + if (mode != mMode) + { + if (mode == LLRender::QUADS) + { + mQuadCycle = 1; + } + + if (mMode == LLRender::QUADS || + mMode == LLRender::LINES || + mMode == LLRender::TRIANGLES || + mMode == LLRender::POINTS) + { + return getBuffer(count); + } + else if (mCount != 0) + { + LL_ERRS() << "gGL.begin() called redundantly." << LL_ENDL; + } + mMode = mode; + } + return nullptr; +} + +LLVertexBuffer* LLRender::endNoCache(S32& count) +{ + if (mCount == 0) + { + return nullptr; + //IMM_ERRS << "GL begin and end called with no vertices specified." << LL_ENDL; + } + + if ((mMode != LLRender::QUADS && + mMode != LLRender::LINES && + mMode != LLRender::TRIANGLES && + mMode != LLRender::POINTS) || + mCount > 2048) + { + return getBuffer(count); + } + return nullptr; +} + void LLRender::flush() { STOP_GLERROR; @@ -1664,27 +1709,7 @@ void LLRender::flush() else { LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss"); - vb = new LLVertexBuffer(attribute_mask); - vb->allocateBuffer(count, 0); - - vb->setBuffer(); - - vb->setPositionData((LLVector4a*) mVerticesp.get()); - - if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) - { - vb->setTexCoord0Data(mTexcoordsp.get()); - } - - if (attribute_mask & LLVertexBuffer::MAP_COLOR) - { - vb->setColorData(mColorsp.get()); - } - -#if LL_DARWIN - vb->unmapBuffer(); -#endif - vb->unbind(); + vb = genBuffer(attribute_mask, count); sVBCache[vhash] = { vb , std::chrono::steady_clock::now() }; @@ -1712,31 +1737,137 @@ void LLRender::flush() } } - vb->setBuffer(); + 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; + } + + resetStriders(count); + } +} + +LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count) +{ + LLVertexBuffer * vb = new LLVertexBuffer(attribute_mask); + vb->allocateBuffer(count, 0); + + vb->setBuffer(); + + vb->setPositionData((LLVector4a*)mVerticesp.get()); + + if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) + { + vb->setTexCoord0Data(mTexcoordsp.get()); + } - if (mMode == LLRender::QUADS && sGLCoreProfile) + if (attribute_mask & LLVertexBuffer::MAP_COLOR) + { + vb->setColorData(mColorsp.get()); + } + +#if LL_DARWIN + vb->unmapBuffer(); +#endif + vb->unbind(); + + return vb; +} + +void LLRender::drawBuffer(LLVertexBuffer* vb, U32 mode, S32 count) +{ + vb->setBuffer(); + + if (mode == LLRender::QUADS && sGLCoreProfile) + { + vb->drawArrays(LLRender::TRIANGLES, 0, count); + mQuadCycle = 1; + } + else + { + vb->drawArrays(mode, 0, count); + } +} + +void LLRender::resetStriders(S32 count) +{ + mVerticesp[0] = mVerticesp[count]; + mTexcoordsp[0] = mTexcoordsp[count]; + mColorsp[0] = mColorsp[count]; + + mCount = 0; +} + +LLVertexBuffer* LLRender::getBuffer(S32 & count) +{ + STOP_GLERROR; + LLVertexBuffer *vb; + if (mCount > 0) + { + LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE; + llassert(LLGLSLShader::sCurBoundShaderPtr != nullptr); + if (!mUIOffset.empty()) + { + sUICalls++; + sUIVerts += mCount; + } + + //store mCount in a local variable to avoid re-entrance (drawArrays may call flush) + count = mCount; + + if (mMode == LLRender::QUADS && !sGLCoreProfile) + { + if (mCount % 4 != 0) { - vb->drawArrays(LLRender::TRIANGLES, 0, count); - mQuadCycle = 1; + count -= (mCount % 4); + LL_WARNS() << "Incomplete quad requested." << LL_ENDL; } - else + } + + if (mMode == LLRender::TRIANGLES) + { + if (mCount % 3 != 0) { - vb->drawArrays(mMode, 0, count); + count -= (mCount % 3); + LL_WARNS() << "Incomplete triangle requested." << LL_ENDL; } } + + if (mMode == LLRender::LINES) + { + if (mCount % 2 != 0) + { + count -= (mCount % 2); + LL_WARNS() << "Incomplete line requested." << LL_ENDL; + } + } + + mCount = 0; + + if (mBuffer) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss"); + + U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; + vb = genBuffer(attribute_mask, count); + 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; } - - mVerticesp[0] = mVerticesp[count]; - mTexcoordsp[0] = mTexcoordsp[count]; - mColorsp[0] = mColorsp[count]; - - mCount = 0; + resetStriders(count); } + else + { + count = 0; + } + + return vb; } void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) -- cgit v1.2.3 From 5c64e5e13d9a75cac510aac3128fc6ee780ab243 Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Fri, 30 Aug 2024 18:51:48 +0300 Subject: viewer#2411 LLFontGL::render optimizations #3 --- indra/llrender/llrender.cpp | 175 +++++++++++++++++++++++++++----------------- 1 file changed, 109 insertions(+), 66 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 8cd2d9cc15..182f61f907 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -77,6 +77,7 @@ struct LLVBCache }; static std::unordered_map sVBCache; +static thread_local std::list *sBufferDataList = nullptr; static const GLenum sGLTextureType[] = { @@ -1528,6 +1529,25 @@ void LLRender::clearErrors() } } +void LLRender::beginList(std::list *list) +{ + flush(); + sBufferDataList = list; +} + +void LLRender::endList() +{ + if (sBufferDataList) + { + flush(); + sBufferDataList = nullptr; + } + else + { + llassert(false); // something failed to provide a list or closed it twice + } +} + void LLRender::begin(const GLuint& mode) { if (mode != mMode) @@ -1663,78 +1683,26 @@ void LLRender::flush() if (mBuffer) { - HBXXH64 hash; - U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; + LLVertexBuffer *vb; - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash"); - - hash.update((U8*)mVerticesp.get(), count * sizeof(LLVector4a)); - if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) - { - hash.update((U8*)mTexcoordsp.get(), count * sizeof(LLVector2)); - } - - if (attribute_mask & LLVertexBuffer::MAP_COLOR) - { - hash.update((U8*)mColorsp.get(), count * sizeof(LLColor4U)); - } - - hash.finalize(); - } - - - U64 vhash = hash.digest(); - - // 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::iterator cache = sVBCache.find(vhash); - - LLPointer vb; + U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; - if (cache != sVBCache.end()) + if (sBufferDataList) { - 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(); + vb = genBuffer(attribute_mask, count); + sBufferDataList->emplace_back( + vb, + mMode, + count, + gGL.getTexUnit(0)->mCurrTexture, + mMatrix[MM_MODELVIEW][mMatIdx[MM_MODELVIEW]], + mMatrix[MM_PROJECTION][mMatIdx[MM_PROJECTION]], + mMatrix[MM_TEXTURE0][mMatIdx[MM_TEXTURE0]] + ); } else { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss"); - vb = genBuffer(attribute_mask, count); - - sVBCache[vhash] = { vb , std::chrono::steady_clock::now() }; - - static U32 miss_count = 0; - miss_count++; - if (miss_count > 1024) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache clean"); - miss_count = 0; - auto now = std::chrono::steady_clock::now(); - - 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::iterator iter = sVBCache.begin(); iter != sVBCache.end(); ) - { - if (now - iter->second.touched > 1s) - { - iter = sVBCache.erase(iter); - } - else - { - ++iter; - } - } - } + vb = bufferfromCache(attribute_mask, count); } drawBuffer(vb, mMode, count); @@ -1749,6 +1717,81 @@ void LLRender::flush() } } +LLVertexBuffer* LLRender::bufferfromCache(U32 attribute_mask, U32 count) +{ + LLVertexBuffer *vb = nullptr; + HBXXH64 hash; + + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hash"); + + hash.update((U8*)mVerticesp.get(), count * sizeof(LLVector4a)); + if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) + { + hash.update((U8*)mTexcoordsp.get(), count * sizeof(LLVector2)); + } + + if (attribute_mask & LLVertexBuffer::MAP_COLOR) + { + hash.update((U8*)mColorsp.get(), count * sizeof(LLColor4U)); + } + + hash.finalize(); + } + + U64 vhash = hash.digest(); + + // 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::iterator cache = sVBCache.find(vhash); + + if (cache != sVBCache.end()) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache hit"); + // cache hit, just use the cached buffer + vb = cache->second.vb; + cache->second.touched = std::chrono::steady_clock::now(); + } + else + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss"); + vb = genBuffer(attribute_mask, count); + + sVBCache[vhash] = { vb , std::chrono::steady_clock::now() }; + + static U32 miss_count = 0; + miss_count++; + if (miss_count > 1024) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache clean"); + miss_count = 0; + auto now = std::chrono::steady_clock::now(); + + 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::iterator iter = sVBCache.begin(); iter != sVBCache.end(); ) + { + if (now - iter->second.touched > 1s) + { + iter = sVBCache.erase(iter); + } + else + { + ++iter; + } + } + } + } + return vb; +} + LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count) { LLVertexBuffer * vb = new LLVertexBuffer(attribute_mask); -- cgit v1.2.3 From 57ab1a410f9cb3534bb403e034743505758579d8 Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Mon, 2 Sep 2024 13:46:13 +0300 Subject: viewer#2411 A bit more coverage for font buffer --- indra/llrender/llrender.cpp | 1 + 1 file changed, 1 insertion(+) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 182f61f907..b906563b08 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1531,6 +1531,7 @@ void LLRender::clearErrors() void LLRender::beginList(std::list *list) { + llassert(LLGLSLShader::sCurBoundShaderPtr == &gUIProgram); flush(); sBufferDataList = list; } -- cgit v1.2.3 From 4cc0edb189c4deaa205f986d20a6959aa83fb25c Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Mon, 2 Sep 2024 22:14:42 +0300 Subject: viewer#2411 Cleanup --- indra/llrender/llrender.cpp | 114 -------------------------------------------- 1 file changed, 114 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index b906563b08..a40a668a70 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1592,50 +1592,6 @@ void LLRender::end() } } -LLVertexBuffer* LLRender::beginNoCache(const GLuint& mode, S32& count) -{ - if (mode != mMode) - { - if (mode == LLRender::QUADS) - { - mQuadCycle = 1; - } - - if (mMode == LLRender::QUADS || - mMode == LLRender::LINES || - mMode == LLRender::TRIANGLES || - mMode == LLRender::POINTS) - { - return getBuffer(count); - } - else if (mCount != 0) - { - LL_ERRS() << "gGL.begin() called redundantly." << LL_ENDL; - } - mMode = mode; - } - return nullptr; -} - -LLVertexBuffer* LLRender::endNoCache(S32& count) -{ - if (mCount == 0) - { - return nullptr; - //IMM_ERRS << "GL begin and end called with no vertices specified." << LL_ENDL; - } - - if ((mMode != LLRender::QUADS && - mMode != LLRender::LINES && - mMode != LLRender::TRIANGLES && - mMode != LLRender::POINTS) || - mCount > 2048) - { - return getBuffer(count); - } - return nullptr; -} - void LLRender::flush() { STOP_GLERROR; @@ -1844,76 +1800,6 @@ void LLRender::resetStriders(S32 count) mCount = 0; } -LLVertexBuffer* LLRender::getBuffer(S32 & count) -{ - STOP_GLERROR; - LLVertexBuffer *vb; - if (mCount > 0) - { - LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE; - llassert(LLGLSLShader::sCurBoundShaderPtr != nullptr); - if (!mUIOffset.empty()) - { - sUICalls++; - sUIVerts += mCount; - } - - //store mCount in a local variable to avoid re-entrance (drawArrays may call flush) - 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; - } - } - - if (mMode == LLRender::LINES) - { - if (mCount % 2 != 0) - { - count -= (mCount % 2); - LL_WARNS() << "Incomplete line requested." << LL_ENDL; - } - } - - mCount = 0; - - if (mBuffer) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb cache miss"); - - U32 attribute_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; - vb = genBuffer(attribute_mask, count); - 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; - } - - resetStriders(count); - } - else - { - count = 0; - } - - return vb; -} - void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) { //the range of mVerticesp, mColorsp and mTexcoordsp is [0, 4095] -- cgit v1.2.3 From 2fc8d5ff3cfa1b9ad00b310cd4a6cdb557b9415c Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Tue, 3 Sep 2024 17:22:04 +0300 Subject: viewer#2411 Cleanup #2 --- indra/llrender/llrender.cpp | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index a40a668a70..828a509971 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1531,6 +1531,10 @@ void LLRender::clearErrors() void LLRender::beginList(std::list *list) { + if (sBufferDataList) + { + LL_ERRS() << "beginList called while another list is open." << LL_ENDL; + } llassert(LLGLSLShader::sCurBoundShaderPtr == &gUIProgram); flush(); sBufferDataList = list; @@ -1545,7 +1549,7 @@ void LLRender::endList() } else { - llassert(false); // something failed to provide a list or closed it twice + llassert(false); // endList called without an open list } } -- cgit v1.2.3 From b713f56d07cffb21cad5928bf30d6d4b6dc79de9 Mon Sep 17 00:00:00 2001 From: Rye Mutt Date: Thu, 12 Sep 2024 09:22:10 -0700 Subject: Replace glh_linear usage with GLM (#2554) --- indra/llrender/llrender.cpp | 216 ++++++++++++++++++++------------------------ 1 file changed, 98 insertions(+), 118 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 828a509971..16a8309a9a 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, @@ -57,8 +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]; @@ -735,10 +736,10 @@ 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(glm::make_vec4(position.mV)); + const glm::mat4& mat = gGL.getModelviewMatrix(); + pos = mat * pos; + mPosition.set(glm::value_ptr(pos)); } void LLLightState::setConstantAttenuation(const F32& atten) @@ -790,13 +791,13 @@ 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(glm::make_vec3(direction.mV)); + const glm::mat3 mat(gGL.getModelviewMatrix()); + dir = mat * dir; - mSpotDirection.set(dir.v); + mSpotDirection.set(glm::value_ptr(dir)); } LLRender::LLRender() @@ -830,6 +831,10 @@ 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(); + } mMatIdx[i] = 0; mMatHash[i] = 0; mCurMatHash[i] = 0xFFFFFFFF; @@ -991,12 +996,12 @@ 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) @@ -1006,15 +1011,15 @@ 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 @@ -1023,17 +1028,17 @@ 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); @@ -1041,7 +1046,7 @@ 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 @@ -1054,36 +1059,36 @@ 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())); } - 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) @@ -1096,12 +1101,12 @@ 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)); } } } @@ -1110,7 +1115,7 @@ 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]; } } @@ -1129,12 +1134,7 @@ 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]++; } } @@ -1144,12 +1144,7 @@ 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]++; } } @@ -1159,13 +1154,7 @@ 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]++; } } @@ -1175,19 +1164,7 @@ 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]++; } } @@ -1229,7 +1206,7 @@ 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]++; } } @@ -1238,9 +1215,7 @@ 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]++; } } @@ -1283,17 +1258,17 @@ void LLRender::loadIdentity() { llassert_always(mMatrixMode < NUM_MATRIX_MODES) ; - mMatrix[mMatrixMode][mMatIdx[mMatrixMode]].make_identity(); + mMatrix[mMatrixMode][mMatIdx[mMatrixMode]] = glm::identity(); 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]]; } @@ -2214,85 +2189,90 @@ void LLRender::debugTexUnits(void) LL_INFOS("TextureUnit") << "Active TexUnit Enabled : " << active_enabled << LL_ENDL; } - - -glh::matrix4f copy_matrix(F32* src) -{ - glh::matrix4f ret; - ret.set_value(src); - return ret; -} - -glh::matrix4f get_current_modelview() +glm::mat4 get_current_modelview() { - return copy_matrix(gGLModelView); + return glm::make_mat4(gGLModelView); } -glh::matrix4f get_current_projection() +glm::mat4 get_current_projection() { - return copy_matrix(gGLProjection); + return glm::make_mat4(gGLProjection); } -glh::matrix4f get_last_modelview() +glm::mat4 get_last_modelview() { - return copy_matrix(gGLLastModelView); + return glm::make_mat4(gGLLastModelView); } -glh::matrix4f get_last_projection() +glm::mat4 get_last_projection() { - return copy_matrix(gGLLastProjection); + return glm::make_mat4(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; + //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 + //); + 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); + + s.mul(x); + t.mul(y); + p.mul(z); + q.add(s); + t.add(p); + q.add(t); - 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); + 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()); } -- cgit v1.2.3 From d9da5bbb33ce70e3bc799ba6696c8b10de0e5f04 Mon Sep 17 00:00:00 2001 From: Ansariel Hiller Date: Wed, 18 Sep 2024 16:09:51 +0200 Subject: Remove quads rendering mode entirely (#2593) --- indra/llrender/llrender.cpp | 212 ++++++-------------------------------------- 1 file changed, 27 insertions(+), 185 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 16a8309a9a..f432b09107 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -803,7 +803,6 @@ void LLLightState::setSpotDirection(const LLVector3& direction) LLRender::LLRender() : mDirty(false), mCount(0), - mQuadCycle(0), mMode(LLRender::TRIANGLES), mCurrTextureUnitIndex(0) { @@ -1532,13 +1531,7 @@ 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) { @@ -1561,8 +1554,7 @@ 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 > 2048) @@ -1587,28 +1579,19 @@ 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 (mMode == LLRender::TRIANGLES) + { + if (mCount%3 != 0) { - if (mCount%3 != 0) - { - count -= (mCount % 3); - LL_WARNS() << "Incomplete triangle requested." << LL_ENDL; - } + count -= (mCount % 3); + LL_WARNS() << "Incomplete triangle requested." << LL_ENDL; } + } - if (mMode == LLRender::LINES) + if (mMode == LLRender::LINES) + { + if (mCount%2 != 0) { - if (mCount%2 != 0) - { count -= (mCount % 2); LL_WARNS() << "Incomplete line requested." << LL_ENDL; } @@ -1758,16 +1741,7 @@ LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count) void LLRender::drawBuffer(LLVertexBuffer* vb, U32 mode, S32 count) { vb->setBuffer(); - - if (mode == LLRender::QUADS && sGLCoreProfile) - { - vb->drawArrays(LLRender::TRIANGLES, 0, count); - mQuadCycle = 1; - } - else - { - vb->drawArrays(mode, 0, count); - } + vb->drawArrays(mode, 0, count); } void LLRender::resetStriders(S32 count) @@ -1788,7 +1762,6 @@ void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) { 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; } } @@ -1809,25 +1782,6 @@ void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) mVerticesp[mCount] = vert; } - 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]; - } - } - mCount++; mVerticesp[mCount] = mVerticesp[mCount-1]; mColorsp[mCount] = mColorsp[mCount-1]; @@ -1842,50 +1796,13 @@ void LLRender::vertexBatchPreTransformed(LLVector3* verts, S32 vert_count) 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 @@ -1900,50 +1817,13 @@ void LLRender::vertexBatchPreTransformed(LLVector3* verts, LLVector2* uvs, S32 v 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) @@ -1961,51 +1841,13 @@ void LLRender::vertexBatchPreTransformed(LLVector3* verts, LLVector2* uvs, LLCol 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) -- cgit v1.2.3 From 6d842ac0af814a088c56f437dc885e4ce58b61a8 Mon Sep 17 00:00:00 2001 From: Andrey Kleshchev Date: Wed, 18 Sep 2024 17:11:03 +0300 Subject: Expose LLVector4a in LLRender Avoid using a bunch of allocators. Make sure we use LLVector4a's SSE logic instead of LLVector3's. Some minor optimizations. --- indra/llrender/llrender.cpp | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'indra/llrender/llrender.cpp') diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index f432b09107..c8a9fb4893 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1298,7 +1298,7 @@ void LLRender::pushUIMatrix() { if (mUIOffset.empty()) { - mUIOffset.push_back(LLVector3(0,0,0)); + mUIOffset.emplace_back(0.f,0.f,0.f); } else { @@ -1307,7 +1307,7 @@ void LLRender::pushUIMatrix() if (mUIScale.empty()) { - mUIScale.push_back(LLVector3(1,1,1)); + mUIScale.emplace_back(1.f,1.f,1.f); } else { @@ -1718,7 +1718,7 @@ LLVertexBuffer* LLRender::genBuffer(U32 attribute_mask, S32 count) vb->setBuffer(); - vb->setPositionData((LLVector4a*)mVerticesp.get()); + vb->setPositionData(mVerticesp.get()); if (attribute_mask & LLVertexBuffer::MAP_TEXCOORD0) { @@ -1774,12 +1774,12 @@ void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) if (mUIOffset.empty()) { - mVerticesp[mCount] = LLVector3(x,y,z); + mVerticesp[mCount].set(x,y,z); } else { LLVector3 vert = (LLVector3(x,y,z)+mUIOffset.back()).scaledVec(mUIScale.back()); - mVerticesp[mCount] = vert; + mVerticesp[mCount].set(vert.mV[VX], vert.mV[VY], vert.mV[VZ]); } mCount++; @@ -1788,7 +1788,7 @@ void LLRender::vertex3f(const GLfloat& x, const GLfloat& y, const GLfloat& z) mTexcoordsp[mCount] = mTexcoordsp[mCount-1]; } -void LLRender::vertexBatchPreTransformed(LLVector3* verts, S32 vert_count) +void LLRender::vertexBatchPreTransformed(LLVector4a* verts, S32 vert_count) { if (mCount + vert_count > 4094) { @@ -1809,7 +1809,7 @@ void LLRender::vertexBatchPreTransformed(LLVector3* verts, S32 vert_count) mVerticesp[mCount] = mVerticesp[mCount-1]; } -void LLRender::vertexBatchPreTransformed(LLVector3* verts, LLVector2* uvs, S32 vert_count) +void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, S32 vert_count) { if (mCount + vert_count > 4094) { @@ -1833,7 +1833,7 @@ void LLRender::vertexBatchPreTransformed(LLVector3* verts, LLVector2* uvs, S32 v } } -void LLRender::vertexBatchPreTransformed(LLVector3* verts, LLVector2* uvs, LLColor4U* colors, S32 vert_count) +void LLRender::vertexBatchPreTransformed(LLVector4a* verts, LLVector2* uvs, LLColor4U* colors, S32 vert_count) { if (mCount + vert_count > 4094) { -- cgit v1.2.3