diff options
Diffstat (limited to 'indra')
| -rwxr-xr-x | indra/llmath/llvector4a.inl | 20 | ||||
| -rwxr-xr-x | indra/llmath/llvolume.cpp | 83 | ||||
| -rwxr-xr-x | indra/llrender/llrendertarget.cpp | 4 | ||||
| -rwxr-xr-x | indra/newview/app_settings/settings.xml | 12 | ||||
| -rwxr-xr-x | indra/newview/llface.cpp | 6 | ||||
| -rwxr-xr-x | indra/newview/pipeline.cpp | 21 |
6 files changed, 96 insertions, 50 deletions
diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl index 7c52ffef21..558fe09323 100755 --- a/indra/llmath/llvector4a.inl +++ b/indra/llmath/llvector4a.inl @@ -409,6 +409,26 @@ inline void LLVector4a::normalize3fast() mQ = _mm_mul_ps( mQ, approxRsqrt ); } +inline void LLVector4a::normalize3fast_checked(LLVector4a* d) +{ + if (!isFinite3()) + { + *this = d ? *d : LLVector4a(0,1,0,1); + return; + } + + LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); + + if (lenSqrd.getF32ptr()[0] <= FLT_EPSILON) + { + *this = d ? *d : LLVector4a(0,1,0,1); + return; + } + + const LLQuad approxRsqrt = _mm_rsqrt_ps(lenSqrd.mQ); + mQ = _mm_mul_ps( mQ, approxRsqrt ); +} + // Return true if this vector is normalized with respect to x,y,z up to tolerance inline LLBool32 LLVector4a::isNormalized3( F32 tolerance ) const { diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 14cebfe5aa..58cac57e7f 100755 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4472,6 +4472,9 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, continue; //skip degenerate face } + LLVector4a default_norm; + default_norm.set(0,1,0,1); + //for each edge for (S32 k = 0; k < 3; k++) { S32 index = face.mEdge[j*3+k]; @@ -4493,14 +4496,14 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, norm_mat.rotate(n[v1], t); - t.normalize3fast(); + t.normalize3fast_checked(&default_norm); normals.push_back(LLVector3(t[0], t[1], t[2])); mat.affineTransform(v[v2], t); vertices.push_back(LLVector3(t[0], t[1], t[2])); norm_mat.rotate(n[v2], t); - t.normalize3fast(); + t.normalize3fast_checked(&default_norm); normals.push_back(LLVector3(t[0], t[1], t[2])); } } @@ -6096,6 +6099,9 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) { VertexData corners[4]; VertexData baseVert; + LLVector4a default_norm; + default_norm.set(0,1,0,1); + for(S32 t = 0; t < 4; t++) { corners[t].getPosition().load3( mesh[offset + (grid_size*t)].mPos.mV); @@ -6108,8 +6114,8 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) lhs.setSub(corners[1].getPosition(), corners[0].getPosition()); LLVector4a rhs; rhs.setSub(corners[2].getPosition(), corners[1].getPosition()); - baseVert.getNormal().setCross3(lhs, rhs); - baseVert.getNormal().normalize3fast(); + baseVert.getNormal().setCross3(lhs, rhs); + baseVert.getNormal().normalize3fast_checked(&default_norm); } if(!(mTypeMask & TOP_MASK)) @@ -6559,17 +6565,12 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) d1.setSub(mPositions[mIndices[2]], mPositions[mIndices[0]]); LLVector4a normal; - normal.setCross3(d0,d1); - - if (normal.dot3(normal).getF32() > F_APPROXIMATELY_ZERO) - { - normal.normalize3fast(); - } - else - { //degenerate, make up a value - normal.set(0,0,1); - } + LLVector4a default_norm; + default_norm.set(0,1,0,1); + normal.setCross3(d0,d1); + normal.normalize3fast_checked(&default_norm); + llassert(llfinite(normal.getF32ptr()[0])); llassert(llfinite(normal.getF32ptr()[1])); llassert(llfinite(normal.getF32ptr()[2])); @@ -6611,11 +6612,13 @@ void LLVolumeFace::createTangents() CalculateTangentArray(mNumVertices, mPositions, mNormals, mTexCoords, mNumIndices/3, mIndices, mTangents); //normalize tangents + LLVector4a default_norm; + default_norm.set(0,1,0,1); for (U32 i = 0; i < mNumVertices; i++) { //binorm[i].normalize3fast(); //bump map/planar projection code requires normals to be normalized - mNormals[i].normalize3fast(); + mNormals[i].normalize3fast_checked(&default_norm); } } } @@ -6793,6 +6796,9 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat mat.loadu(mat_in); norm_mat.loadu(norm_mat_in); + LLVector4a default_norm; + default_norm.set(0,1,0,1); + for (U32 i = 0; i < face.mNumVertices; ++i) { //transform appended face position and store @@ -6800,7 +6806,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat //transform appended face normal and store norm_mat.rotate(src_norm[i], dst_norm[i]); - dst_norm[i].normalize3fast(); + dst_norm[i].normalize3fast_checked(&default_norm); //copy appended face texture coordinate dst_tc[i] = src_tc[i]; @@ -7213,42 +7219,41 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal, const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent) { - //LLVector4a *tan1 = new LLVector4a[vertexCount * 2]; LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a)); - LLVector4a* tan2 = tan1 + vertexCount; + LLVector4a* tan2 = tan1 + vertexCount; memset(tan1, 0, vertexCount*2*sizeof(LLVector4a)); - for (U32 a = 0; a < triangleCount; a++) - { - U32 i1 = *index_array++; - U32 i2 = *index_array++; - U32 i3 = *index_array++; + for (U32 a = 0; a < triangleCount; a++) + { + U32 i1 = *index_array++; + U32 i2 = *index_array++; + U32 i3 = *index_array++; - const LLVector4a& v1 = vertex[i1]; - const LLVector4a& v2 = vertex[i2]; - const LLVector4a& v3 = vertex[i3]; + const LLVector4a& v1 = vertex[i1]; + const LLVector4a& v2 = vertex[i2]; + const LLVector4a& v3 = vertex[i3]; - const LLVector2& w1 = texcoord[i1]; - const LLVector2& w2 = texcoord[i2]; - const LLVector2& w3 = texcoord[i3]; + const LLVector2& w1 = texcoord[i1]; + const LLVector2& w2 = texcoord[i2]; + const LLVector2& w3 = texcoord[i3]; const F32* v1ptr = v1.getF32ptr(); const F32* v2ptr = v2.getF32ptr(); const F32* v3ptr = v3.getF32ptr(); - float x1 = v2ptr[0] - v1ptr[0]; - float x2 = v3ptr[0] - v1ptr[0]; - float y1 = v2ptr[1] - v1ptr[1]; - float y2 = v3ptr[1] - v1ptr[1]; - float z1 = v2ptr[2] - v1ptr[2]; - float z2 = v3ptr[2] - v1ptr[2]; + float x1 = v2ptr[0] - v1ptr[0]; + float x2 = v3ptr[0] - v1ptr[0]; + float y1 = v2ptr[1] - v1ptr[1]; + float y2 = v3ptr[1] - v1ptr[1]; + float z1 = v2ptr[2] - v1ptr[2]; + float z2 = v3ptr[2] - v1ptr[2]; - float s1 = w2.mV[0] - w1.mV[0]; - float s2 = w3.mV[0] - w1.mV[0]; - float t1 = w2.mV[1] - w1.mV[1]; - float t2 = w3.mV[1] - w1.mV[1]; + float s1 = w2.mV[0] - w1.mV[0]; + float s2 = w3.mV[0] - w1.mV[0]; + float t1 = w2.mV[1] - w1.mV[1]; + float t2 = w3.mV[1] - w1.mV[1]; F32 rd = s1*t2-s2*t1; diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp index 6e22712b94..09ef780424 100755 --- a/indra/llrender/llrendertarget.cpp +++ b/indra/llrender/llrendertarget.cpp @@ -572,8 +572,10 @@ void LLRenderTarget::copyContentsToFramebuffer(LLRenderTarget& source, S32 srcX0 { if (!source.mFBO) { - llerrs << "Cannot copy framebuffer contents for non FBO render targets." << llendl; + llwarns << "Cannot copy framebuffer contents for non FBO render targets." << llendl; + return; } + { GLboolean write_depth = mask & GL_DEPTH_BUFFER_BIT ? TRUE : FALSE; diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index 344079b640..ebb01fb330 100755 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -8592,6 +8592,18 @@ </array> </map> + <key>RenderSpecularPrecision</key> + <map> + <key>Comment</key> + <string>Force 32-bit floating point LUT</string> + <key>Persist</key> + <integer>1</integer> + <key>Type</key> + <string>U32</string> + <key>Value</key> + <real>0</real> + </map> + <key>RenderSpecularResX</key> <map> <key>Comment</key> diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index f021f4ed0f..700b31f8d3 100755 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -817,12 +817,6 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f, size.mul(scale); } - // Catch potential badness from normalization before it happens - // - llassert(mat_normal.mMatrix[0].isFinite3() && (mat_normal.mMatrix[0].dot3(mat_normal.mMatrix[0]).getF32() > F_APPROXIMATELY_ZERO)); - llassert(mat_normal.mMatrix[1].isFinite3() && (mat_normal.mMatrix[1].dot3(mat_normal.mMatrix[1]).getF32() > F_APPROXIMATELY_ZERO)); - llassert(mat_normal.mMatrix[2].isFinite3() && (mat_normal.mMatrix[2].dot3(mat_normal.mMatrix[2]).getF32() > F_APPROXIMATELY_ZERO)); - mat_normal.mMatrix[0].normalize3fast(); mat_normal.mMatrix[1].normalize3fast(); mat_normal.mMatrix[2].normalize3fast(); diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index d37b086ae3..442b0b6d2d 100755 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -1201,7 +1201,15 @@ void LLPipeline::releaseLUTBuffers() { if (mLightFunc) { - LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, GL_R16F, 0, 1, &mLightFunc); + U32 use_high_precision = gSavedSettings.getU32("RenderSpecularPrecision"); + + U32 pix_format = use_high_precision ? GL_R32F : GL_R16F; + +#if LL_DARWIN + pix_format = GL_R32F; +#endif + + LLImageGL::deleteTextures(LLTexUnit::TT_TEXTURE, pix_format, 0, 1, &mLightFunc); mLightFunc = 0; } } @@ -1400,13 +1408,18 @@ void LLPipeline::createLUTBuffers() ls[y*lightResX+x] = k;*/ } } - - U32 pix_format = GL_R16F; -#if LL_DARWIN // Need to work around limited precision with 10.6.8 and older drivers // + U32 use_high_precision = gSavedSettings.getU32("RenderSpecularPrecision"); + + U32 pix_format = use_high_precision ? GL_R32F : GL_R16F; + +#if LL_DARWIN + // Forced to work around 10.6.8. driver bugs on most every GPU + // pix_format = GL_R32F; #endif + LLImageGL::generateTextures(LLTexUnit::TT_TEXTURE, pix_format, 1, &mLightFunc); gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mLightFunc); LLImageGL::setManualImage(LLTexUnit::getInternalType(LLTexUnit::TT_TEXTURE), 0, pix_format, lightResX, lightResY, GL_RED, GL_FLOAT, ls, false); |