GLSL: Implement GL_EXT_mesh_shader.

shaders/mesh/mesh-shader-basic-triangle.mesh

@@ -0,0 +1,63 @@
+#version 450
+#extension GL_EXT_mesh_shader : require
+layout(local_size_x = 2, local_size_y = 3, local_size_z = 4) in;
+layout(triangles, max_vertices = 24, max_primitives = 22) out;
+
+out gl_MeshPerVertexEXT
+{
+    vec4 gl_Position;
+    float gl_PointSize;
+    float gl_ClipDistance[1];
+    float gl_CullDistance[2];
+} gl_MeshVerticesEXT[];
+
+layout(location = 0) out vec4 vOut[];
+layout(location = 1) perprimitiveEXT out vec4 vPrim[];
+
+layout(location = 2) out BlockOut
+{
+	vec4 a;
+	vec4 b;
+} outputs[];
+
+layout(location = 4) perprimitiveEXT out BlockOutPrim
+{
+	vec4 a;
+	vec4 b;
+} prim_outputs[];
+
+shared float shared_float[16];
+
+struct TaskPayload
+{
+	float a;
+	float b;
+	int c;
+};
+
+taskPayloadSharedEXT TaskPayload payload;
+
+void main()
+{
+	SetMeshOutputsEXT(24, 22);
+	gl_MeshVerticesEXT[gl_LocalInvocationIndex].gl_Position = vec4(gl_GlobalInvocationID, 1.0);
+	gl_MeshVerticesEXT[gl_LocalInvocationIndex].gl_PointSize = 2.0;
+	gl_MeshVerticesEXT[gl_LocalInvocationIndex].gl_ClipDistance[0] = 4.0;
+	gl_MeshVerticesEXT[gl_LocalInvocationIndex].gl_CullDistance[1] = 5.0;
+	vOut[gl_LocalInvocationIndex] = vec4(gl_GlobalInvocationID, 2.0);
+	outputs[gl_LocalInvocationIndex].a = vec4(5.0);
+	outputs[gl_LocalInvocationIndex].b = vec4(6.0);
+	barrier();
+	if (gl_LocalInvocationIndex < 22)
+	{
+		vPrim[gl_LocalInvocationIndex] = vec4(gl_WorkGroupID, 3.0);
+		prim_outputs[gl_LocalInvocationIndex].a = vec4(payload.a);
+		prim_outputs[gl_LocalInvocationIndex].b = vec4(payload.b);
+		gl_PrimitiveTriangleIndicesEXT[gl_LocalInvocationIndex] = uvec3(0, 1, 2) + gl_LocalInvocationIndex;
+		gl_MeshPrimitivesEXT[gl_LocalInvocationIndex].gl_PrimitiveID = int(gl_GlobalInvocationID.x);
+		gl_MeshPrimitivesEXT[gl_LocalInvocationIndex].gl_Layer = int(gl_GlobalInvocationID.x) + 1;
+		gl_MeshPrimitivesEXT[gl_LocalInvocationIndex].gl_ViewportIndex = int(gl_GlobalInvocationID.x) + 2;
+		gl_MeshPrimitivesEXT[gl_LocalInvocationIndex].gl_CullPrimitiveEXT = bool(gl_GlobalInvocationID.x & 1);
+		gl_MeshPrimitivesEXT[gl_LocalInvocationIndex].gl_PrimitiveShadingRateEXT = int(gl_GlobalInvocationID.x) + 3;
+	}
+}

spirv_common.hpp

@@ -682,6 +682,7 @@ struct SPIREntryPoint
 	} workgroup_size;
 	uint32_t invocations = 0;
 	uint32_t output_vertices = 0;
+	uint32_t output_primitives = 0;
 	spv::ExecutionModel model = spv::ExecutionModelMax;
 	bool geometry_passthrough = false;
 };

spirv_cross.cpp

@@ -154,6 +154,10 @@ bool Compiler::block_is_pure(const SPIRBlock &block)
 		case OpEmitVertex:
 			return false;
 
+		// Mesh shader functions modify global state.
+		case OpSetMeshOutputsEXT:
+			return false;
+
 		// Barriers disallow any reordering, so we should treat blocks with barrier as writing.
 		case OpControlBarrier:
 		case OpMemoryBarrier:
@@ -1069,8 +1073,11 @@ void Compiler::parse_fixup()
 		{
 			auto &var = id.get<SPIRVariable>();
 			if (var.storage == StorageClassPrivate || var.storage == StorageClassWorkgroup ||
+			    var.storage == StorageClassTaskPayloadWorkgroupEXT ||
 			    var.storage == StorageClassOutput)
+			{
 				global_variables.push_back(var.self);
+			}
 			if (variable_storage_is_aliased(var))
 				aliased_variables.push_back(var.self);
 		}
@@ -2177,6 +2184,10 @@ void Compiler::set_execution_mode(ExecutionMode mode, uint32_t arg0, uint32_t ar
 		execution.output_vertices = arg0;
 		break;
 
+	case ExecutionModeOutputPrimitivesEXT:
+		execution.output_primitives = arg0;
+		break;
+
 	default:
 		break;
 	}
@@ -2297,6 +2308,9 @@ uint32_t Compiler::get_execution_mode_argument(spv::ExecutionMode mode, uint32_t
 	case ExecutionModeOutputVertices:
 		return execution.output_vertices;
 
+	case ExecutionModeOutputPrimitivesEXT:
+		return execution.output_primitives;
+
 	default:
 		return 0;
 	}