Move native/gfx_es2/gl_state.cpp/h into GPU/GLES/GLStateCache.cpp/h

CMakeLists.txt

@@ -923,8 +923,6 @@ add_library(native STATIC
 	native/gfx_es2/draw_buffer.h
 	native/gfx_es2/draw_text.cpp
 	native/gfx_es2/draw_text.h
-	native/gfx_es2/gl_state.cpp
-	native/gfx_es2/gl_state.h
 	native/gfx_es2/gpu_features.cpp
 	native/gfx_es2/gpu_features.h
 	native/gfx_es2/glsl_program.cpp
@@ -1487,6 +1485,8 @@ add_library(GPU OBJECT
 	GPU/GLES/FBO.h
 	GPU/GLES/GLES_GPU.cpp
 	GPU/GLES/GLES_GPU.h
+	GPU/GLES/GLStateCache.cpp
+	GPU/GLES/GLStateCache.h
 	GPU/GLES/FragmentShaderGenerator.cpp
 	GPU/GLES/FragmentShaderGenerator.h
 	GPU/GLES/FragmentTestCache.cpp

GPU/Common/DepalettizeShaderCommon.cpp

@@ -17,10 +17,10 @@
 
 #include <stdio.h>
 
-#include "gfx_es2/gl_state.h"
 #include "Common/Log.h"
 #include "Core/Reporting.h"
 #include "GPU/GPUState.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/Common/DepalettizeShaderCommon.h"
 
 

GPU/GLES/DepalettizeShader.cpp

@@ -22,6 +22,7 @@
 #include "Core/Reporting.h"
 #include "DepalettizeShader.h"
 #include "GPU/GLES/TextureCache.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/Common/DepalettizeShaderCommon.h"
 
 static const int DEPAL_TEXTURE_OLD_AGE = 120;

GPU/GLES/DepalettizeShader.h

@@ -18,7 +18,7 @@
 #include <map>
 
 #include "Common/CommonTypes.h"
-#include "gfx_es2/gl_state.h"
+#include "gfx/gl_common.h"
 #include "GPU/ge_constants.h"
 
 class DepalShader {

GPU/GLES/FBO.cpp

@@ -19,8 +19,8 @@
 
 #include "base/logging.h"
 #include "gfx/gl_common.h"
-#include "gfx_es2/gl_state.h"
 
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/FBO.h"
 
 #ifdef IOS

GPU/GLES/FragmentTestCache.h

@@ -19,10 +19,11 @@
 
 #include <map>
 #include "Common/CommonTypes.h"
-#include "gfx_es2/gl_state.h"
-#include "GPU/ge_constants.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/TextureCache.h"
 
+#include "GPU/ge_constants.h"
+
 struct FragmentTestID {
 	union {
 		struct {

GPU/GLES/Framebuffer.cpp

@@ -21,7 +21,6 @@
 #include "profiler/profiler.h"
 
 #include "gfx_es2/glsl_program.h"
-#include "gfx_es2/gl_state.h"
 
 #include "base/timeutil.h"
 #include "math/lin/matrix4x4.h"
@@ -40,6 +39,7 @@
 #include "GPU/Common/TextureDecoder.h"
 #include "GPU/Common/FramebufferCommon.h"
 #include "GPU/Debugger/Stepping.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/FBO.h"
 #include "GPU/GLES/Framebuffer.h"
 #include "GPU/GLES/TextureCache.h"

GPU/GLES/GLES_GPU.cpp

@@ -16,7 +16,6 @@
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
 #include "base/logging.h"
-#include "gfx_es2/gl_state.h"
 #include "profiler/profiler.h"
 
 #include "Common/ChunkFile.h"
@@ -34,6 +33,7 @@
 #include "GPU/GeDisasm.h"
 #include "GPU/Common/FramebufferCommon.h"
 
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/ShaderManager.h"
 #include "GPU/GLES/GLES_GPU.h"
 #include "GPU/GLES/Framebuffer.h"

GPU/GLES/GLStateCache.cpp

@@ -0,0 +1,52 @@
+#include <stdlib.h>
+
+#include "base/logging.h"
+
+#include "GPU/GLES/GLStateCache.h"
+
+OpenGLState glstate;
+
+int OpenGLState::state_count = 0;
+
+void OpenGLState::Restore() {
+	int count = 0;
+
+	blend.restore(); count++;
+	blendEquationSeparate.restore(); count++;
+	blendFuncSeparate.restore(); count++;
+	blendColor.restore(); count++;
+
+	scissorTest.restore(); count++;
+	scissorRect.restore(); count++;
+
+	cullFace.restore(); count++;
+	cullFaceMode.restore(); count++;
+	frontFace.restore(); count++;
+
+	depthTest.restore(); count++;
+	depthRange.restore(); count++;
+	depthFunc.restore(); count++;
+	depthWrite.restore(); count++;
+
+	colorMask.restore(); count++;
+	viewport.restore(); count++;
+
+	stencilTest.restore(); count++;
+	stencilOp.restore(); count++;
+	stencilFunc.restore(); count++;
+	stencilMask.restore(); count++;
+
+	dither.restore(); count++;
+
+#if !defined(USING_GLES2)
+	colorLogicOp.restore(); count++;
+	logicOp.restore(); count++;
+#endif
+
+	arrayBuffer.restore(); count++;
+	elementArrayBuffer.restore(); count++;
+
+	if (count != state_count) {
+		FLOG("OpenGLState::Restore is missing some states");
+	}
+}

GPU/GLES/GLStateCache.h

@@ -0,0 +1,240 @@
+#pragma once
+
+#include <functional>
+#include <string.h>
+#include <string>
+
+#include "gfx/gl_common.h"
+#include "gfx_es2/gpu_features.h"
+
+// OpenGL state cache.
+// Probably only really worth it in rendering cores on weak mobile hardware.
+class OpenGLState {
+private:
+	template<GLenum cap, bool init>
+	class BoolState {
+	private:
+		bool _value;
+	public:
+		BoolState() : _value(init) {
+			OpenGLState::state_count++;
+		}
+
+		inline void set(bool value) {
+			if (value && value != _value) {
+				_value = value;
+				glEnable(cap);
+			}
+			if (!value && value != _value) {
+				_value = value;
+				glDisable(cap);
+			}
+		}
+		inline void enable() {
+			set(true);
+		}
+		inline void disable() {
+			set(false);
+		}
+		operator bool() const {
+			return isset();
+		}
+		inline bool isset() {
+			return _value;
+		}
+		void restore() {
+			if (_value)
+				glEnable(cap);
+			else
+				glDisable(cap);
+		}
+	};
+
+#define STATE1(func, p1type, p1def) \
+	class SavedState1_##func { \
+		p1type p1; \
+	public: \
+		SavedState1_##func() : p1(p1def) { \
+			OpenGLState::state_count++; \
+		} \
+		void set(p1type newp1) { \
+			if (newp1 != p1) { \
+				p1 = newp1; \
+				func(p1); \
+			} \
+		} \
+		void restore() { \
+			func(p1); \
+		} \
+	}
+
+#define STATE2(func, p1type, p2type, p1def, p2def) \
+	class SavedState2_##func { \
+		p1type p1; \
+		p2type p2; \
+	public: \
+		SavedState2_##func() : p1(p1def), p2(p2def) { \
+			OpenGLState::state_count++; \
+		} \
+		inline void set(p1type newp1, p2type newp2) { \
+			if (newp1 != p1 || newp2 != p2) { \
+				p1 = newp1; \
+				p2 = newp2; \
+				func(p1, p2); \
+			} \
+		} \
+		inline void restore() { \
+			func(p1, p2); \
+		} \
+	}
+
+#define STATE3(func, p1type, p2type, p3type, p1def, p2def, p3def) \
+	class SavedState3_##func { \
+		p1type p1; \
+		p2type p2; \
+		p3type p3; \
+	public: \
+		SavedState3_##func() : p1(p1def), p2(p2def), p3(p3def) { \
+			OpenGLState::state_count++; \
+		} \
+		inline void set(p1type newp1, p2type newp2, p3type newp3) { \
+			if (newp1 != p1 || newp2 != p2 || newp3 != p3) { \
+				p1 = newp1; \
+				p2 = newp2; \
+				p3 = newp3; \
+				func(p1, p2, p3); \
+			} \
+		} \
+		inline void restore() { \
+			func(p1, p2, p3); \
+		} \
+	}
+
+	#define STATE4(func, p1type, p2type, p3type, p4type, p1def, p2def, p3def, p4def) \
+	class SavedState4_##func { \
+		p1type p1; \
+		p2type p2; \
+		p3type p3; \
+		p4type p4; \
+	public: \
+		SavedState4_##func() : p1(p1def), p2(p2def), p3(p3def), p4(p4def) { \
+			OpenGLState::state_count++; \
+		} \
+		inline void set(p1type newp1, p2type newp2, p3type newp3, p4type newp4) { \
+			if (newp1 != p1 || newp2 != p2 || newp3 != p3 || newp4 != p4) { \
+				p1 = newp1; \
+				p2 = newp2; \
+				p3 = newp3; \
+				p4 = newp4; \
+				func(p1, p2, p3, p4); \
+			} \
+		} \
+		inline void restore() { \
+			func(p1, p2, p3, p4); \
+		} \
+	}
+
+#define STATEFLOAT4(func, def) \
+	class SavedState4_##func { \
+		float p[4]; \
+	public: \
+		SavedState4_##func() { \
+			for (int i = 0; i < 4; i++) {p[i] = def;} \
+			OpenGLState::state_count++; \
+		} \
+		inline void set(const float v[4]) { \
+			if (memcmp(p,v,sizeof(float)*4)) { \
+				memcpy(p,v,sizeof(float)*4); \
+				func(p[0], p[1], p[2], p[3]); \
+			} \
+		} \
+		inline void restore() { \
+			func(p[0], p[1], p[2], p[3]); \
+		} \
+	}
+
+#define STATEBIND(func, target) \
+	class SavedBind_##func_##target { \
+		GLuint val_; \
+	public: \
+		SavedBind_##func_##target() { \
+			val_ = 0; \
+			OpenGLState::state_count++; \
+		} \
+		inline void bind(GLuint val) { \
+			if (val_ != val) { \
+				func(target, val); \
+				val_ = val; \
+			} \
+		} \
+		inline void unbind() { \
+			bind(0); \
+		} \
+		inline void restore() { \
+			func(target, val_); \
+		} \
+	}
+
+public:
+	static int state_count;
+	OpenGLState() {}
+	void Restore();
+
+	// When adding a state here, don't forget to add it to OpenGLState::Restore() too
+
+	// Blending
+	BoolState<GL_BLEND, false> blend;
+	STATE4(glBlendFuncSeparate, GLenum, GLenum, GLenum, GLenum, GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA, GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA) blendFuncSeparate;
+
+	// On OpenGL ES, using minmax blend requires glBlendEquationEXT (in theory at least but I don't think it's true in practice)
+	STATE2(glBlendEquationSeparate, GLenum, GLenum, GL_FUNC_ADD, GL_FUNC_ADD) blendEquationSeparate;
+	STATEFLOAT4(glBlendColor, 1.0f) blendColor;
+
+	// Logic Ops. Not available on OpenGL ES at all.
+#if !defined(USING_GLES2)
+	BoolState<GL_COLOR_LOGIC_OP, false> colorLogicOp;
+	STATE1(glLogicOp, GLenum, GL_COPY) logicOp;
+#endif
+
+	// Dither
+	BoolState<GL_DITHER, false> dither;
+
+	// Cull Face
+	BoolState<GL_CULL_FACE, false> cullFace;
+	STATE1(glCullFace, GLenum, GL_FRONT) cullFaceMode;
+	STATE1(glFrontFace, GLenum, GL_CCW) frontFace;
+
+	// Depth Test
+	BoolState<GL_DEPTH_TEST, false> depthTest;
+#if defined(USING_GLES2)
+	STATE2(glDepthRangef, float, float, 0.f, 1.f) depthRange;
+#else
+	STATE2(glDepthRange, double, double, 0.0, 1.0) depthRange;
+#endif
+	STATE1(glDepthFunc, GLenum, GL_LESS) depthFunc;
+	STATE1(glDepthMask, GLboolean, GL_TRUE) depthWrite;
+
+	// Color Mask
+	STATE4(glColorMask, bool, bool, bool, bool, true, true, true, true) colorMask;
+
+	// Viewport
+	STATE4(glViewport, GLint, GLint, GLsizei, GLsizei, 0, 0, 128, 128) viewport;
+
+	// Scissor Test
+	BoolState<GL_SCISSOR_TEST, false> scissorTest;
+	STATE4(glScissor, GLint, GLint, GLsizei, GLsizei, 0, 0, 128, 128) scissorRect;
+
+	// Stencil Test
+	BoolState<GL_STENCIL_TEST, false> stencilTest;
+	STATE3(glStencilOp, GLenum, GLenum, GLenum, GL_KEEP, GL_KEEP, GL_KEEP) stencilOp;
+	STATE3(glStencilFunc, GLenum, GLint, GLuint, GL_ALWAYS, 0, 0xFF) stencilFunc;
+	STATE1(glStencilMask, GLuint, 0xFF) stencilMask;
+
+	STATEBIND(glBindBuffer, GL_ARRAY_BUFFER) arrayBuffer;
+	STATEBIND(glBindBuffer, GL_ELEMENT_ARRAY_BUFFER) elementArrayBuffer;
+};
+
+#undef STATE1
+#undef STATE2
+
+extern OpenGLState glstate;

GPU/GLES/ShaderManager.cpp

@@ -27,7 +27,6 @@
 
 #include "base/logging.h"
 #include "math/math_util.h"
-#include "gfx_es2/gl_state.h"
 #include "math/lin/matrix4x4.h"
 #include "profiler/profiler.h"
 
@@ -36,6 +35,7 @@
 #include "GPU/Math3D.h"
 #include "GPU/GPUState.h"
 #include "GPU/ge_constants.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/ShaderManager.h"
 #include "GPU/GLES/TransformPipeline.h"
 #include "UI/OnScreenDisplay.h"

GPU/GLES/StateMapping.cpp

@@ -21,7 +21,6 @@
 
 
 #include "StateMapping.h"
-#include "gfx_es2/gl_state.h"
 #include "profiler/profiler.h"
 
 #include "GPU/Math3D.h"
@@ -31,6 +30,7 @@
 #include "Core/Config.h"
 #include "Core/Reporting.h"
 #include "GPU/GLES/GLES_GPU.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/ShaderManager.h"
 #include "GPU/GLES/TextureCache.h"
 #include "GPU/GLES/Framebuffer.h"

GPU/GLES/StencilBuffer.cpp

@@ -16,8 +16,8 @@
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
 #include "gfx_es2/glsl_program.h"
-#include "gfx_es2/gl_state.h"
 #include "Core/Reporting.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/Framebuffer.h"
 #include "GPU/GLES/ShaderManager.h"
 #include "GPU/GLES/TextureCache.h"

GPU/GLES/TextureCache.cpp

@@ -18,6 +18,8 @@
 #include <algorithm>
 #include <cstring>
 
+#include "ext/xxhash.h"
+#include "math/math_util.h"
 #include "profiler/profiler.h"
 
 #include "Common/ColorConv.h"
@@ -26,6 +28,7 @@
 #include "Core/Reporting.h"
 #include "GPU/ge_constants.h"
 #include "GPU/GPUState.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/TextureCache.h"
 #include "GPU/GLES/Framebuffer.h"
 #include "GPU/GLES/FragmentShaderGenerator.h"
@@ -35,10 +38,6 @@
 #include "Core/Config.h"
 #include "Core/Host.h"
 
-#include "ext/xxhash.h"
-#include "math/math_util.h"
-#include "native/gfx_es2/gl_state.h"
-
 #ifdef _M_SSE
 #include <xmmintrin.h>
 #endif

GPU/GLES/TransformPipeline.cpp

@@ -72,7 +72,6 @@
 #include "Core/Config.h"
 #include "Core/CoreTiming.h"
 
-#include "gfx_es2/gl_state.h"
 #include "profiler/profiler.h"
 
 #include "GPU/Math3D.h"
@@ -83,6 +82,7 @@
 #include "GPU/Common/SplineCommon.h"
 #include "GPU/Common/VertexDecoderCommon.h"
 #include "GPU/Common/SoftwareTransformCommon.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "GPU/GLES/FragmentTestCache.h"
 #include "GPU/GLES/StateMapping.h"
 #include "GPU/GLES/TextureCache.h"

GPU/GPU.vcxproj

@@ -222,6 +222,7 @@
     <ClInclude Include="GLES\FragmentTestCache.h" />
     <ClInclude Include="GLES\Framebuffer.h" />
     <ClInclude Include="GLES\GLES_GPU.h" />
+    <ClInclude Include="GLES\GLStateCache.h" />
     <ClInclude Include="GLES\ShaderManager.h" />
     <ClInclude Include="GLES\StateMapping.h" />
     <ClInclude Include="GLES\TextureCache.h" />
@@ -297,6 +298,7 @@
     <ClCompile Include="GLES\FragmentTestCache.cpp" />
     <ClCompile Include="GLES\Framebuffer.cpp" />
     <ClCompile Include="GLES\GLES_GPU.cpp" />
+    <ClCompile Include="GLES\GLStateCache.cpp" />
     <ClCompile Include="GLES\ShaderManager.cpp" />
     <ClCompile Include="GLES\StateMapping.cpp" />
     <ClCompile Include="GLES\StencilBuffer.cpp" />

GPU/GPU.vcxproj.filters

@@ -189,6 +189,9 @@
     <ClInclude Include="Directx9\helper\dx_fbo.h">
       <Filter>DirectX9\helper</Filter>
     </ClInclude>
+    <ClInclude Include="GLES\GLStateCache.h">
+      <Filter>GLES</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="Math3D.cpp">
@@ -365,5 +368,8 @@
     <ClCompile Include="Directx9\helper\dx_fbo.cpp">
       <Filter>DirectX9\helper</Filter>
     </ClCompile>
+    <ClCompile Include="GLES\GLStateCache.cpp">
+      <Filter>GLES</Filter>
+    </ClCompile>
   </ItemGroup>
 </Project>

GPU/Software/SoftGpu.cpp

@@ -29,7 +29,6 @@
 #include "Core/MIPS/MIPS.h"
 #include "Core/Reporting.h"
 #include "gfx/gl_common.h"
-// #include "gfx_es2/gl_state.h"
 #include "profiler/profiler.h"
 
 #include "GPU/Software/SoftGpu.h"

Qt/Debugger/debugger_memorytex.cpp

@@ -1,5 +1,5 @@
 #include "debugger_memorytex.h"
-#include "gfx_es2/gl_state.h"
+#include "GPU/GLES/GLStateCache.h"
 #include "gfx/gl_common.h"
 #include "gfx/gl_lost_manager.h"
 #include "ui_debugger_memorytex.h"

native

@@ -1 +1 @@
-Subproject commit 65e3d31e249ca811809121529f1971e408c154d3
+Subproject commit 3f02f9fc778b5b4c45db92a96f0a4087f947d5a2