Revert "revert workaround made for #1368, doesn't work on (certain) Intel chipsets. Instead, limit the maximum bone count to 64 so to meet the limit of 1024 uniform components by certain graphics cards (#1285)"

This reverts commit f57bfa089c.

This was committed by mistake. Fixes #1375.

planet/Graphics.ocg/ObjectDefaultVS.glsl

@@ -15,7 +15,11 @@
 
 // Default Vertex Shader for mesh-based objects.
 
-// Input uniforms:
+// Input uniforms, if OC_WA_LOW_MAX_VERTEX_UNIFORM_COMPONENTS is NOT defined:
+//   bones: array of 4x3 bone transformation matrices.
+
+// Input uniforms, if OC_WA_LOW_MAX_VERTEX_UNIFORM_COMPONENTS is defined:
+//   bones: array of 3x4 transposed bone transformation matrices.
 
 // Input vertex attributes:
 //   oc_BoneWeights0 and oc_BoneWeight1: vectors of bone influence weights.
@@ -30,11 +34,15 @@
 // inside the bone matrix array that contains the identity matrix, with a bone
 // weight of 1.0.
 
-#define MAX_BONE_COUNT 64
+#define MAX_BONE_COUNT 80
 
 varying vec3 normalDir;
 
-uniform mat4 bones[MAX_BONE_COUNT];
+#ifndef OC_WA_LOW_MAX_VERTEX_UNIFORM_COMPONENTS
+uniform mat4x3 bones[MAX_BONE_COUNT];
+#else
+uniform mat3x4 bones[MAX_BONE_COUNT];
+#endif
 
 // For more performance, this should be set by the engine, and this shader
 // should be compiled three times: with BONE_COUNT set to 0, 4, and 8,
@@ -49,10 +57,17 @@ attribute vec4 oc_BoneIndices1;
 attribute vec4 oc_BoneWeights1;
 #endif
 
-vec4 merge_bone(vec4 vertex, vec4 original, mat4 bone, float weight)
+#ifndef OC_WA_LOW_MAX_VERTEX_UNIFORM_COMPONENTS
+vec4 merge_bone(vec4 vertex, vec4 original, mat4x3 bone, float weight)
 {
-	return (bone * original) * weight + vertex;
+	return (mat4(bone) * original) * weight + vertex;
 }
+#else
+vec4 merge_bone(vec4 vertex, vec4 original, mat3x4 bone, float weight)
+{
+	return (mat4(transpose(bone)) * original) * weight + vertex;
+}
+#endif
 
 slice(position)
 {

src/graphics/C4DrawGL.cpp

@@ -260,6 +260,15 @@ CStdGLCtx *CStdGL::CreateContext(C4Window * pWindow, C4AbstractApp *pApp)
 				LogSilentF("GLExt: %s", gl_extensions ? gl_extensions : "");
 			}
 		}
+
+		// Check which workarounds we have to apply
+		{
+			// If we have less than 2048 uniform components available, we
+			// need to upload bone matrices in a different way
+			GLint count;
+			glGetIntegerv(GL_MAX_VERTEX_UNIFORM_COMPONENTS, &count);
+			Workarounds.LowMaxVertexUniformCount = count < 2048;
+		}
 	}
 	if (!success)
 	{
@@ -819,6 +828,7 @@ void CStdGL::Default()
 	iPixelFormat=0;
 	sfcFmt=0;
 	iClrDpt=0;
+	Workarounds.LowMaxVertexUniformCount = false;
 }
 
 #endif // USE_CONSOLE

src/graphics/C4DrawGL.h

@@ -190,6 +190,11 @@ public:
 	C4Shader* GetSpriteShader(int ssc);
 	C4Shader* GetSpriteShader(bool haveBase, bool haveOverlay, bool haveNormal);
 
+	struct
+	{
+		bool LowMaxVertexUniformCount;
+	} Workarounds;
+
 protected:
 	bool CreatePrimarySurfaces(unsigned int iXRes, unsigned int iYRes, int iColorDepth, unsigned int iMonitor);
 

src/graphics/C4DrawMeshGL.cpp

@@ -128,7 +128,10 @@ namespace
 			);
 		}
 
-		return buf;
+		if (pGL->Workarounds.LowMaxVertexUniformCount)
+			return StdStrBuf("#define OC_WA_LOW_MAX_VERTEX_UNIFORM_COMPONENTS\n") + buf;
+		else
+			return buf;
 	}
 
 	// Note this only gets the code which inserts the slices specific for the pass
@@ -474,7 +477,7 @@ namespace
 		// Or, even better, we could upload them into a UBO, but Intel doesn't support them prior to Sandy Bridge.
 		struct BoneTransform
 		{
-			float m[4][4];
+			float m[3][4];
 		};
 		std::vector<BoneTransform> bones;
 		if (mesh_instance.GetBoneCount() == 0)
@@ -483,7 +486,7 @@ namespace
 			static const BoneTransform dummy_bone = {
 				1.0f, 0.0f, 0.0f, 0.0f,
 				0.0f, 1.0f, 0.0f, 0.0f,
-				0.0f, 0.0f, 1.0f, 1.0f
+				0.0f, 0.0f, 1.0f, 0.0f
 			};
 			bones.push_back(dummy_bone);
 		}
@@ -496,8 +499,7 @@ namespace
 				BoneTransform cooked_bone = {
 					bone(0, 0), bone(0, 1), bone(0, 2), bone(0, 3),
 					bone(1, 0), bone(1, 1), bone(1, 2), bone(1, 3),
-					bone(2, 0), bone(2, 1), bone(2, 2), bone(2, 3),
-					0, 0, 0, 1
+					bone(2, 0), bone(2, 1), bone(2, 2), bone(2, 3)
 				};
 				bones.push_back(cooked_bone);
 			}
@@ -622,7 +624,10 @@ namespace
 			// Upload the current bone transformation matrixes (if there are any)
 			if (!bones.empty())
 			{
-				call.SetUniformMatrix4x4fv(C4SSU_Bones, bones.size(), &bones[0].m[0][0]);
+				if (pGL->Workarounds.LowMaxVertexUniformCount)
+					glUniformMatrix3x4fv(shader->GetUniform(C4SSU_Bones), bones.size(), GL_FALSE, &bones[0].m[0][0]);
+				else
+					glUniformMatrix4x3fv(shader->GetUniform(C4SSU_Bones), bones.size(), GL_TRUE, &bones[0].m[0][0]);
 			}
 
 			// Bind the vertex data of the mesh