diff options
author | Dave Parks <davep@lindenlab.com> | 2021-09-20 18:58:09 +0000 |
---|---|---|
committer | Dave Parks <davep@lindenlab.com> | 2021-09-20 18:58:09 +0000 |
commit | f79890669dcf8e44b5ec3ce1abbd1d1fdd34eb3b (patch) | |
tree | 749f11a7763fbe38a2ecfc51479db23faecaaf0d /indra | |
parent | c09532ac13a3ca2f1ca0854cf00594a26ddcb593 (diff) |
SL-16006 and SL-16009 Rigged mesh rendering optimization pass
Diffstat (limited to 'indra')
25 files changed, 521 insertions, 354 deletions
diff --git a/indra/llcharacter/lljoint.cpp b/indra/llcharacter/lljoint.cpp index dee642310e..d72282ab42 100644 --- a/indra/llcharacter/lljoint.cpp +++ b/indra/llcharacter/lljoint.cpp @@ -922,6 +922,13 @@ const LLMatrix4 &LLJoint::getWorldMatrix() return mXform.getWorldMatrix(); } +const LLMatrix4a& LLJoint::getWorldMatrix4a() +{ + updateWorldMatrixParent(); + + return mWorldMatrix; +} + //-------------------------------------------------------------------- // setWorldMatrix() @@ -1003,6 +1010,7 @@ void LLJoint::updateWorldMatrix() { sNumUpdates++; mXform.updateMatrix(FALSE); + mWorldMatrix.loadu(mXform.getWorldMatrix()); mDirtyFlags = 0x0; } } diff --git a/indra/llcharacter/lljoint.h b/indra/llcharacter/lljoint.h index 1b646b641f..ba821667c7 100644 --- a/indra/llcharacter/lljoint.h +++ b/indra/llcharacter/lljoint.h @@ -38,6 +38,7 @@ #include "m4math.h" #include "llquaternion.h" #include "xform.h" +#include "llmatrix4a.h" const S32 LL_CHARACTER_MAX_JOINTS_PER_MESH = 15; // Need to set this to count of animate-able joints, @@ -123,6 +124,7 @@ protected: // explicit transformation members LLXformMatrix mXform; + LLMatrix4a mWorldMatrix; LLVector3 mDefaultPosition; LLVector3 mDefaultScale; @@ -259,6 +261,8 @@ public: const LLMatrix4 &getWorldMatrix(); void setWorldMatrix( const LLMatrix4& mat ); + const LLMatrix4a& getWorldMatrix4a(); + void updateWorldMatrixChildren(); void updateWorldMatrixParent(); diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h index 7ba347062f..5291a05607 100644 --- a/indra/llmath/llmatrix4a.h +++ b/indra/llmath/llmatrix4a.h @@ -36,6 +36,26 @@ class LLMatrix4a public: LL_ALIGN_16(LLVector4a mMatrix[4]); + LLMatrix4a() + { + + } + + explicit LLMatrix4a(const LLMatrix4& val) + { + loadu(val); + } + + inline F32* getF32ptr() + { + return (F32*) &mMatrix; + } + + inline const F32* getF32ptr() const + { + return (F32*)&mMatrix; + } + inline void clear() { mMatrix[0].clear(); @@ -44,6 +64,14 @@ public: mMatrix[3].clear(); } + inline void setIdentity() + { + mMatrix[0].set(1.f, 0.f, 0.f, 0.f); + mMatrix[1].set(0.f, 1.f, 0.f, 0.f); + mMatrix[2].set(0.f, 0.f, 1.f, 0.f); + mMatrix[3].set(0.f, 0.f, 0.f, 1.f); + } + inline void loadu(const LLMatrix4& src) { mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]); @@ -105,7 +133,7 @@ public: mMatrix[3].setAdd(a.mMatrix[3],d3); } - inline void rotate(const LLVector4a& v, LLVector4a& res) + inline void rotate(const LLVector4a& v, LLVector4a& res) const { LLVector4a y,z; @@ -151,6 +179,8 @@ public: { affineTransformSSE(v,res); } + + const LLVector4a& getTranslation() const { return mMatrix[3]; } }; inline LLVector4a rowMul(const LLVector4a &row, const LLMatrix4a &mat) @@ -176,6 +206,15 @@ inline void matMul(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res) res.mMatrix[3] = row3; } +//Faster version of matMul wehere res must not be a or b +inline void matMulUnsafe(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res) +{ + res.mMatrix[0] = rowMul(a.mMatrix[0], b); + res.mMatrix[1] = rowMul(a.mMatrix[1], b); + res.mMatrix[2] = rowMul(a.mMatrix[2], b); + res.mMatrix[3] = rowMul(a.mMatrix[3], b); +} + inline std::ostream& operator<<(std::ostream& s, const LLMatrix4a& m) { s << "[" << m.mMatrix[0] << ", " << m.mMatrix[1] << ", " << m.mMatrix[2] << ", " << m.mMatrix[3] << "]"; diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h index 27abf39537..5a02928374 100644 --- a/indra/llmath/llvector4a.h +++ b/indra/llmath/llvector4a.h @@ -46,10 +46,9 @@ class LLRotation; // of this writing, July 08, 2010) about getting it implemented before you resort to // LLVector3/LLVector4. ///////////////////////////////// -struct LLVector4a; LL_ALIGN_PREFIX(16) -struct LLVector4a +class LLVector4a { public: @@ -138,10 +137,10 @@ public: // BASIC GET/SET //////////////////////////////////// - // Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + // Return a "this" as an F32 pointer. inline F32* getF32ptr(); - // Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + // Return a "this" as a const F32 pointer. inline const F32* const getF32ptr() const; // Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl index 69d3d01efe..8be1c1b114 100644 --- a/indra/llmath/llvector4a.inl +++ b/indra/llmath/llvector4a.inl @@ -58,13 +58,13 @@ inline void LLVector4a::store4a(F32* dst) const // BASIC GET/SET //////////////////////////////////// -// Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +// Return a "this" as an F32 pointer. F32* LLVector4a::getF32ptr() { return (F32*) &mQ; } -// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +// Return a "this" as a const F32 pointer. const F32* const LLVector4a::getF32ptr() const { return (const F32* const) &mQ; diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp index 3baf1bad18..6e40dae30b 100644 --- a/indra/llmath/m4math.cpp +++ b/indra/llmath/m4math.cpp @@ -32,8 +32,7 @@ #include "m4math.h" #include "m3math.h" #include "llquaternion.h" - - +#include "llmatrix4a.h" // LLMatrix4 @@ -115,6 +114,12 @@ LLMatrix4::LLMatrix4(const LLQuaternion &q) *this = initRotation(q); } +LLMatrix4::LLMatrix4(const LLMatrix4a& mat) + : LLMatrix4(mat.getF32ptr()) +{ + +} + LLMatrix4::LLMatrix4(const LLQuaternion &q, const LLVector4 &pos) { *this = initRotTrans(q, pos); diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h index bf60adb9b6..b9da970cde 100644 --- a/indra/llmath/m4math.h +++ b/indra/llmath/m4math.h @@ -32,6 +32,7 @@ class LLVector4; class LLMatrix3; class LLQuaternion; +class LLMatrix4a; // NOTA BENE: Currently assuming a right-handed, x-forward, y-left, z-up universe @@ -104,6 +105,7 @@ public: explicit LLMatrix4(const F32 *mat); // Initializes Matrix to values in mat explicit LLMatrix4(const LLMatrix3 &mat); // Initializes Matrix to values in mat and sets position to (0,0,0) explicit LLMatrix4(const LLQuaternion &q); // Initializes Matrix with rotation q and sets position to (0,0,0) + explicit LLMatrix4(const LLMatrix4a& mat); LLMatrix4(const LLMatrix3 &mat, const LLVector4 &pos); // Initializes Matrix to values in mat and pos diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp index b04c67d926..93010d2250 100644 --- a/indra/llmath/v3math.cpp +++ b/indra/llmath/v3math.cpp @@ -316,6 +316,12 @@ LLVector3::LLVector3(const LLVector4 &vec) mV[VZ] = (F32)vec.mV[VZ]; } +LLVector3::LLVector3(const LLVector4a& vec) + : LLVector3(vec.getF32ptr()) +{ + +} + LLVector3::LLVector3(const LLSD& sd) { setValue(sd); diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h index 6f857d7061..068f489020 100644 --- a/indra/llmath/v3math.h +++ b/indra/llmath/v3math.h @@ -33,6 +33,7 @@ #include "llsd.h" class LLVector2; class LLVector4; +class LLVector4a; class LLMatrix3; class LLMatrix4; class LLVector3d; @@ -62,7 +63,9 @@ class LLVector3 explicit LLVector3(const LLVector2 &vec); // Initializes LLVector3 to (vec[0]. vec[1], 0) explicit LLVector3(const LLVector3d &vec); // Initializes LLVector3 to (vec[0]. vec[1], vec[2]) explicit LLVector3(const LLVector4 &vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2]) - explicit LLVector3(const LLSD& sd); + explicit LLVector3(const LLVector4a& vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2]) + explicit LLVector3(const LLSD& sd); + LLSD getValue() const; diff --git a/indra/llprimitive/lldaeloader.cpp b/indra/llprimitive/lldaeloader.cpp index dfa29fb539..8343de0cbc 100644 --- a/indra/llprimitive/lldaeloader.cpp +++ b/indra/llprimitive/lldaeloader.cpp @@ -1173,17 +1173,19 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do LLMeshSkinInfo& skin_info = model->mSkinInfo; + LLMatrix4 mat; for (int i = 0; i < 4; i++) { for(int j = 0; j < 4; j++) { - skin_info.mBindShapeMatrix.mMatrix[i][j] = dom_value[i + j*4]; + mat.mMatrix[i][j] = dom_value[i + j*4]; } } - LLMatrix4 trans = normalized_transformation; - trans *= skin_info.mBindShapeMatrix; - skin_info.mBindShapeMatrix = trans; + skin_info.mBindShapeMatrix.loadu(mat); + + LLMatrix4a trans(normalized_transformation); + matMul(trans, skin_info.mBindShapeMatrix, skin_info.mBindShapeMatrix); } @@ -1401,7 +1403,7 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do mat.mMatrix[i][j] = transform[k*16 + i + j*4]; } } - model->mSkinInfo.mInvBindMatrix.push_back(mat); + model->mSkinInfo.mInvBindMatrix.push_back(LLMatrix4a(mat)); } } } @@ -1475,9 +1477,9 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do if (mJointMap.find(lookingForJoint) != mJointMap.end() && model->mSkinInfo.mInvBindMatrix.size() > i) { - LLMatrix4 newInverse = model->mSkinInfo.mInvBindMatrix[i]; + LLMatrix4 newInverse = LLMatrix4(model->mSkinInfo.mInvBindMatrix[i].getF32ptr()); newInverse.setTranslation( mJointList[lookingForJoint].getTranslation() ); - model->mSkinInfo.mAlternateBindMatrix.push_back( newInverse ); + model->mSkinInfo.mAlternateBindMatrix.push_back( LLMatrix4a(newInverse) ); } else { diff --git a/indra/llprimitive/llmodel.cpp b/indra/llprimitive/llmodel.cpp index 702a1b5238..a23b991f1d 100644 --- a/indra/llprimitive/llmodel.cpp +++ b/indra/llprimitive/llmodel.cpp @@ -1396,7 +1396,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) } } - mInvBindMatrix.push_back(mat); + mInvBindMatrix.push_back(LLMatrix4a(mat)); } if (mJointNames.size() != mInvBindMatrix.size()) @@ -1410,13 +1410,15 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) if (skin.has("bind_shape_matrix")) { + LLMatrix4 mat; for (U32 j = 0; j < 4; j++) { for (U32 k = 0; k < 4; k++) { - mBindShapeMatrix.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal(); + mat.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal(); } } + mBindShapeMatrix.loadu(mat); } if (skin.has("alt_inverse_bind_matrix")) @@ -1432,7 +1434,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin) } } - mAlternateBindMatrix.push_back(mat); + mAlternateBindMatrix.push_back(LLMatrix4a(mat)); } } diff --git a/indra/llprimitive/llmodel.h b/indra/llprimitive/llmodel.h index 51fa2f8079..96d4582b4f 100644 --- a/indra/llprimitive/llmodel.h +++ b/indra/llprimitive/llmodel.h @@ -33,6 +33,8 @@ #include "m4math.h" #include <queue> +#include <boost/align/aligned_allocator.hpp> + class daeElement; class domMesh; @@ -49,10 +51,11 @@ public: LLUUID mMeshID; std::vector<std::string> mJointNames; mutable std::vector<S32> mJointNums; - std::vector<LLMatrix4> mInvBindMatrix; - std::vector<LLMatrix4> mAlternateBindMatrix; + typedef std::vector<LLMatrix4a, boost::alignment::aligned_allocator<LLMatrix4a, 16>> matrix_list_t; + matrix_list_t mInvBindMatrix; + matrix_list_t mAlternateBindMatrix; - LLMatrix4 mBindShapeMatrix; + LLMatrix4a mBindShapeMatrix; float mPelvisOffset; bool mLockScaleIfJointPosition; bool mInvalidJointsScrubbed; diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 4351f6e2c8..8bd9dbf9b8 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -1346,6 +1346,7 @@ void LLGLSLShader::uniformMatrix3fv(U32 index, U32 count, GLboolean transpose, c void LLGLSLShader::uniformMatrix3x4fv(U32 index, U32 count, GLboolean transpose, const GLfloat *v) { + LL_PROFILE_ZONE_SCOPED; if (mProgramObject) { if (mUniform.size() <= index) diff --git a/indra/newview/llcontrolavatar.cpp b/indra/newview/llcontrolavatar.cpp index fab249f988..606e670805 100644 --- a/indra/newview/llcontrolavatar.cpp +++ b/indra/newview/llcontrolavatar.cpp @@ -241,7 +241,7 @@ void LLControlAvatar::matchVolumeTransform() if (skin_info) { LL_DEBUGS("BindShape") << getFullname() << " bind shape " << skin_info->mBindShapeMatrix << LL_ENDL; - bind_rot = LLSkinningUtil::getUnscaledQuaternion(skin_info->mBindShapeMatrix); + bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(skin_info->mBindShapeMatrix)); } #endif setRotation(bind_rot*obj_rot); diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 6c4844f9ee..c7aa104ca5 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -163,6 +163,7 @@ void LLDrawPoolAvatar::prerender() { LLVOAvatar* avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); updateRiggedVertexBuffers(avatarp); + updateSkinInfoMatrixPalettes(avatarp); } } } @@ -428,221 +429,230 @@ S32 LLDrawPoolAvatar::getNumShadowPasses() void LLDrawPoolAvatar::beginShadowPass(S32 pass) { LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (pass == SHADOW_PASS_AVATAR_OPAQUE) - { - sVertexProgram = &gDeferredAvatarShadowProgram; - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + if (pass == SHADOW_PASS_AVATAR_OPAQUE) + { + sVertexProgram = &gDeferredAvatarShadowProgram; - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - sVertexProgram = &gDeferredAvatarAlphaShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAvatarAlphaShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) - { - sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) - { - sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; - gGL.diffuseColor4f(1,1,1,1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) - { - sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } + + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - gGL.diffuseColor4f(1,1,1,1); - } - else // SHADOW_PASS_ATTACHMENT_OPAQUE - { - sVertexProgram = &gDeferredAttachmentShadowProgram; - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } + + gGL.diffuseColor4f(1, 1, 1, 1); + } + else // SHADOW_PASS_ATTACHMENT_OPAQUE { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } - sVertexProgram->bind(); - } + sVertexProgram = &gDeferredAttachmentShadowProgram; + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } + sVertexProgram->bind(); + } + } } void LLDrawPoolAvatar::endShadowPass(S32 pass) { LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) - { - LLVertexBuffer::unbind(); - } + if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) + { + LLVertexBuffer::unbind(); + } - if (sShaderLevel > 0) - { - sVertexProgram->unbind(); - } - sVertexProgram = NULL; - sRenderingSkinned = FALSE; - LLDrawPoolAvatar::sShadowPass = -1; + if (sShaderLevel > 0) + { + sVertexProgram->unbind(); + } + sVertexProgram = NULL; + sRenderingSkinned = FALSE; + LLDrawPoolAvatar::sShadowPass = -1; + } } void LLDrawPoolAvatar::renderShadow(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + { + LL_PROFILE_ZONE_SCOPED; - if (mDrawFace.empty()) - { - return; - } + if (mDrawFace.empty()) + { + return; + } - const LLFace *facep = mDrawFace[0]; - if (!facep->getDrawable()) - { - return; - } - LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); + const LLFace *facep = mDrawFace[0]; + if (!facep->getDrawable()) + { + return; + } + LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); - if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) - { - return; - } - LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); - BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); - if (oa == LLVOAvatar::AOA_INVISIBLE || - (impostor && oa == LLVOAvatar::AOA_JELLYDOLL)) - { - // No shadows for jellydolled or invisible avs. - return; - } - - LLDrawPoolAvatar::sShadowPass = pass; + if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) + { + return; + } + LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); + BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); + if (oa == LLVOAvatar::AOA_INVISIBLE || + (impostor && oa == LLVOAvatar::AOA_JELLYDOLL)) + { + // No shadows for jellydolled or invisible avs. + return; + } - if (pass == SHADOW_PASS_AVATAR_OPAQUE) - { - LLDrawPoolAvatar::sSkipTransparent = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipTransparent = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); - renderRigged(avatarp, RIGGED_ALPHA); - renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_SPECMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); - renderRigged(avatarp, RIGGED_NORMMAP_MASK); - renderRigged(avatarp, RIGGED_SPECMAP_MASK); - renderRigged(avatarp, RIGGED_NORMSPEC_MASK); - renderRigged(avatarp, RIGGED_GLOW); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE - { - LLDrawPoolAvatar::sSkipTransparent = true; - renderRigged(avatarp, RIGGED_MATERIAL); - renderRigged(avatarp, RIGGED_SPECMAP); - renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMMAP); - renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMSPEC); - renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); - renderRigged(avatarp, RIGGED_SIMPLE); - renderRigged(avatarp, RIGGED_FULLBRIGHT); - renderRigged(avatarp, RIGGED_SHINY); - renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_DEFERRED_BUMP); - renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); - LLDrawPoolAvatar::sSkipTransparent = false; - } + LLDrawPoolAvatar::sShadowPass = pass; + + if (pass == SHADOW_PASS_AVATAR_OPAQUE) + { + LLDrawPoolAvatar::sSkipTransparent = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipTransparent = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); + renderRigged(avatarp, RIGGED_ALPHA); + renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_SPECMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); + renderRigged(avatarp, RIGGED_NORMMAP_MASK); + renderRigged(avatarp, RIGGED_SPECMAP_MASK); + renderRigged(avatarp, RIGGED_NORMSPEC_MASK); + renderRigged(avatarp, RIGGED_GLOW); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE + { + LLDrawPoolAvatar::sSkipTransparent = true; + renderRigged(avatarp, RIGGED_MATERIAL); + renderRigged(avatarp, RIGGED_SPECMAP); + renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMMAP); + renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMSPEC); + renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); + renderRigged(avatarp, RIGGED_SIMPLE); + renderRigged(avatarp, RIGGED_FULLBRIGHT); + renderRigged(avatarp, RIGGED_SHINY); + renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_DEFERRED_BUMP); + renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); + LLDrawPoolAvatar::sSkipTransparent = false; + } + } } S32 LLDrawPoolAvatar::getNumPasses() @@ -1794,7 +1804,7 @@ void LLDrawPoolAvatar::getRiggedGeometry( U16 offset = 0; - LLMatrix4 mat_vert = skin->mBindShapeMatrix; + LLMatrix4 mat_vert = LLMatrix4(skin->mBindShapeMatrix); glh::matrix4f m((F32*) mat_vert.mMatrix); m = m.inverse().transpose(); @@ -1968,14 +1978,11 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( skin = vobj->getSkinInfo(); } - //build matrix palette - LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - U32 count = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar); - LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin); + const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin); + const LLMatrix4a* mat = &(mpc.mMatrixPalette[0]); - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + LLSkinningUtil::checkSkinWeights(weights, buffer->getNumVerts(), skin); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS U8* joint_indices_cursor = vol_face.mJointIndices; @@ -2043,6 +2050,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) { + LL_PROFILE_ZONE_SCOPED; + if (!avatar->shouldRenderRigged()) { return; @@ -2050,15 +2059,18 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) stop_glerror(); + const LLMeshSkinInfo* lastSkin = nullptr; + for (U32 i = 0; i < mRiggedFace[type].size(); ++i) { + LL_PROFILE_ZONE_NAMED("Render Rigged Face"); LLFace* face = mRiggedFace[type][i]; S32 offset = face->getIndicesStart(); U32 count = face->getIndicesCount(); U16 start = face->getGeomStart(); - U16 end = start + face->getGeomCount()-1; + U16 end = start + face->getGeomCount()-1; LLDrawable* drawable = face->getDrawable(); if (!drawable) @@ -2180,52 +2192,32 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) } if (buff) - { + { if (sShaderLevel > 0) { - // upload matrix palette to shader - LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - U32 count = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, count, skin, avatar); - - stop_glerror(); - - F32 mp[LL_MAX_JOINTS_PER_MESH_OBJECT*12]; - - for (U32 i = 0; i < count; ++i) - { - F32* m = (F32*) mat[i].mMatrix[0].getF32ptr(); - - U32 idx = i*12; - - mp[idx+0] = m[0]; - mp[idx+1] = m[1]; - mp[idx+2] = m[2]; - mp[idx+3] = m[12]; - - mp[idx+4] = m[4]; - mp[idx+5] = m[5]; - mp[idx+6] = m[6]; - mp[idx+7] = m[13]; + if (lastSkin != skin) // <== only upload matrix palette to GL if the skininfo changed + { + // upload matrix palette to shader + const MatrixPaletteCache& mpc = updateSkinInfoMatrixPalette(avatar, skin); + U32 count = mpc.mMatrixPalette.size(); - mp[idx+8] = m[8]; - mp[idx+9] = m[9]; - mp[idx+10] = m[10]; - mp[idx+11] = m[14]; - } + stop_glerror(); - LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX, - count, - FALSE, - (GLfloat*) mp); + LLDrawPoolAvatar::sVertexProgram->uniformMatrix3x4fv(LLViewerShaderMgr::AVATAR_MATRIX, + count, + FALSE, + (GLfloat*) &(mpc.mGLMp[0])); - stop_glerror(); + stop_glerror(); + } } else { data_mask &= ~LLVertexBuffer::MAP_WEIGHT4; } + lastSkin = skin; + /*if (glow) { gGL.diffuseColor4f(0,0,0,face->getTextureEntry()->getGlow()); @@ -2391,6 +2383,70 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) } } +void LLDrawPoolAvatar::updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp) +{ + LL_PROFILE_ZONE_SCOPED; + //evict matrix palettes from the cache that haven't been updated in 10 frames + for (matrix_palette_cache_t::iterator iter = mMatrixPaletteCache.begin(); iter != mMatrixPaletteCache.end(); ) + { + if (gFrameCount - iter->second.mFrame > 10) + { + iter = mMatrixPaletteCache.erase(iter); + } + else + { + ++iter; + } + } +} + +const LLDrawPoolAvatar::MatrixPaletteCache& LLDrawPoolAvatar::updateSkinInfoMatrixPalette(LLVOAvatar * avatarp, const LLMeshSkinInfo* skin) +{ + MatrixPaletteCache& entry = mMatrixPaletteCache[skin]; + + if (entry.mFrame != gFrameCount) + { + LL_PROFILE_ZONE_SCOPED; + entry.mFrame = gFrameCount; + //build matrix palette + U32 count = LLSkinningUtil::getMeshJointCount(skin); + entry.mMatrixPalette.resize(count); + LLSkinningUtil::initSkinningMatrixPalette(&(entry.mMatrixPalette[0]), count, skin, avatarp); + + const LLMatrix4a* mat = &(entry.mMatrixPalette[0]); + + stop_glerror(); + + entry.mGLMp.resize(count * 12); + + F32* mp = &(entry.mGLMp[0]); + + for (U32 i = 0; i < count; ++i) + { + F32* m = (F32*)mat[i].mMatrix[0].getF32ptr(); + + U32 idx = i * 12; + + mp[idx + 0] = m[0]; + mp[idx + 1] = m[1]; + mp[idx + 2] = m[2]; + mp[idx + 3] = m[12]; + + mp[idx + 4] = m[4]; + mp[idx + 5] = m[5]; + mp[idx + 6] = m[6]; + mp[idx + 7] = m[13]; + + mp[idx + 8] = m[8]; + mp[idx + 9] = m[9]; + mp[idx + 10] = m[10]; + mp[idx + 11] = m[14]; + } + } + + return entry; +} + void LLDrawPoolAvatar::renderRiggedSimple(LLVOAvatar* avatar) { renderRigged(avatar, RIGGED_SIMPLE); diff --git a/indra/newview/lldrawpoolavatar.h b/indra/newview/lldrawpoolavatar.h index 9b26266ced..0c1ee2cced 100644 --- a/indra/newview/lldrawpoolavatar.h +++ b/indra/newview/lldrawpoolavatar.h @@ -28,15 +28,18 @@ #define LL_LLDRAWPOOLAVATAR_H #include "lldrawpool.h" +#include "llmodel.h" + +#include <unordered_map> class LLVOAvatar; class LLVOVolume; class LLGLSLShader; class LLFace; -class LLMeshSkinInfo; class LLVolume; class LLVolumeFace; +extern U32 gFrameCount; class LLDrawPoolAvatar : public LLFacePool { @@ -259,6 +262,8 @@ typedef enum LLVolumeFace& vol_face); void updateRiggedVertexBuffers(LLVOAvatar* avatar); + void updateSkinInfoMatrixPalettes(LLVOAvatar* avatarp); + void renderRigged(LLVOAvatar* avatar, U32 type, bool glow = false); void renderRiggedSimple(LLVOAvatar* avatar); void renderRiggedAlpha(LLVOAvatar* avatar); @@ -278,6 +283,26 @@ typedef enum std::vector<LLFace*> mRiggedFace[NUM_RIGGED_PASSES]; + class MatrixPaletteCache + { + public: + U32 mFrame; + LLMeshSkinInfo::matrix_list_t mMatrixPalette; + + // Float array ready to be sent to GL + std::vector<F32> mGLMp; + + MatrixPaletteCache() : + mFrame(gFrameCount-1) + { + } + }; + + const MatrixPaletteCache& updateSkinInfoMatrixPalette(LLVOAvatar* avatarp, const LLMeshSkinInfo* skin); + + typedef std::unordered_map<const LLMeshSkinInfo*, MatrixPaletteCache> matrix_palette_cache_t; + matrix_palette_cache_t mMatrixPaletteCache; + /*virtual*/ LLViewerTexture *getDebugTexture(); /*virtual*/ LLColor3 getDebugColor() const; // For AGP debug display diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp index d9edd4dc30..0e54b66ea9 100644 --- a/indra/newview/llfloatermodelpreview.cpp +++ b/indra/newview/llfloatermodelpreview.cpp @@ -1357,31 +1357,31 @@ void LLFloaterModelPreview::clearAvatarTab() } void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) - { +{ S32 display_lod = mModelPreview->mPreviewLOD; if (mModelPreview->mModel[display_lod].empty()) - { + { mSelectedJointName.clear(); return; - } + } // Joints will be listed as long as they are listed in mAlternateBindMatrix // even if they are for some reason identical to defaults. // Todo: Are overrides always identical for all lods? They normally are, but there might be situations where they aren't. if (mJointOverrides[display_lod].empty()) - { + { // populate map for (LLModelLoader::scene::iterator iter = mModelPreview->mScene[display_lod].begin(); iter != mModelPreview->mScene[display_lod].end(); ++iter) - { + { for (LLModelLoader::model_instance_list::iterator model_iter = iter->second.begin(); model_iter != iter->second.end(); ++model_iter) - { + { LLModelInstance& instance = *model_iter; LLModel* model = instance.mModel; const LLMeshSkinInfo *skin = &model->mSkinInfo; U32 joint_count = LLSkinningUtil::getMeshJointCount(skin); U32 bind_count = highlight_overrides ? skin->mAlternateBindMatrix.size() : 0; // simply do not include overrides if data is not needed if (bind_count > 0 && bind_count != joint_count) - { + { std::ostringstream out; out << "Invalid joint overrides for model " << model->getName(); out << ". Amount of joints " << joint_count; @@ -1390,68 +1390,68 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) addStringToLog(out.str(), true); // Disable overrides for this model bind_count = 0; - } + } if (bind_count > 0) - { + { for (U32 j = 0; j < joint_count; ++j) - { - const LLVector3& joint_pos = skin->mAlternateBindMatrix[j].getTranslation(); + { + const LLVector3& joint_pos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation()); LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]]; LLJoint* pJoint = LLModelPreview::lookupJointByName(skin->mJointNames[j], mModelPreview); if (pJoint) - { + { // see how voavatar uses aboveJointPosThreshold if (pJoint->aboveJointPosThreshold(joint_pos)) - { + { // valid override if (data.mPosOverrides.size() > 0 && (data.mPosOverrides.begin()->second - joint_pos).lengthSquared() > (LL_JOINT_TRESHOLD_POS_OFFSET * LL_JOINT_TRESHOLD_POS_OFFSET)) - { + { // File contains multiple meshes with conflicting joint offsets // preview may be incorrect, upload result might wary (depends onto // mesh_id that hasn't been generated yet). data.mHasConflicts = true; - } + } data.mPosOverrides[model->getName()] = joint_pos; - } - else - { + } + else + { // default value, it won't be accounted for by avatar data.mModelsNoOverrides.insert(model->getName()); - } - } - } - } - else - { + } + } + } + } + else + { for (U32 j = 0; j < joint_count; ++j) - { + { LLJointOverrideData &data = mJointOverrides[display_lod][skin->mJointNames[j]]; data.mModelsNoOverrides.insert(model->getName()); } } - } - } - } + } + } + } LLPanel *panel = mTabContainer->getPanelByName("rigging_panel"); LLScrollListCtrl *joints_list = panel->getChild<LLScrollListCtrl>("joints_list"); if (joints_list->isEmpty()) - { + { // Populate table - std::map<std::string, std::string> joint_alias_map; + std::map<std::string, std::string> joint_alias_map; mModelPreview->getJointAliases(joint_alias_map); - + S32 conflicts = 0; joint_override_data_map_t::iterator joint_iter = mJointOverrides[display_lod].begin(); joint_override_data_map_t::iterator joint_end = mJointOverrides[display_lod].end(); while (joint_iter != joint_end) - { + { const std::string& listName = joint_iter->first; - + LLScrollListItem::Params item_params; item_params.value(listName); @@ -1459,38 +1459,38 @@ void LLFloaterModelPreview::updateAvatarTab(bool highlight_overrides) cell_params.font = LLFontGL::getFontSansSerif(); cell_params.value = listName; if (joint_alias_map.find(listName) == joint_alias_map.end()) - { + { // Missing names cell_params.color = LLColor4::red; - } + } if (joint_iter->second.mHasConflicts) - { + { // Conflicts cell_params.color = LLColor4::orange; conflicts++; - } + } if (highlight_overrides && joint_iter->second.mPosOverrides.size() > 0) - { + { cell_params.font.style = "BOLD"; - } + } item_params.columns.add(cell_params); joints_list->addRow(item_params, ADD_BOTTOM); joint_iter++; - } + } joints_list->selectFirstItem(); LLScrollListItem *selected = joints_list->getFirstSelected(); if (selected) -{ + { mSelectedJointName = selected->getValue().asString(); - } + } LLTextBox *joint_conf_descr = panel->getChild<LLTextBox>("conflicts_description"); joint_conf_descr->setTextArg("[CONFLICTS]", llformat("%d", conflicts)); joint_conf_descr->setTextArg("[JOINTS_COUNT]", llformat("%d", mJointOverrides[display_lod].size())); - } - } + } +} //----------------------------------------------------------------------------- // addStringToLogTab() diff --git a/indra/newview/llmodelpreview.cpp b/indra/newview/llmodelpreview.cpp index a9e80ab5da..01bddd781d 100644 --- a/indra/newview/llmodelpreview.cpp +++ b/indra/newview/llmodelpreview.cpp @@ -591,7 +591,7 @@ void LLModelPreview::rebuildUploadData() bool upload_skinweights = fmp && fmp->childGetValue("upload_skin").asBoolean(); if (upload_skinweights && high_lod_model->mSkinInfo.mJointNames.size() > 0) { - LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(high_lod_model->mSkinInfo.mBindShapeMatrix); + LLQuaternion bind_rot = LLSkinningUtil::getUnscaledQuaternion(LLMatrix4(high_lod_model->mSkinInfo.mBindShapeMatrix)); LLQuaternion identity; if (!bind_rot.isEqualEps(identity, 0.01)) { @@ -3298,7 +3298,7 @@ BOOL LLModelPreview::render() LLJoint *joint = getPreviewAvatar()->getJoint(skin->mJointNums[j]); if (joint) { - const LLVector3& jointPos = skin->mAlternateBindMatrix[j].getTranslation(); + const LLVector3& jointPos = LLVector3(skin->mAlternateBindMatrix[j].getTranslation()); if (joint->aboveJointPosThreshold(jointPos)) { bool override_changed; @@ -3340,11 +3340,10 @@ BOOL LLModelPreview::render() //build matrix palette LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, joint_count, + LLSkinningUtil::initSkinningMatrixPalette(mat, joint_count, skin, getPreviewAvatar()); - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; U32 max_joints = LLSkinningUtil::getMaxJointCount(); for (U32 j = 0; j < buffer->getNumVerts(); ++j) { diff --git a/indra/newview/llskinningutil.cpp b/indra/newview/llskinningutil.cpp index e02b21f036..dc12de29fb 100644 --- a/indra/newview/llskinningutil.cpp +++ b/indra/newview/llskinningutil.cpp @@ -35,7 +35,6 @@ #include "llrigginginfo.h" #define DEBUG_SKINNING LL_DEBUG -#define MAT_USE_SSE 1 void dump_avatar_and_skin_state(const std::string& reason, LLVOAvatar *avatar, const LLMeshSkinInfo *skin) { @@ -120,36 +119,26 @@ void LLSkinningUtil::scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin skin->mInvalidJointsScrubbed = true; } -#define MAT_USE_SSE 1 - void LLSkinningUtil::initSkinningMatrixPalette( - LLMatrix4* mat, + LLMatrix4a* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar) { + LL_PROFILE_ZONE_SCOPED; + initJointNums(const_cast<LLMeshSkinInfo*>(skin), avatar); + + LLMatrix4a world[LL_CHARACTER_MAX_ANIMATED_JOINTS]; + for (U32 j = 0; j < count; ++j) { S32 joint_num = skin->mJointNums[j]; - LLJoint *joint = NULL; - if (joint_num >= 0 && joint_num < LL_CHARACTER_MAX_ANIMATED_JOINTS) - { - joint = avatar->getJoint(joint_num); - } - llassert(joint); + LLJoint *joint = avatar->getJoint(joint_num); + if (joint) { -#ifdef MAT_USE_SSE - LLMatrix4a bind, world, res; - bind.loadu(skin->mInvBindMatrix[j]); - world.loadu(joint->getWorldMatrix()); - matMul(bind,world,res); - memcpy(mat[j].mMatrix,res.mMatrix,16*sizeof(float)); -#else - mat[j] = skin->mInvBindMatrix[j]; - mat[j] *= joint->getWorldMatrix(); -#endif + world[j] = joint->getWorldMatrix4a(); } else { @@ -159,16 +148,27 @@ void LLSkinningUtil::initSkinningMatrixPalette( // rendering should be disabled unless all joints are // valid. In other cases of skinned rendering, invalid // joints should already have been removed during scrubInvalidJoints(). - LL_WARNS_ONCE("Avatar") << avatar->getFullname() - << " rigged to invalid joint name " << skin->mJointNames[j] - << " num " << skin->mJointNums[j] << LL_ENDL; - LL_WARNS_ONCE("Avatar") << avatar->getFullname() - << " avatar build state: isBuilt() " << avatar->isBuilt() - << " mInitFlags " << avatar->mInitFlags << LL_ENDL; + LL_WARNS_ONCE("Avatar") << avatar->getFullname() + << " rigged to invalid joint name " << skin->mJointNames[j] + << " num " << skin->mJointNums[j] << LL_ENDL; + LL_WARNS_ONCE("Avatar") << avatar->getFullname() + << " avatar build state: isBuilt() " << avatar->isBuilt() + << " mInitFlags " << avatar->mInitFlags << LL_ENDL; #endif dump_avatar_and_skin_state("initSkinningMatrixPalette joint not found", avatar, skin); } } + + //NOTE: pointer striders used here as a micro-optimization over vector/array lookups + const LLMatrix4a* invBind = &(skin->mInvBindMatrix[0]); + const LLMatrix4a* w = world; + LLMatrix4a* m = mat; + LLMatrix4a* end = m + count; + + while (m < end) + { + matMulUnsafe(*(invBind++), *(w++), *(m++)); + } } void LLSkinningUtil::checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin) @@ -212,7 +212,7 @@ void LLSkinningUtil::scrubSkinWeights(LLVector4a* weights, U32 num_vertices, con void LLSkinningUtil::getPerVertexSkinMatrix( F32* weights, - LLMatrix4a* mat, + const LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints) @@ -270,6 +270,7 @@ void LLSkinningUtil::initJointNums(LLMeshSkinInfo* skin, LLVOAvatar *avatar) { if (!skin->mJointNumsInitialized) { + LL_PROFILE_ZONE_SCOPED; for (U32 j = 0; j < skin->mJointNames.size(); ++j) { #if DEBUG_SKINNING @@ -357,13 +358,11 @@ void LLSkinningUtil::updateRiggingInfo(const LLMeshSkinInfo* skin, LLVOAvatar *a rig_info_tab[joint_num].setIsRiggedTo(true); // FIXME could precompute these matMuls. - LLMatrix4a bind_shape; - LLMatrix4a inv_bind; + const LLMatrix4a& bind_shape = skin->mBindShapeMatrix; + const LLMatrix4a& inv_bind = skin->mInvBindMatrix[joint_index]; LLMatrix4a mat; LLVector4a pos_joint_space; - bind_shape.loadu(skin->mBindShapeMatrix); - inv_bind.loadu(skin->mInvBindMatrix[joint_index]); matMul(bind_shape, inv_bind, mat); mat.affineTransform(pos, pos_joint_space); @@ -426,3 +425,4 @@ LLQuaternion LLSkinningUtil::getUnscaledQuaternion(const LLMatrix4& mat4) bind_rot.normalize(); return bind_rot; } + diff --git a/indra/newview/llskinningutil.h b/indra/newview/llskinningutil.h index efe7c85997..807418f983 100644 --- a/indra/newview/llskinningutil.h +++ b/indra/newview/llskinningutil.h @@ -42,10 +42,10 @@ namespace LLSkinningUtil S32 getMaxJointCount(); U32 getMeshJointCount(const LLMeshSkinInfo *skin); void scrubInvalidJoints(LLVOAvatar *avatar, LLMeshSkinInfo* skin); - void initSkinningMatrixPalette(LLMatrix4* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar); + void initSkinningMatrixPalette(LLMatrix4a* mat, S32 count, const LLMeshSkinInfo* skin, LLVOAvatar *avatar); void checkSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin); void scrubSkinWeights(LLVector4a* weights, U32 num_vertices, const LLMeshSkinInfo* skin); - void getPerVertexSkinMatrix(F32* weights, LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints); + void getPerVertexSkinMatrix(F32* weights, const LLMatrix4a* mat, bool handle_bad_scale, LLMatrix4a& final_mat, U32 max_joints); LL_FORCE_INLINE void getPerVertexSkinMatrixWithIndices( F32* weights, diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index ca01bb46aa..af55f68cd2 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -208,6 +208,7 @@ void LLViewerTextureManager::findTextures(const LLUUID& id, std::vector<LLViewe LLViewerFetchedTexture* LLViewerTextureManager::findFetchedTexture(const LLUUID& id, S32 tex_type) { + LL_PROFILE_ZONE_SCOPED; return gTextureList.findImage(id, (ETexListType)tex_type); } diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp index 561319ca5d..12495078e9 100644 --- a/indra/newview/llviewertexturelist.cpp +++ b/indra/newview/llviewertexturelist.cpp @@ -620,6 +620,7 @@ void LLViewerTextureList::findTexturesByID(const LLUUID &image_id, std::vector<L LLViewerFetchedTexture *LLViewerTextureList::findImage(const LLTextureKey &search_key) { + LL_PROFILE_ZONE_SCOPED; uuid_map_t::iterator iter = mUUIDMap.find(search_key); if (iter == mUUIDMap.end()) return NULL; diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index ba9f8d14cb..f260e4f750 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -4886,6 +4886,8 @@ bool LLVOAvatar::shouldAlphaMask() //----------------------------------------------------------------------------- U32 LLVOAvatar::renderSkinned() { + LL_PROFILE_ZONE_SCOPED; + U32 num_indices = 0; if (!mIsBuilt) @@ -6160,27 +6162,29 @@ LLJoint *LLVOAvatar::getJoint( const std::string &name ) LLJoint *LLVOAvatar::getJoint( S32 joint_num ) { LLJoint *pJoint = NULL; - S32 collision_start = mNumBones; - S32 attachment_start = mNumBones + mNumCollisionVolumes; - if (joint_num>=attachment_start) + if (joint_num >= 0) { - // Attachment IDs start at 1 - S32 attachment_id = joint_num - attachment_start + 1; - attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id); - if (iter != mAttachmentPoints.end()) + if (joint_num < mNumBones) { - pJoint = iter->second; + pJoint = mSkeleton[joint_num]; + } + else if (joint_num < mNumBones + mNumCollisionVolumes) + { + S32 collision_id = joint_num - mNumBones; + pJoint = &mCollisionVolumes[collision_id]; + } + else + { + // Attachment IDs start at 1 + S32 attachment_id = joint_num - (mNumBones + mNumCollisionVolumes) + 1; + attachment_map_t::iterator iter = mAttachmentPoints.find(attachment_id); + if (iter != mAttachmentPoints.end()) + { + pJoint = iter->second; + } } } - else if (joint_num>=collision_start) - { - S32 collision_id = joint_num-collision_start; - pJoint = &mCollisionVolumes[collision_id]; - } - else if (joint_num>=0) - { - pJoint = mSkeleton[joint_num]; - } + llassert(!pJoint || pJoint->getJointNum() == joint_num); return pJoint; } @@ -6515,7 +6519,7 @@ void LLVOAvatar::addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LL LLJoint* pJoint = getJoint( lookingForJoint ); if (pJoint) { - const LLVector3& jointPos = pSkinData->mAlternateBindMatrix[i].getTranslation(); + const LLVector3& jointPos = LLVector3(pSkinData->mAlternateBindMatrix[i].getTranslation()); if (pJoint->aboveJointPosThreshold(jointPos)) { bool override_changed; @@ -7864,6 +7868,8 @@ void LLVOAvatar::onGlobalColorChanged(const LLTexGlobalColor* global_color) // Do rigged mesh attachments display with this av? bool LLVOAvatar::shouldRenderRigged() const { + LL_PROFILE_ZONE_SCOPED; + if (getOverallAppearance() == AOA_NORMAL) { return true; @@ -10951,6 +10957,7 @@ void LLVOAvatar::updateOverallAppearanceAnimations() // Based on isVisuallyMuted(), but has 3 possible results. LLVOAvatar::AvatarOverallAppearance LLVOAvatar::getOverallAppearance() const { + LL_PROFILE_ZONE_SCOPED; AvatarOverallAppearance result = AOA_NORMAL; // Priority order (highest priority first) diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h index 74ef589ca4..39adaab8ca 100644 --- a/indra/newview/llvoavatar.h +++ b/indra/newview/llvoavatar.h @@ -209,6 +209,11 @@ public: virtual LLJoint* getJoint(const std::string &name); LLJoint* getJoint(S32 num); + //if you KNOW joint_num is a valid animated joint index, use getSkeletonJoint for efficiency + inline LLJoint* getSkeletonJoint(S32 joint_num) { return mSkeleton[joint_num]; } + inline size_t getSkeletonJointCount() const { return mSkeleton.size(); } + + void addAttachmentOverridesForObject(LLViewerObject *vo, std::set<LLUUID>* meshes_seen = NULL, bool recursive = true); void removeAttachmentOverridesForObject(const LLUUID& mesh_id); void removeAttachmentOverridesForObject(LLViewerObject *vo); diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index f9b2285989..e5a4b0f374 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -4807,7 +4807,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons LLMatrix4a mat[kMaxJoints]; U32 maxJoints = LLSkinningUtil::getMeshJointCount(skin); - LLSkinningUtil::initSkinningMatrixPalette((LLMatrix4*)mat, maxJoints, skin, avatar); + LLSkinningUtil::initSkinningMatrixPalette(mat, maxJoints, skin, avatar); S32 rigged_vert_count = 0; S32 rigged_face_count = 0; @@ -4823,8 +4823,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons if ( weight ) { LLSkinningUtil::checkSkinWeights(weight, dst_face.mNumVertices, skin); - LLMatrix4a bind_shape_matrix; - bind_shape_matrix.loadu(skin->mBindShapeMatrix); + const LLMatrix4a& bind_shape_matrix = skin->mBindShapeMatrix; LLVector4a* pos = dst_face.mPositions; |