From f424bf73ae5f514d3f82bcf1d64c4d63dcdfdf09 Mon Sep 17 00:00:00 2001
From: Henrik Rydgard <hrydgard@gmail.com>
Date: Sun, 14 Dec 2014 20:59:21 +0100
Subject: [PATCH] GLES: Implement a custom modulo function for PowerVR. Fixes
 blockiness, see #7153 and #7150

---
 GPU/GLES/FragmentShaderGenerator.cpp | 19 ++++++++++++++-----
 GPU/GLES/VertexShaderGenerator.cpp   |  7 +++++--
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/GPU/GLES/FragmentShaderGenerator.cpp b/GPU/GLES/FragmentShaderGenerator.cpp
index 330fbb229a..6ffd9fdb4a 100644
--- a/GPU/GLES/FragmentShaderGenerator.cpp
+++ b/GPU/GLES/FragmentShaderGenerator.cpp
@@ -487,6 +487,7 @@ void GenerateFragmentShader(char *buffer) {
 	const char *texture = "texture2D";
 	const char *texelFetch = NULL;
 	bool highpFog = false;
+	bool highpTexcoord = false;
 	bool bitwiseOps = false;
 
 #if defined(USING_GLES2)
@@ -511,7 +512,8 @@ void GenerateFragmentShader(char *buffer) {
 	// PowerVR needs highp to do the fog in MHU correctly.
 	// Others don't, and some can't handle highp in the fragment shader.
 	highpFog = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR;
-	
+	highpTexcoord = highpFog;
+
 	// GL_NV_shader_framebuffer_fetch available on mobile platform and ES 2.0 only but not desktop
 	if (gl_extensions.NV_shader_framebuffer_fetch) {
 		WRITE(p, "#extension GL_NV_shader_framebuffer_fetch : require\n");
@@ -635,9 +637,9 @@ void GenerateFragmentShader(char *buffer) {
 	}
 	if (doTexture) {
 		if (doTextureProjection)
-			WRITE(p, "%s mediump vec3 v_texcoord;\n", varying);
+			WRITE(p, "%s %s vec3 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
 		else
-			WRITE(p, "%s mediump vec2 v_texcoord;\n", varying);
+			WRITE(p, "%s %s vec2 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
 	}
 
 	if (!g_Config.bFragmentTestCache) {
@@ -668,6 +670,11 @@ void GenerateFragmentShader(char *buffer) {
 		WRITE(p, "out vec4 fragColor0;\n");
 	}
 
+	// PowerVR needs a custom modulo function. For some reason, this has far higher precision than the builtin one.
+	if (gl_extensions.gpuVendor == GPU_VENDOR_POWERVR && gstate_c.needShaderTexClamp) {
+		WRITE(p, "float mymod(float a, float b) { return a - b * floor(a / b); }\n");
+	}
+
 	WRITE(p, "void main() {\n");
 
 	if (gstate.isModeClear()) {
@@ -700,15 +707,17 @@ void GenerateFragmentShader(char *buffer) {
 					vcoord = "1.0 - " + vcoord;
 				}
 
+				std::string modulo = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR ? "mymod" : "mod";
+
 				if (gstate.isTexCoordClampedS()) {
 					ucoord = "clamp(" + ucoord + ", u_texclamp.z, u_texclamp.x - u_texclamp.z)";
 				} else {
-					ucoord = "mod(" + ucoord + ", u_texclamp.x)";
+					ucoord = modulo + "(" + ucoord + ", u_texclamp.x)";
 				}
 				if (gstate.isTexCoordClampedT()) {
 					vcoord = "clamp(" + vcoord + ", u_texclamp.w, u_texclamp.y - u_texclamp.w)";
 				} else {
-					vcoord = "mod(" + vcoord + ", u_texclamp.y)";
+					vcoord = modulo + "(" + vcoord + ", u_texclamp.y)";
 				}
 				if (textureAtOffset) {
 					ucoord = "(" + ucoord + " + u_texclampoff.x)";
diff --git a/GPU/GLES/VertexShaderGenerator.cpp b/GPU/GLES/VertexShaderGenerator.cpp
index 68b3ab6139..e8f1d6841f 100644
--- a/GPU/GLES/VertexShaderGenerator.cpp
+++ b/GPU/GLES/VertexShaderGenerator.cpp
@@ -163,6 +163,7 @@ void GenerateVertexShader(int prim, u32 vertType, char *buffer, bool useHWTransf
 	const char *attribute = "attribute";
 	const char * const * boneWeightDecl = boneWeightAttrDecl;
 	bool highpFog = false;
+	bool highpTexcoord = false;
 
 #if defined(USING_GLES2)
 	// Let's wait until we have a real use for this.
@@ -178,6 +179,8 @@ void GenerateVertexShader(int prim, u32 vertType, char *buffer, bool useHWTransf
 	// PowerVR needs highp to do the fog in MHU correctly.
 	// Others don't, and some can't handle highp in the fragment shader.
 	highpFog = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR;
+	highpTexcoord = gl_extensions.gpuVendor == GPU_VENDOR_POWERVR;
+
 #elif !defined(FORCE_OPENGL_2_0)
 	if (gl_extensions.VersionGEThan(3, 3, 0)) {
 		glslES30 = true;
@@ -339,9 +342,9 @@ void GenerateVertexShader(int prim, u32 vertType, char *buffer, bool useHWTransf
 	}
 	if (doTexture) {
 		if (doTextureProjection)
-			WRITE(p, "%s mediump vec3 v_texcoord;\n", varying);
+			WRITE(p, "%s %s vec3 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
 		else
-			WRITE(p, "%s mediump vec2 v_texcoord;\n", varying);
+			WRITE(p, "%s %s vec2 v_texcoord;\n", varying, highpTexcoord ? "highp" : "mediump");
 	}