From 70318fda375a3ed322afbfb54ebad6999bde4203 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sun, 12 Aug 2018 19:49:08 -0700
Subject: [PATCH] GLES: Enable shader depal with Vulkan fixes.

---
 GPU/GLES/FragmentShaderGeneratorGLES.cpp | 42 ++++++++++++++----------
 GPU/GLES/TextureCacheGLES.cpp            |  6 +---
 2 files changed, 26 insertions(+), 22 deletions(-)

diff --git a/GPU/GLES/FragmentShaderGeneratorGLES.cpp b/GPU/GLES/FragmentShaderGeneratorGLES.cpp
index 0c11d82344..674cbaa42d 100644
--- a/GPU/GLES/FragmentShaderGeneratorGLES.cpp
+++ b/GPU/GLES/FragmentShaderGeneratorGLES.cpp
@@ -346,29 +346,37 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, uint64_t *uniform
 				doTextureProjection = false;
 			}
 
-			if (doTextureProjection) {
-				WRITE(p, "  vec4 t = %sProj(tex, %s);\n", texture, texcoord);
-				if (shaderDepal) {
-					WRITE(p, "  vec4 t1 = %sProjOffset(tex, %s, ivec2(1, 0));\n", texture, texcoord);
-					WRITE(p, "  vec4 t2 = %sProjOffset(tex, %s, ivec2(0, 1));\n", texture, texcoord);
-					WRITE(p, "  vec4 t3 = %sProjOffset(tex, %s, ivec2(1, 1));\n", texture, texcoord);
+			if (!shaderDepal) {
+				if (doTextureProjection) {
+					WRITE(p, "  vec4 t = %sProj(tex, %s);\n", texture, texcoord);
+				} else {
+					WRITE(p, "  vec4 t = %s(tex, %s.xy);\n", texture, texcoord);
 				}
 			} else {
-				WRITE(p, "  vec4 t = %s(tex, %s.xy);\n", texture, texcoord);
-				if (shaderDepal) {
-					WRITE(p, "  vec4 t1 = %sOffset(tex, %s.xy, ivec2(1, 0));\n", texture, texcoord);
-					WRITE(p, "  vec4 t2 = %sOffset(tex, %s.xy, ivec2(0, 1));\n", texture, texcoord);
-					WRITE(p, "  vec4 t3 = %sOffset(tex, %s.xy, ivec2(1, 1));\n", texture, texcoord);
+				if (doTextureProjection) {
+					// We don't use textureProj because we need better control and it's probably not much of a savings anyway.
+					WRITE(p, "  vec2 uv = %s.xy/%s.z;\n  vec2 uv_round;\n", texcoord, texcoord);
+				} else {
+					WRITE(p, "  vec2 uv = %s.xy;\n  vec2 uv_round;\n", texcoord);
 				}
-			}
-
-			if (shaderDepal) {
+				WRITE(p, "  vec2 tsize = textureSize(tex, 0);\n");
+				WRITE(p, "  vec2 fraction;\n");
+				WRITE(p, "  bool bilinear = (u_depal >> 31) != 0;\n");
+				WRITE(p, "  if (bilinear) {\n");
+				WRITE(p, "    uv_round = uv * tsize - vec2(0.5, 0.5);\n");
+				WRITE(p, "    fraction = fract(uv_round);\n");
+				WRITE(p, "    uv_round = (uv_round - fraction + vec2(0.5, 0.5)) / tsize;\n");  // We want to take our four point samples at pixel centers.
+				WRITE(p, "  } else {\n");
+				WRITE(p, "    uv_round = uv;\n");
+				WRITE(p, "  }\n");
+				WRITE(p, "  vec4 t = %s(tex, uv_round);\n", texture);
+				WRITE(p, "  vec4 t1 = %sOffset(tex, uv_round, ivec2(1, 0));\n", texture);
+				WRITE(p, "  vec4 t2 = %sOffset(tex, uv_round, ivec2(0, 1));\n", texture);
+				WRITE(p, "  vec4 t3 = %sOffset(tex, uv_round, ivec2(1, 1));\n", texture);
 				WRITE(p, "  int depalMask = (u_depal & 0xFF);\n");
 				WRITE(p, "  int depalShift = ((u_depal >> 8) & 0xFF);\n");
 				WRITE(p, "  int depalOffset = (((u_depal >> 16) & 0xFF) << 4);\n");
 				WRITE(p, "  int depalFmt = ((u_depal >> 24) & 0x3);\n");
-				WRITE(p, "  bool bilinear = (u_depal >> 31) != 0;\n");
-				WRITE(p, "  vec2 fraction = fract(%s.xy * vec2(textureSize(tex, 0).xy));\n", texcoord);
 				WRITE(p, "  ivec4 col; int index0; int index1; int index2; int index3;\n");
 				WRITE(p, "  switch (depalFmt) {\n");  // We might want to include fmt in the shader ID if this is a performance issue.
 				WRITE(p, "  case 0:\n");  // 565
@@ -422,7 +430,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, uint64_t *uniform
 				WRITE(p, "  };\n");
 				WRITE(p, "  index0 = ((index0 >> depalShift) & depalMask) | depalOffset;\n");
 				WRITE(p, "  t = texelFetch(pal, ivec2(index0, 0), 0);\n");
-				WRITE(p, "  if (bilinear) {\n");
+				WRITE(p, "  if (bilinear && !(index0 == index1 && index1 == index2 && index2 == index3)) {\n");
 				WRITE(p, "    index1 = ((index1 >> depalShift) & depalMask) | depalOffset;\n");
 				WRITE(p, "    index2 = ((index2 >> depalShift) & depalMask) | depalOffset;\n");
 				WRITE(p, "    index3 = ((index3 >> depalShift) & depalMask) | depalOffset;\n");
diff --git a/GPU/GLES/TextureCacheGLES.cpp b/GPU/GLES/TextureCacheGLES.cpp
index 659f9540fe..5f1c3e4a0f 100644
--- a/GPU/GLES/TextureCacheGLES.cpp
+++ b/GPU/GLES/TextureCacheGLES.cpp
@@ -443,11 +443,7 @@ void TextureCacheGLES::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFram
 	DepalShader *depal = nullptr;
 	uint32_t clutMode = gstate.clutformat & 0xFFFFFF;
 
-#if 0
-	bool useShaderDepal = gstate_c.Supports(GPU_SUPPORTS_GLSL_ES_300);
-#else
-	bool useShaderDepal = false;
-#endif
+	bool useShaderDepal = framebufferManager_->GetCurrentRenderVFB() != framebuffer && gstate_c.Supports(GPU_SUPPORTS_GLSL_ES_300);
 
 	if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
 		if (useShaderDepal) {