#1530 Increase joint limit for GLTF Assets (#1582)

* Migrate GLTF scene rendering to its own shaders
* Add support for ambient occlusion map separate from metallic roughness map (or absent)
* Use UBO's for GLTF joints
* Better error handling of downloading GLTF assets

7 files changed, 52 insertions, 20 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index ac697b72be..0586c34e50 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -1238,6 +1238,11 @@ bool LLGLManager::initGL()
     glGetIntegerv(GL_MAX_INTEGER_SAMPLES, &mMaxIntegerSamples);
     glGetIntegerv(GL_MAX_SAMPLE_MASK_WORDS, &mMaxSampleMaskWords);
     glGetIntegerv(GL_MAX_SAMPLES, &mMaxSamples);
+    glGetIntegerv(GL_MAX_UNIFORM_BLOCK_SIZE, &mMaxUniformBlockSize);
+
+    // sanity clamp max uniform block size to 64k just in case 
+    // there's some implementation that reports a crazy value
+    mMaxUniformBlockSize = llmin(mMaxUniformBlockSize, 65536);
 
     if (mGLVersion >= 4.59f)
     {
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 75a7c5d3b2..2f538d0844 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -87,6 +87,7 @@ public:
     S32 mGLMaxIndexRange;
     S32 mGLMaxTextureSize;
     F32 mMaxAnisotropy = 0.f;
+    S32 mMaxUniformBlockSize = 0;
 
     // GL 4.x capabilities
     bool mHasCubeMapArray = false;
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index ecd0c6908b..8ea134393a 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -983,17 +983,25 @@ bool LLGLSLShader::mapUniforms(const vector<LLStaticHashedString>* uniforms)
     }
     //........................................................................................................................................
 
-    if (mFeatures.hasReflectionProbes) // Set up block binding, in a way supported by Apple (rather than binding = 1 in .glsl).
-    {   // See slide 35 and more of https://docs.huihoo.com/apple/wwdc/2011/session_420__advances_in_opengl_for_mac_os_x_lion.pdf
-        static const GLuint BLOCKBINDING = 1; //picked by us
-        //Get the index, similar to a uniform location
-        GLuint UBOBlockIndex = glGetUniformBlockIndex(mProgramObject, "ReflectionProbes");
+    // Set up block binding, in a way supported by Apple (rather than binding = 1 in .glsl).
+    // See slide 35 and more of https://docs.huihoo.com/apple/wwdc/2011/session_420__advances_in_opengl_for_mac_os_x_lion.pdf
+    const char* ubo_names[] =
+    {
+        "ReflectionProbes", // UB_REFLECTION_PROBES
+        "GLTFJoints", // UB_GLTF_JOINTS
+    };
+
+    llassert(LL_ARRAY_SIZE(ubo_names) == NUM_UNIFORM_BLOCKS);
+
+    for (U32 i = 0; i < NUM_UNIFORM_BLOCKS; ++i)
+    {
+        GLuint UBOBlockIndex = glGetUniformBlockIndex(mProgramObject, ubo_names[i]);
         if (UBOBlockIndex != GL_INVALID_INDEX)
         {
-            //Set this index to a binding index
-            glUniformBlockBinding(mProgramObject, UBOBlockIndex, BLOCKBINDING);
+            glUniformBlockBinding(mProgramObject, UBOBlockIndex, i);
         }
     }
+    
     unbind();
 
     LL_DEBUGS("ShaderUniform") << "Total Uniform Size: " << mTotalUniformSize << LL_ENDL;
@@ -1049,9 +1057,10 @@ void LLGLSLShader::bind()
     }
 }
 
-void LLGLSLShader::bind(LLGLSLShader::GLTFVariant variant)
+void LLGLSLShader::bind(U32 variant)
 {
     llassert(mGLTFVariants.size() == LLGLSLShader::NUM_GLTF_VARIANTS);
+    llassert(variant < LLGLSLShader::NUM_GLTF_VARIANTS);
     mGLTFVariants[variant].bind();
 }
 
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index fa01d212e1..8ebea2deca 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -146,6 +146,14 @@ public:
         SG_COUNT
     } eGroup;
 
+    enum UniformBlock : GLuint
+    {
+        UB_REFLECTION_PROBES,
+        UB_GLTF_JOINTS,
+        NUM_UNIFORM_BLOCKS
+    };
+
+
     static std::set<LLGLSLShader*> sInstances;
     static bool sProfileEnabled;
 
@@ -320,20 +328,24 @@ public:
     LLGLSLShader* mRiggedVariant = nullptr;
 
     // variants for use by GLTF renderer
-    // "this" is considered to be OPAQUE
-    enum GLTFVariant
+    // bit 0 = alpha mode blend (1) or opaque (0)
+    // bit 1 = rigged (1) or static (0)
+    struct GLTFVariant
     {
-        STATIC_OPAQUE,
-        STATIC_BLEND,
-        RIGGED_OPAQUE,
-        RIGGED_BLEND,
-        NUM_GLTF_VARIANTS
+        constexpr static U32 RIGGED = 2;
+        constexpr static U32 ALPHA = 1;
+        constexpr static U32 OPAQUE_STATIC = 0;
+        constexpr static U32 ALPHA_STATIC = 1;
+        constexpr static U32 OPAQUE_RIGGED = 2;
+        constexpr static U32 ALPHA_RIGGED = 3;
     };
 
+    constexpr static U32 NUM_GLTF_VARIANTS = 4;
+
     std::vector<LLGLSLShader> mGLTFVariants;
 
     //helper to bind GLTF variant
-    void bind(GLTFVariant variant);
+    void bind(U32 variant);
 
     // hacky flag used for optimization in LLDrawPoolAlpha
     bool mCanBindFast = false;
diff --git a/indra/llrender/llshadermgr.cpp b/indra/llrender/llshadermgr.cpp
index 694bcbeeb9..4e8adb2fb3 100644
--- a/indra/llrender/llshadermgr.cpp
+++ b/indra/llrender/llshadermgr.cpp
@@ -1225,6 +1225,9 @@ void LLShaderMgr::initAttribsAndUniforms()
     mReservedUniforms.push_back("diffuseMap");
     mReservedUniforms.push_back("altDiffuseMap");
     mReservedUniforms.push_back("specularMap");
+    mReservedUniforms.push_back("metallicRoughnessMap");
+    mReservedUniforms.push_back("normalMap");
+    mReservedUniforms.push_back("occlusionMap");
     mReservedUniforms.push_back("emissiveMap");
     mReservedUniforms.push_back("bumpMap");
     mReservedUniforms.push_back("bumpMap2");
@@ -1348,7 +1351,6 @@ void LLShaderMgr::initAttribsAndUniforms()
 
     llassert(mReservedUniforms.size() == LLShaderMgr::DEFERRED_SHADOW5+1);
 
-    mReservedUniforms.push_back("normalMap");
     mReservedUniforms.push_back("positionMap");
     mReservedUniforms.push_back("diffuseRect");
     mReservedUniforms.push_back("specularRect");
diff --git a/indra/llrender/llshadermgr.h b/indra/llrender/llshadermgr.h
index c3e5a2aafd..03803c0e96 100644
--- a/indra/llrender/llshadermgr.h
+++ b/indra/llrender/llshadermgr.h
@@ -93,6 +93,9 @@ public:
         DIFFUSE_MAP,                        //  "diffuseMap"
         ALTERNATE_DIFFUSE_MAP,              //  "altDiffuseMap"
         SPECULAR_MAP,                       //  "specularMap"
+        METALLIC_ROUGHNESS_MAP,             //  "metallicRoughnessMap"
+        NORMAL_MAP,                         //  "normalMap"
+        OCCLUSION_MAP,                      //  "occlusionMap"
         EMISSIVE_MAP,                       //  "emissiveMap"
         BUMP_MAP,                           //  "bumpMap"
         BUMP_MAP2,                          //  "bumpMap2"
@@ -202,7 +205,6 @@ public:
         DEFERRED_SHADOW3,                   //  "shadowMap3"
         DEFERRED_SHADOW4,                   //  "shadowMap4"
         DEFERRED_SHADOW5,                   //  "shadowMap5"
-        DEFERRED_NORMAL,                    //  "normalMap"
         DEFERRED_POSITION,                  //  "positionMap"
         DEFERRED_DIFFUSE,                   //  "diffuseRect"
         DEFERRED_SPECULAR,                  //  "specularRect"
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 7caf20f40b..8cb124d406 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -687,6 +687,7 @@ bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of
     }
 
     {
+#if 0  // not a reliable test for VBOs that are not backed by a CPU buffer
         U16* idx = (U16*) mMappedIndexData+indices_offset;
         for (U32 i = 0; i < count; ++i)
         {
@@ -724,6 +725,7 @@ bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of
                 }
             }
         }
+#endif
     }
 
     return true;
@@ -1036,8 +1038,7 @@ bool LLVertexBuffer::updateNumIndices(U32 nindices)
 
 bool LLVertexBuffer::allocateBuffer(U32 nverts, U32 nindices)
 {
-    if (nverts < 0 || nindices < 0 ||
-        nverts > 65536)
+    if (nverts < 0 || nindices < 0)
     {
         LL_ERRS() << "Bad vertex buffer allocation: " << nverts << " : " << nindices << LL_ENDL;
     }