rsx/interpreter: Improve instructions support

- Must statically write the gl_ClipDistance registers else you get uninitialized trash.
  This problem is more readily apparent on NVIDIA technology but even AMD is not completely immune.

rpcs3/Emu/RSX/Common/Interpreter/FragmentInterpreter.glsl

@@ -176,9 +176,9 @@ vec4 read_src(const in int index)
 			// TODO: wpos
 			value = vec4(0.); break;
 		case 1:
-			value = gl_FrontFacing? in_regs[1] : in_regs[3]; break;
+			value = gl_FrontFacing? in_regs[3] : in_regs[1]; break;
 		case 2:
-			value = gl_FrontFacing? in_regs[2] : in_regs[4]; break;
+			value = gl_FrontFacing? in_regs[4] : in_regs[2]; break;
 		case 3:
 			value = fetch_fog_value(fog_mode, in_regs[5]); break;
 		case 13:

rpcs3/Emu/RSX/Common/Interpreter/VertexInterpreter.glsl

@@ -258,6 +258,13 @@ void write_output(const in int oid, const in int mask_bit)
 	}
 }
 
+// Cannot dynamically index into the gl_ClipDistance array without causing problems due to it's unknown size
+#define write_clip_distance(plane, mask_bit, test, value)\
+	if (test && attribute_enabled(1 << mask_bit))\
+		gl_ClipDistance[plane] = value;\
+	else\
+		gl_ClipDistance[plane] = 0.5f;\
+
 ivec4 read_addr_reg()
 {
 	return a[d0.addr_reg_sel_1];
@@ -524,15 +531,21 @@ void main()
 		}
 	}
 
-	// TODO: 2-sided lighting
+	// Unconditionally update COLOR0 and SPECULAR0
-	if (!attribute_enabled(1 << 0 | 1 << 2))
+	write_output(1, 0);
-	{
+	write_output(2, 1);
-		dest[1] = dest[3] = vec4(0, 0, 0, 1);
-	}
 
-	if (!attribute_enabled(1 << 1 | 1 << 3))
+	// Conditionally update COLOR1 and SPECULAR1 depending on 2-sided mask
+	if (control == 0)
 	{
-		dest[2] = dest[4] = vec4(0, 0, 0, 1);
+		dest[3] = dest[1];
+		dest[4] = dest[2];
+	}
+	else
+	{
+		// 2-sided lighting
+		write_output(3, 2);
+		write_output(4, 3);
 	}
 
 	if (!attribute_enabled(1 << 4))
@@ -549,19 +562,12 @@ void main()
 		gl_PointSize = point_size;
 	}
 
-	if (attribute_enabled(1 << 6 | 1 << 7 | 1 << 8))
+	write_clip_distance(0, 6, user_clip_enabled[0].x > 0, dest[5].y * user_clip_factor[0].x);
-	{
+	write_clip_distance(1, 7, user_clip_enabled[0].y > 0, dest[5].z * user_clip_factor[0].y);
-		gl_ClipDistance[0] = (user_clip_enabled[0].x > 0)? dest[5].y * user_clip_factor[0].x : 0.5f;
+	write_clip_distance(2, 8, user_clip_enabled[0].z > 0, dest[5].w * user_clip_factor[0].z);
-		gl_ClipDistance[1] = (user_clip_enabled[0].y > 0)? dest[5].z * user_clip_factor[0].y : 0.5f;
+	write_clip_distance(3, 9, user_clip_enabled[0].w > 0, dest[6].y * user_clip_factor[0].w);
-		gl_ClipDistance[2] = (user_clip_enabled[0].z > 0)? dest[5].w * user_clip_factor[0].z : 0.5f;
+	write_clip_distance(4, 10, user_clip_enabled[1].x > 0, dest[6].z * user_clip_factor[1].x);
-	}
+	write_clip_distance(5, 11, user_clip_enabled[1].y > 0, dest[6].w * user_clip_factor[1].y);
-
-	if (attribute_enabled(1 << 9 | 1 << 10 | 1 << 11))
-	{
-		gl_ClipDistance[3] = (user_clip_enabled[0].w > 0)? dest[6].y * user_clip_factor[0].w : 0.5f;
-		gl_ClipDistance[4] = (user_clip_enabled[1].x > 0)? dest[6].z * user_clip_factor[1].x : 0.5f;
-		gl_ClipDistance[5] = (user_clip_enabled[1].y > 0)? dest[6].w * user_clip_factor[1].y : 0.5f;
-	}
 
 	write_output(15, 12);
 	write_output(6, 13);

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -778,6 +778,7 @@ void GLGSRender::load_program_env()
 			vp_config[0] = current_vertex_program.base_address;
 			vp_config[1] = current_vertex_program.entry;
 			vp_config[2] = current_vertex_program.output_mask;
+			vp_config[3] = rsx::method_registers.two_side_light_en() ? 1u : 0u;
 
 			std::memcpy(vp_buf + 16, current_vertex_program.data.data(), current_vp_metadata.ucode_length);
 

rpcs3/Emu/RSX/GL/GLShaderInterpreter.cpp

@@ -114,7 +114,7 @@ namespace gl
 		"	uint base_address;\n"
 		"	uint entry;\n"
 		"	uint output_mask;\n"
-		"	uint reserved;\n"
+		"	uint control;\n"
 		"	uvec4 vp_instructions[];\n"
 		"};\n\n";
 
@@ -285,26 +285,22 @@ namespace gl
 			return;
 		}
 
-		if (get_driver_caps().vendor_AMD)
+		// Overlapping texture bindings are trouble. Cannot bind one TIU to two types of samplers simultaneously
+		for (unsigned i = 0; i < replacement_map.size(); ++i)
 		{
-			// AMD drivers don't like texture bindings overlapping which means workarounds are needed
+			for (int j = 0; j < 4; ++j)
-			// Technically this is accurate to spec, but makes efficient usage of shader resources difficult
-			for (unsigned i = 0; i < replacement_map.size(); ++i)
 			{
-				for (int j = 0; j < 4; ++j)
+				auto& pool = allocator.pools[j];
+				for (int k = pool.num_used; k < pool.pool_size; ++k)
 				{
-					auto& pool = allocator.pools[j];
+					if (pool.allocated[k] == replacement_map[i].second)
-					for (int k = pool.num_used; k < pool.pool_size; ++k)
 					{
-						if (pool.allocated[k] == replacement_map[i].second)
+						pool.allocated[k] = replacement_map[i].first;
-						{
+						pool.flags |= static_cast<u32>(interpreter::texture_pool_flags::dirty);
-							pool.allocated[k] = replacement_map[i].first;
-							pool.flags |= static_cast<u32>(interpreter::texture_pool_flags::dirty);
 
-							// Exit nested loop
+						// Exit nested loop
-							j = 4;
+						j = 4;
-							break;
+						break;
-						}
 					}
 				}
 			}

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -467,8 +467,8 @@ VKGSRender::VKGSRender() : GSRender()
 
 	if (g_cfg.video.shader_interpreter_mode != shader_interpreter_mode::disabled)
 	{
-		m_vertex_instructions_buffer.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, 16 * 0x100000, "vertex instructions buffer", 512 * 16);
+		m_vertex_instructions_buffer.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, 64 * 0x100000, "vertex instructions buffer", 512 * 16);
-		m_fragment_instructions_buffer.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, 16 * 0x100000, "fragment instructions buffer", 2048);
+		m_fragment_instructions_buffer.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, 64 * 0x100000, "fragment instructions buffer", 2048);
 	}
 
 	const auto limits = m_device->gpu().get_limits();
@@ -1759,6 +1759,7 @@ void VKGSRender::load_program_env()
 			vp_config[0] = current_vertex_program.base_address;
 			vp_config[1] = current_vertex_program.entry;
 			vp_config[2] = current_vertex_program.output_mask;
+			vp_config[3] = rsx::method_registers.two_side_light_en()? 1u: 0u;
 
 			std::memcpy(vp_buf + 16, current_vertex_program.data.data(), current_vp_metadata.ucode_length);
 			m_vertex_instructions_buffer.unmap();

rpcs3/Emu/RSX/VK/VKShaderInterpreter.cpp

@@ -34,7 +34,7 @@ namespace vk
 		"	uint base_address;\n"
 		"	uint entry;\n"
 		"	uint output_mask;\n"
-		"	uint reserved;\n"
+		"	uint control;\n"
 		"	uvec4 vp_instructions[];\n"
 		"};\n\n";