From 14f6bbadef2c39e58a3b54c0c6212949acf50e45 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 8 Aug 2011 15:29:23 -0500 Subject: SH-2242 Work in progress migrating to glVertexAttrib everywhere --- indra/newview/llface.cpp | 48 ++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 44 insertions(+), 4 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 432e61f6d8..f5a8013f4d 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -362,8 +362,8 @@ void LLFace::setSize(S32 num_vertices, S32 num_indices, bool align) { if (align) { - //allocate vertices in blocks of 4 for alignment - num_vertices = (num_vertices + 0x3) & ~0x3; + //allocate vertices in blocks of 16 for alignment + num_vertices = (num_vertices + 0xF) & ~0xF; } if (mGeomCount != num_vertices || @@ -503,7 +503,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) glMultMatrixf((GLfloat*)mDrawablep->getRegion()->mRenderMatrix.mMatrix); } - glColor4fv(color.mV); + gGL.diffuseColor4fv(color.mV); if (mDrawablep->isState(LLDrawable::RIGGED)) { @@ -1055,6 +1055,7 @@ static LLFastTimer::DeclareTimer FTM_FACE_GEOM_POSITION("Position"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_NORMAL("Normal"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_TEXTURE("Texture"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_COLOR("Color"); +static LLFastTimer::DeclareTimer FTM_FACE_GEOM_EMISSIVE("Emissive"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_WEIGHTS("Weights"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_BINORMAL("Binormal"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_INDEX("Index"); @@ -1124,6 +1125,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, bool rebuild_pos = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_POSITION); bool rebuild_color = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_COLOR); + bool rebuild_emissive = rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_EMISSIVE); bool rebuild_tcoord = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_TCOORD); bool rebuild_normal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL); bool rebuild_binormal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_BINORMAL); @@ -1758,6 +1760,44 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } } + if (rebuild_emissive) + { + LLFastTimer t(FTM_FACE_GEOM_EMISSIVE); + LLStrider emissive; + mVertexBuffer->getEmissiveStrider(emissive, mGeomIndex, mGeomCount, map_range); + + U8 glow = (U8) llclamp((S32) (getTextureEntry()->getGlow()*255), 0, 255); + + LLVector4a src; + + + U32 glow32 = glow | + (glow << 8) | + (glow << 16) | + (glow << 24); + + U32 vec[4]; + vec[0] = vec[1] = vec[2] = vec[3] = glow32; + + src.loadua((F32*) vec); + + LLVector4a* dst = (LLVector4a*) emissive.get(); + S32 num_vecs = num_vertices/16; + if (num_vertices%16 > 0) + { + ++num_vecs; + } + + for (S32 i = 0; i < num_vecs; i++) + { + dst[i] = src; + } + + if (map_range) + { + mVertexBuffer->setBuffer(0); + } + } if (rebuild_tcoord) { mTexExtents[0].setVec(0,0); @@ -2095,7 +2135,7 @@ void LLFace::renderSetColor() const { const LLColor4* color = &(getRenderColor()); - glColor4fv(color->mV); + gGL.diffuseColor4fv(color->mV); } } -- cgit v1.2.3 From 7c95af74f195c9ec4ebc0fc0264d98cd4a85be49 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 14 Sep 2011 16:30:45 -0500 Subject: SH-2243 work in progress -- application side matrix stack management --- indra/newview/llface.cpp | 27 ++++++++++++++------------- 1 file changed, 14 insertions(+), 13 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index f5a8013f4d..7ae11bff94 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -496,11 +496,11 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) gGL.pushMatrix(); if (mDrawablep->isActive()) { - glMultMatrixf((GLfloat*)mDrawablep->getRenderMatrix().mMatrix); + gGL.multMatrix((GLfloat*)mDrawablep->getRenderMatrix().mMatrix); } else { - glMultMatrixf((GLfloat*)mDrawablep->getRegion()->mRenderMatrix.mMatrix); + gGL.multMatrix((GLfloat*)mDrawablep->getRegion()->mRenderMatrix.mMatrix); } gGL.diffuseColor4fv(color.mV); @@ -515,7 +515,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) { LLGLEnable offset(GL_POLYGON_OFFSET_FILL); glPolygonOffset(-1.f, -1.f); - glMultMatrixf((F32*) volume->getRelativeXform().mMatrix); + gGL.multMatrix((F32*) volume->getRelativeXform().mMatrix); const LLVolumeFace& vol_face = rigged->getVolumeFace(getTEOffset()); LLVertexBuffer::unbind(); glVertexPointer(3, GL_FLOAT, 16, vol_face.mPositions); @@ -524,6 +524,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) glEnableClientState(GL_TEXTURE_COORD_ARRAY); glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords); } + gGL.syncMatrices(); glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices); glDisableClientState(GL_TEXTURE_COORD_ARRAY); } @@ -557,17 +558,17 @@ void LLFace::renderSelectedUV() // add green dither pattern on top of red/blue gradient gGL.blendFunc(LLRender::BF_ONE, LLRender::BF_ONE); - glMatrixMode(GL_TEXTURE); - glPushMatrix(); + gGL.matrixMode(LLRender::MM_TEXTURE); + gGL.pushMatrix(); // make green pattern repeat once per texel in red/blue texture - glScalef(256.f, 256.f, 1.f); - glMatrixMode(GL_MODELVIEW); + gGL.scalef(256.f, 256.f, 1.f); + gGL.matrixMode(LLRender::MM_MODELVIEW); renderSelected(green_imagep, LLColor4::white); - glMatrixMode(GL_TEXTURE); - glPopMatrix(); - glMatrixMode(GL_MODELVIEW); + gGL.matrixMode(LLRender::MM_TEXTURE); + gGL.popMatrix(); + gGL.matrixMode(LLRender::MM_MODELVIEW); gGL.blendFunc(LLRender::BF_SOURCE_ALPHA, LLRender::BF_ONE_MINUS_SOURCE_ALPHA); } */ @@ -2170,10 +2171,10 @@ S32 LLFace::renderElements(const U16 *index_array) const } else { - glPushMatrix(); - glMultMatrixf((float*)getRenderMatrix().mMatrix); + gGL.pushMatrix(); + gGL.multMatrix((float*)getRenderMatrix().mMatrix); ret = pushVertices(index_array); - glPopMatrix(); + gGL.popMatrix(); } return ret; -- cgit v1.2.3 From b9926e8f57787eb146b06260cc3d0260e34330ce Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 23 Sep 2011 02:29:53 -0500 Subject: SH-2244 Better VAO support -- still slower than non-VAO implementation for some reason --- indra/newview/llface.cpp | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 7ae11bff94..d36379b0e7 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1216,7 +1216,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1441,7 +1441,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } else @@ -1588,7 +1588,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } if (do_bump) @@ -1625,7 +1625,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } } @@ -1675,7 +1675,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1695,7 +1695,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1715,7 +1715,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1727,7 +1727,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLVector4a::memcpyNonAliased16((F32*) weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32)); if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1757,7 +1757,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } @@ -1796,7 +1796,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (map_range) { - mVertexBuffer->setBuffer(0); + mVertexBuffer->flush(); } } if (rebuild_tcoord) -- cgit v1.2.3 From 54ba56dbe995a3d215f85932c83948145a871ac4 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 26 Sep 2011 18:31:55 -0500 Subject: SH-2244 changes to run in a core context on AMD hardware without generating deprecation or performance warnings --- indra/newview/llface.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index d36379b0e7..36b88ebbd4 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1764,7 +1764,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_emissive) { LLFastTimer t(FTM_FACE_GEOM_EMISSIVE); - LLStrider emissive; + LLStrider emissive; mVertexBuffer->getEmissiveStrider(emissive, mGeomIndex, mGeomCount, map_range); U8 glow = (U8) llclamp((S32) (getTextureEntry()->getGlow()*255), 0, 255); @@ -1783,8 +1783,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, src.loadua((F32*) vec); LLVector4a* dst = (LLVector4a*) emissive.get(); - S32 num_vecs = num_vertices/16; - if (num_vertices%16 > 0) + S32 num_vecs = num_vertices/4; + if (num_vertices%4 > 0) { ++num_vecs; } -- cgit v1.2.3 From 7b6723d1e0158d5dc326266a0332e87f634f9755 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 13 Oct 2011 01:19:45 -0500 Subject: SH-1650 Mitigate memory fragmentation by holding onto and reusing VBOs --- indra/newview/llface.cpp | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 36b88ebbd4..eab3dcfadd 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -362,8 +362,8 @@ void LLFace::setSize(S32 num_vertices, S32 num_indices, bool align) { if (align) { - //allocate vertices in blocks of 16 for alignment - num_vertices = (num_vertices + 0xF) & ~0xF; + //allocate vertices in blocks of 4 for alignment + num_vertices = (num_vertices + 0x3) & ~0x3; } if (mGeomCount != num_vertices || @@ -1073,6 +1073,11 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, S32 num_vertices = (S32)vf.mNumVertices; S32 num_indices = (S32) vf.mNumIndices; + if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_OCTREE)) + { + updateRebuildFlags(); + } + bool map_range = gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange; if (mVertexBuffer.notNull()) @@ -2055,7 +2060,7 @@ BOOL LLFace::verify(const U32* indices_array) const } // First, check whether the face data fits within the pool's range. - if ((mGeomIndex + mGeomCount) > mVertexBuffer->getRequestedVerts()) + if ((mGeomIndex + mGeomCount) > mVertexBuffer->getNumVerts()) { ok = FALSE; llinfos << "Face references invalid vertices!" << llendl; @@ -2074,7 +2079,7 @@ BOOL LLFace::verify(const U32* indices_array) const llinfos << "Face has bogus indices count" << llendl; } - if (mIndicesIndex + mIndicesCount > mVertexBuffer->getRequestedIndices()) + if (mIndicesIndex + mIndicesCount > mVertexBuffer->getNumIndices()) { ok = FALSE; llinfos << "Face references invalid indices!" << llendl; -- cgit v1.2.3 From 5c690db63ff118b6bc66fe1607624a59f4162b30 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 18 Nov 2011 15:04:26 -0600 Subject: SH-2700 Fix for random black textures due to texture index out of bounds. Also change shader to show bright pink on index out of bounds and add assertions to help prevent this sort of bug in the future. --- indra/newview/llface.cpp | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index eab3dcfadd..6dbeae6677 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1658,6 +1658,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, while(dst < end); F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0); + + llassert(index <= LLGLSLShader::sIndexedTextureChannels-1); F32 *index_dst = (F32*) vertices; F32 *index_end = (F32*) end; -- cgit v1.2.3 From 5546041301a67da6e471a433b29374bc75a4f0d0 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 21 Dec 2011 15:55:14 -0600 Subject: SH-2794 WIP -- work on getting glMapBuffer and friends to behave --- indra/newview/llface.cpp | 194 +++++++++++++++++++++++++++++++++++------------ 1 file changed, 147 insertions(+), 47 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 6dbeae6677..941b8db2cf 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -693,6 +693,49 @@ static void xform(LLVector2 &tex_coord, F32 cosAng, F32 sinAng, F32 offS, F32 of tex_coord.mV[1] = t; } +// Transform the texture coordinates for this face. +static void xform4a(LLVector4a &tex_coord, const LLVector4a& trans, const LLVector4Logical& mask, const LLVector4a& rot0, const LLVector4a& rot1, const LLVector4a& offset, const LLVector4a& scale) +{ + //tex coord is two coords, + LLVector4a st; + + // Texture transforms are done about the center of the face. + st.setAdd(tex_coord, trans); + + // Handle rotation + LLVector4a rot_st; + + // + LLVector4a s0; + s0.splat(st, 0); + LLVector4a s1; + s1.splat(st, 2); + LLVector4a ss; + ss.setSelectWithMask(mask, s1, s0); + + LLVector4a a; + a.setMul(rot0, ss); + + // + LLVector4a t0; + t0.splat(st, 1); + LLVector4a t1; + t1.splat(st, 3); + LLVector4a tt; + tt.setSelectWithMask(mask, t1, t0); + + LLVector4a b; + b.setMul(rot1, tt); + + st.setAdd(a,b); + + // Then scale + st.mul(scale); + + // Then offset + tex_coord.setAdd(st, offset); +} + bool less_than_max_mag(const LLVector4a& vec) { @@ -1060,6 +1103,16 @@ static LLFastTimer::DeclareTimer FTM_FACE_GEOM_EMISSIVE("Emissive"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_WEIGHTS("Weights"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_BINORMAL("Binormal"); static LLFastTimer::DeclareTimer FTM_FACE_GEOM_INDEX("Index"); +static LLFastTimer::DeclareTimer FTM_FACE_GEOM_INDEX_TAIL("Tail"); +static LLFastTimer::DeclareTimer FTM_FACE_POSITION_STORE("Pos"); +static LLFastTimer::DeclareTimer FTM_FACE_TEXTURE_INDEX_STORE("TexIdx"); +static LLFastTimer::DeclareTimer FTM_FACE_POSITION_PAD("Pad"); +static LLFastTimer::DeclareTimer FTM_FACE_TEX_DEFAULT("Default"); +static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK("Quick"); +static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_NO_XFORM("No Xform"); +static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_XFORM("Xform"); + +static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_PLANAR("Quick Planar"); BOOL LLFace::getGeometryVolume(const LLVolume& volume, const S32 &f, @@ -1104,16 +1157,12 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } LLStrider vert; - LLVector4a* vertices = NULL; LLStrider tex_coords; LLStrider tex_coords2; - LLVector4a* normals = NULL; LLStrider norm; LLStrider colors; - LLVector4a* binormals = NULL; LLStrider binorm; LLStrider indicesp; - LLVector4a* weights = NULL; LLStrider wght; BOOL full_rebuild = force_rebuild || mDrawablep->isState(LLDrawable::REBUILD_VOLUME); @@ -1202,7 +1251,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLFastTimer t(FTM_FACE_GEOM_INDEX); mVertexBuffer->getIndexStrider(indicesp, mIndicesIndex, mIndicesCount, map_range); - __m128i* dst = (__m128i*) indicesp.get(); + volatile __m128i* dst = (__m128i*) indicesp.get(); __m128i* src = (__m128i*) vf.mIndices; __m128i offset = _mm_set1_epi16(index_offset); @@ -1211,12 +1260,17 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, for (S32 i = 0; i < end; i++) { __m128i res = _mm_add_epi16(src[i], offset); - _mm_storeu_si128(dst+i, res); + _mm_storeu_si128((__m128i*) dst++, res); } - for (S32 i = end*8; i < num_indices; ++i) { - indicesp[i] = vf.mIndices[i]+index_offset; + LLFastTimer t(FTM_FACE_GEOM_INDEX_TAIL); + U16* idx = (U16*) dst; + + for (S32 i = end*8; i < num_indices; ++i) + { + *idx++ = vf.mIndices[i]+index_offset; + } } if (map_range) @@ -1373,19 +1427,48 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (texgen != LLTextureEntry::TEX_GEN_PLANAR) { + LLFastTimer t(FTM_FACE_TEX_QUICK); if (!do_tex_mat) { if (!do_xform) { + LLFastTimer t(FTM_FACE_TEX_QUICK_NO_XFORM); LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32)); } else { - for (S32 i = 0; i < num_vertices; i++) + LLFastTimer t(FTM_FACE_TEX_QUICK_XFORM); + F32* dst = (F32*) tex_coords.get(); + LLVector4a* src = (LLVector4a*) vf.mTexCoords; + + LLVector4a trans; + trans.splat(-0.5f); + + LLVector4a rot0; + rot0.set(cos_ang, -sin_ang, cos_ang, -sin_ang); + + LLVector4a rot1; + rot1.set(sin_ang, cos_ang, sin_ang, cos_ang); + + LLVector4a scale; + scale.set(ms, mt, ms, mt); + + LLVector4a offset; + offset.set(os+0.5f, ot+0.5f, os+0.5f, ot+0.5f); + + LLVector4Logical mask; + mask.clear(); + mask.setElement<2>(); + mask.setElement<3>(); + + U32 count = num_vertices/2 + num_vertices%2; + + for (S32 i = 0; i < count; i++) { - LLVector2 tc(vf.mTexCoords[i]); - xform(tc, cos_ang, sin_ang, os, ot, ms, mt); - *tex_coords++ = tc; + LLVector4a res = *src++; + xform4a(res, trans, mask, rot0, rot1, offset, scale); + res.store4a(dst); + dst += 4; } } } @@ -1407,6 +1490,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } else { //no bump, no atlas, tex gen planar + LLFastTimer t(FTM_FACE_TEX_QUICK_PLANAR); if (do_tex_mat) { for (S32 i = 0; i < num_vertices; i++) @@ -1451,6 +1535,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } else { //either bump mapped or in atlas, just do the whole expensive loop + LLFastTimer t(FTM_FACE_TEX_DEFAULT); mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount, map_range); std::vector bump_tc; @@ -1642,44 +1727,55 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, llassert(num_vertices > 0); mVertexBuffer->getVertexStrider(vert, mGeomIndex, mGeomCount, map_range); - vertices = (LLVector4a*) vert.get(); - + + LLMatrix4a mat_vert; mat_vert.loadu(mat_vert_in); LLVector4a* src = vf.mPositions; - LLVector4a* dst = vertices; + volatile F32* dst = (volatile F32*) vert.get(); - LLVector4a* end = dst+num_vertices; - do - { - mat_vert.affineTransform(*src++, *dst++); - } - while(dst < end); + volatile F32* end = dst+num_vertices*4; + LLVector4a res; - F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0); + LLVector4a texIdx; + F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0); llassert(index <= LLGLSLShader::sIndexedTextureChannels-1); - F32 *index_dst = (F32*) vertices; - F32 *index_end = (F32*) end; - index_dst += 3; - index_end += 3; - do + LLVector4Logical mask; + mask.clear(); + mask.setElement<3>(); + + texIdx.set(0,0,0,index); + { - *index_dst = index; - index_dst += 4; + LLFastTimer t(FTM_FACE_POSITION_STORE); + LLVector4a tmp; + + do + { + mat_vert.affineTransform(*src++, res); + tmp.setSelectWithMask(mask, texIdx, res); + tmp.store4a((F32*) dst); + dst += 4; + } + while(dst < end); } - while (index_dst < index_end); - - S32 aligned_pad_vertices = mGeomCount - num_vertices; - LLVector4a* last_vec = end - 1; - while (aligned_pad_vertices > 0) + { - --aligned_pad_vertices; - *dst++ = *last_vec; + LLFastTimer t(FTM_FACE_POSITION_PAD); + S32 aligned_pad_vertices = mGeomCount - num_vertices; + res.set(res[0], res[1], res[2], 0.f); + + while (aligned_pad_vertices > 0) + { + --aligned_pad_vertices; + res.store4a((F32*) dst); + dst += 4; + } } - + if (map_range) { mVertexBuffer->flush(); @@ -1690,14 +1786,15 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, { LLFastTimer t(FTM_FACE_GEOM_NORMAL); mVertexBuffer->getNormalStrider(norm, mGeomIndex, mGeomCount, map_range); - normals = (LLVector4a*) norm.get(); + F32* normals = (F32*) norm.get(); for (S32 i = 0; i < num_vertices; i++) { LLVector4a normal; mat_normal.rotate(vf.mNormals[i], normal); normal.normalize3fast(); - normals[i] = normal; + normal.store4a(normals); + normals += 4; } if (map_range) @@ -1710,14 +1807,15 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, { LLFastTimer t(FTM_FACE_GEOM_BINORMAL); mVertexBuffer->getBinormalStrider(binorm, mGeomIndex, mGeomCount, map_range); - binormals = (LLVector4a*) binorm.get(); + F32* binormals = (F32*) binorm.get(); for (S32 i = 0; i < num_vertices; i++) { LLVector4a binormal; mat_normal.rotate(vf.mBinormals[i], binormal); binormal.normalize3fast(); - binormals[i] = binormal; + binormal.store4a(binormals); + binormals += 4; } if (map_range) @@ -1730,8 +1828,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, { LLFastTimer t(FTM_FACE_GEOM_WEIGHTS); mVertexBuffer->getWeight4Strider(wght, mGeomIndex, mGeomCount, map_range); - weights = (LLVector4a*) wght.get(); - LLVector4a::memcpyNonAliased16((F32*) weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32)); + F32* weights = (F32*) wght.get(); + LLVector4a::memcpyNonAliased16(weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32)); if (map_range) { mVertexBuffer->flush(); @@ -1750,7 +1848,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, src.loadua((F32*) vec); - LLVector4a* dst = (LLVector4a*) colors.get(); + F32* dst = (F32*) colors.get(); S32 num_vecs = num_vertices/4; if (num_vertices%4 > 0) { @@ -1759,7 +1857,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, for (S32 i = 0; i < num_vecs; i++) { - dst[i] = src; + src.store4a(dst); + dst += 4; } if (map_range) @@ -1789,7 +1888,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, src.loadua((F32*) vec); - LLVector4a* dst = (LLVector4a*) emissive.get(); + F32* dst = (F32*) emissive.get(); S32 num_vecs = num_vertices/4; if (num_vertices%4 > 0) { @@ -1798,7 +1897,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, for (S32 i = 0; i < num_vecs; i++) { - dst[i] = src; + src.store4a(dst); + dst += 4; } if (map_range) -- cgit v1.2.3 From 0c1fc78bd94014ee19da690f16cef64c13e50771 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 18 Jan 2012 12:40:11 -0600 Subject: SH-2794 Map buffer now performs as well as buffersubdata --- indra/newview/llface.cpp | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 941b8db2cf..cd33a19a2a 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1131,7 +1131,9 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, updateRebuildFlags(); } - bool map_range = gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange; + + //don't use map range (generates many redundant unmap calls) + bool map_range = false; //gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange; if (mVertexBuffer.notNull()) { @@ -1921,6 +1923,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, mTexExtents[1][1] *= et ; } + mLastVertexBuffer = mVertexBuffer; mLastGeomCount = mGeomCount; mLastGeomIndex = mGeomIndex; -- cgit v1.2.3 From 3710c6110d65d3a604f7b419cd764cf5b9b98600 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 10 Feb 2012 20:04:19 -0600 Subject: SH-2908 Rework indexed texture rendering to use a uvec4 instead of a float for texture indices in the data stream. Also rework gl_FragColor overrides to not collide with some odd driver implementations. --- indra/newview/llface.cpp | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index cd33a19a2a..838e541145 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1742,14 +1742,22 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLVector4a texIdx; - F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0); + U8 index = mTextureIndex < 255 ? mTextureIndex : 0; + + F32 val = 0.f; + U8* vp = (U8*) &val; + vp[0] = index; + vp[1] = 0; + vp[2] = 0; + vp[3] = 0; + llassert(index <= LLGLSLShader::sIndexedTextureChannels-1); LLVector4Logical mask; mask.clear(); mask.setElement<3>(); - texIdx.set(0,0,0,index); + texIdx.set(0,0,0,val); { LLFastTimer t(FTM_FACE_POSITION_STORE); -- cgit v1.2.3 From 38ad8611c988e6ca0caae95d397e3ce67f1ceb36 Mon Sep 17 00:00:00 2001 From: Oz Linden Date: Tue, 27 Mar 2012 10:55:15 -0400 Subject: MAINT-710: Fix for VertexBufferStrider could not find valid vertex data. (from Firestorm) --- indra/newview/llface.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/newview/llface.cpp') diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index cd33a19a2a..cfb4147e71 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -1838,7 +1838,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } } - if (rebuild_color) + if (rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_COLOR) ) { LLFastTimer t(FTM_FACE_GEOM_COLOR); mVertexBuffer->getColorStrider(colors, mGeomIndex, mGeomCount, map_range); -- cgit v1.2.3