rsx: Fix support for gamma pack/unpack instructions

rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp

@@ -207,6 +207,8 @@ void GLFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
 	m_shader_props.require_texture_ops = properties.has_tex_op;
 	m_shader_props.require_shadow_ops = properties.shadow_sampler_mask != 0;
 	m_shader_props.require_texture_expand = properties.has_exp_tex_op;
+	m_shader_props.require_srgb_to_linear = properties.has_upg;
+	m_shader_props.require_linear_to_srgb = properties.has_pkg;
 	m_shader_props.emulate_coverage_tests = true; // g_cfg.video.antialiasing_level == msaa_level::none;
 	m_shader_props.emulate_shadow_compare = device_props.emulate_depth_compare;
 	m_shader_props.low_precision_tests = ::gl::get_driver_caps().vendor_NVIDIA;

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -359,6 +359,8 @@ void GLGSRender::on_exit()
 		gl::g_typeless_transfer_buffer.remove();
 	}
 
+	gl::debug::g_vis_texture.reset(); // TODO
+
 	gl::destroy_pipe_compiler();
 
 	m_prog_buffer.clear();

rpcs3/Emu/RSX/GL/GLPresent.cpp

@@ -4,6 +4,22 @@
 
 LOG_CHANNEL(screenshot_log, "SCREENSHOT");
 
+namespace gl
+{
+	namespace debug
+	{
+		std::unique_ptr<texture> g_vis_texture;
+
+		void set_vis_texture(texture* visual)
+		{
+			const auto target = static_cast<GLenum>(visual->get_target());
+			const auto ifmt = static_cast<GLenum>(visual->get_internal_format());
+			g_vis_texture.reset(new texture(target, visual->width(), visual->height(), 1, 1, ifmt, visual->format_class()));
+			glCopyImageSubData(visual->id(), target, 0, 0, 0, 0, g_vis_texture->id(), target, 0, 0, 0, 0, visual->width(), visual->height(), 1);
+		}
+	}
+}
+
 gl::texture* GLGSRender::get_present_source(gl::present_surface_info* info, const rsx::avconf& avconfig)
 {
 	gl::texture* image = nullptr;
@@ -334,6 +350,30 @@ void GLGSRender::flip(const rsx::display_flip_info_t& info)
 		m_text_printer.print_text(4, 180, width, height, fmt::format("Texture uploads: %15u (%u from CPU - %02u%%)", num_texture_upload, num_texture_upload_miss, texture_upload_miss_ratio));
 	}
 
+	if (gl::debug::g_vis_texture)
+	{
+		// Optionally renders a single debug texture to framebuffer.
+		// Only programmatic access provided at the moment.
+		// TODO: Migrate to use overlay system. (kd-11)
+		gl::fbo m_vis_buffer;
+		m_vis_buffer.create();
+		m_vis_buffer.bind();
+		m_vis_buffer.color = gl::debug::g_vis_texture->id();
+		m_vis_buffer.read_buffer(m_vis_buffer.color);
+		m_vis_buffer.draw_buffer(m_vis_buffer.color);
+
+		const u32 vis_width = 320;
+		const u32 vis_height = 240;
+		areai display_view = areai(aspect_ratio).flipped_vertical();
+		display_view.x1 = display_view.x2 - vis_width;
+		display_view.y1 = vis_height;
+
+		// Blit
+		const auto src_region = areau{ 0u, 0u, gl::debug::g_vis_texture->width(), gl::debug::g_vis_texture->height() };
+		m_vis_buffer.blit(gl::screen, static_cast<areai>(src_region), display_view, gl::buffers::color, gl::filter::linear);
+		m_vis_buffer.remove();
+	}
+
 	m_frame->flip(m_context);
 	rsx::thread::flip(info);
 

rpcs3/Emu/RSX/GL/GLTexture.cpp

@@ -10,6 +10,11 @@
 
 namespace gl
 {
+	namespace debug
+	{
+		extern void set_vis_texture(texture*);
+	}
+
 	buffer g_typeless_transfer_buffer;
 
 	GLenum get_target(rsx::texture_dimension_extended type)

rpcs3/Emu/RSX/GL/GLTexture.h

@@ -119,5 +119,10 @@ namespace gl
 		void apply_defaults(GLenum default_filter = GL_NEAREST);
 	};
 
+	namespace debug
+	{
+		extern std::unique_ptr<texture> g_vis_texture;
+	}
+
 	extern buffer g_typeless_transfer_buffer;
 }

rpcs3/Emu/RSX/Program/FragmentProgramDecompiler.cpp

@@ -830,6 +830,26 @@ std::string FragmentProgramDecompiler::BuildCode()
 	"#define _builtin_rsq(x) (1. / _builtin_sqrt(x))\n"
 	"#define _builtin_div(x, y) (x / y)\n\n";
 
+	if (properties.has_pkg)
+	{
+		OS <<
+		"vec4 _builtin_pkg(const in vec4 value)\n"
+		"{\n"
+		"	vec4 convert = linear_to_srgb(value);\n"
+		"	return uintBitsToFloat(packUnorm4x8(convert)).xxxx;\n"
+		"}\n\n";
+	}
+
+	if (properties.has_upg)
+	{
+		OS <<
+		"vec4 _builtin_upg(const in float value)\n"
+		"{\n"
+		"	vec4 raw = unpackUnorm4x8(floatBitsToUint(value));\n"
+		"	return srgb_to_linear(raw);\n"
+		"}\n\n";
+	}
+
 	if (properties.has_divsq)
 	{
 		// Define RSX-compliant DIVSQ
@@ -954,6 +974,9 @@ bool FragmentProgramDecompiler::handle_sct_scb(u32 opcode)
 	case RSX_FP_OPCODE_PK16: SetDst(getFloatTypeName(4) + "(uintBitsToFloat(packUnorm2x16($0.xy)))"); return true;
 	case RSX_FP_OPCODE_PKG:
 		// Should be similar to PKB but with gamma correction, see description of PK4UBG in khronos page
+		properties.has_pkg = true;
+		SetDst("_builtin_pkg($0)");
+		return true;
 	case RSX_FP_OPCODE_PKB: SetDst(getFloatTypeName(4) + "(uintBitsToFloat(packUnorm4x8($0)))"); return true;
 	case RSX_FP_OPCODE_SIN: SetDst("sin($0.xxxx)"); return true;
 	}
@@ -1120,7 +1143,10 @@ bool FragmentProgramDecompiler::handle_tex_srb(u32 opcode)
 	case RSX_FP_OPCODE_UP4: SetDst("unpackSnorm4x8(floatBitsToUint($0.x))"); return true;
 	case RSX_FP_OPCODE_UP16: SetDst("unpackUnorm2x16(floatBitsToUint($0.x)).xyxy"); return true;
 	case RSX_FP_OPCODE_UPG:
-	// Same as UPB with gamma correction
+		// Same as UPB with gamma correction
+		properties.has_upg = true;
+		SetDst("_builtin_upg($0.x)");
+		return true;
 	case RSX_FP_OPCODE_UPB: SetDst("(unpackUnorm4x8(floatBitsToUint($0.x)))"); return true;
 	}
 	return false;

rpcs3/Emu/RSX/Program/FragmentProgramDecompiler.h

@@ -285,6 +285,8 @@ class FragmentProgramDecompiler
 		bool has_clamp = false;
 		bool has_w_access = false;
 		bool has_exp_tex_op = false;
+		bool has_pkg = false;
+		bool has_upg = false;
 	}
 	properties;
 

rpcs3/Emu/RSX/Program/GLSLCommon.cpp

@@ -673,7 +673,7 @@ namespace glsl
 			"}\n\n";
 		}
 
-		if (!props.fp32_outputs)
+		if (!props.fp32_outputs || props.require_linear_to_srgb)
 		{
 			OS <<
 			"vec4 linear_to_srgb(const in vec4 cl)\n"
@@ -685,6 +685,17 @@ namespace glsl
 			"}\n\n";
 		}
 
+		if (props.require_texture_ops || props.require_srgb_to_linear)
+		{
+			OS <<
+			"vec4 srgb_to_linear(const in vec4 cs)\n"
+			"{\n"
+			"	vec4 a = cs / 12.92;\n"
+			"	vec4 b = pow((cs + 0.055) / 1.055, vec4(2.4));\n"
+			"	return _select(a, b, greaterThan(cs, vec4(0.04045)));\n"
+			"}\n\n";
+		}
+
 		if (props.require_depth_conversion)
 		{
 			ensure(props.require_texture_ops);
@@ -763,12 +774,6 @@ namespace glsl
 			"	return mix(direct, indexed, choice);\n"
 			"}\n\n"
 #endif
-			"vec4 srgb_to_linear(const in vec4 cs)\n"
-			"{\n"
-			"	vec4 a = cs / 12.92;\n"
-			"	vec4 b = pow((cs + 0.055) / 1.055, vec4(2.4));\n"
-			"	return _select(a, b, greaterThan(cs, vec4(0.04045)));\n"
-			"}\n\n"
 
 			//TODO: Move all the texture read control operations here
 			"vec4 process_texel(in vec4 rgba, const in uint control_bits)\n"

rpcs3/Emu/RSX/Program/GLSLTypes.h

@@ -28,6 +28,8 @@ namespace glsl
 		bool require_texture_ops;
 		bool require_shadow_ops;
 		bool require_texture_expand;
+		bool require_srgb_to_linear;
+		bool require_linear_to_srgb;
 		bool emulate_coverage_tests;
 		bool emulate_shadow_compare;
 		bool emulate_zclip_transform;

rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp

@@ -244,6 +244,8 @@ void VKFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
 	m_shader_props.require_texture_ops = properties.has_tex_op;
 	m_shader_props.require_shadow_ops = properties.shadow_sampler_mask != 0;
 	m_shader_props.require_texture_expand = properties.has_exp_tex_op;
+	m_shader_props.require_srgb_to_linear = properties.has_upg;
+	m_shader_props.require_linear_to_srgb = properties.has_pkg;
 	m_shader_props.emulate_coverage_tests = g_cfg.video.antialiasing_level == msaa_level::none;
 	m_shader_props.emulate_shadow_compare = device_props.emulate_depth_compare;
 	m_shader_props.low_precision_tests = vk::get_driver_vendor() == vk::driver_vendor::NVIDIA;