Apply rendered CLUT to depal as well.

GPU/Common/DepalettizeShaderCommon.cpp

@@ -39,6 +39,7 @@ void GenerateDepalShader300(char *buffer, GEBufferFormat pixelFormat) {
 	WRITE(p, "out vec4 fragColor0;\n");
 	WRITE(p, "uniform sampler2D tex;\n");
 	WRITE(p, "uniform sampler2D pal;\n");
+	WRITE(p, "uniform vec2 u_offset;\n");
 
 	WRITE(p, "void main() {\n");
 	WRITE(p, "  vec4 color = texture(tex, v_texcoord0);\n");
@@ -99,7 +100,7 @@ void GenerateDepalShader300(char *buffer, GEBufferFormat pixelFormat) {
 		WRITE(p, ";\n");
 	}
 
-	WRITE(p, "  fragColor0 = texture(pal, vec2((float(index) + 0.5) * (1.0 / %f), 0.0));\n", texturePixels);
+	WRITE(p, "  fragColor0 = texture(pal, vec2((float(index) + 0.5) * (1.0 / %f) * u_offset.x + u_offset.y, 0.0));\n", texturePixels);
 	WRITE(p, "}\n");
 }
 
@@ -228,17 +229,19 @@ void GenerateDepalShaderFloat(char *buffer, GEBufferFormat pixelFormat, ShaderLa
 		WRITE(p, "varying vec2 v_texcoord0;\n");
 		WRITE(p, "uniform sampler2D tex;\n");
 		WRITE(p, "uniform sampler2D pal;\n");
+		WRITE(p, "uniform vec2 u_offset;\n");
 		WRITE(p, "void main() {\n");
 		WRITE(p, "  vec4 index = texture2D(tex, v_texcoord0);\n");
-		WRITE(p, "  float coord = (%s * %f)%s;\n", lookupMethod, index_multiplier, offset);
+		WRITE(p, "  float coord = (%s * %f * u_offset.x)%s + u_offset.y;\n", lookupMethod, index_multiplier, offset);
 		WRITE(p, "  gl_FragColor = texture2D(pal, vec2(coord, 0.0));\n");
 		WRITE(p, "}\n");
 	} else if (lang == HLSL_DX9) {
 		WRITE(p, "sampler tex: register(s0);\n");
 		WRITE(p, "sampler pal: register(s1);\n");
+		WRITE(p, "float2 u_offset : register(c%i);\n", CONST_PS_DEPAL_OFFSET);
 		WRITE(p, "float4 main(float2 v_texcoord0 : TEXCOORD0) : COLOR0 {\n");
 		WRITE(p, "  float4 index = tex2D(tex, v_texcoord0);\n");
-		WRITE(p, "  float coord = (%s * %f)%s;\n", lookupMethod, index_multiplier, offset);
+		WRITE(p, "  float coord = (%s * %f * u_offset.x)%s + u_offset.y;\n", lookupMethod, index_multiplier, offset);
 		WRITE(p, "  return tex2D(pal, float2(coord, 0.0)).bgra;\n");
 		WRITE(p, "}\n");
 	}

GPU/Directx9/TextureCacheDX9.cpp

@@ -1000,7 +1000,19 @@ void TextureCacheDX9::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFrame
 	}
 
 	if (pshader) {
-		LPDIRECT3DTEXTURE9 clutTexture = depalShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBuf_);
+		LPDIRECT3DTEXTURE9 clutTexture = nullptr;
+
+		VirtualFramebuffer *clutVfb = nullptr;
+		for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
+			auto clutFramebuffer = fbCache_[i];
+			if (clutFramebuffer->fb_address == clutRenderAddress_) {
+				clutVfb = clutFramebuffer;
+			}
+		}
+
+		if (!clutVfb) {
+			clutTexture = depalShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBuf_);
+		}
 
 		FBO_DX9 *depalFBO = framebufferManager_->GetTempFBO(framebuffer->renderWidth, framebuffer->renderHeight, FBO_8888);
 		fbo_bind_as_render_target(depalFBO);
@@ -1014,7 +1026,22 @@ void TextureCacheDX9::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFrame
 
 		shaderApply.Use(depalShaderCache_->GetDepalettizeVertexShader());
 
-		pD3Ddevice->SetTexture(1, clutTexture);
+		if (clutVfb) {
+			// We scale by the width of the CLUT - to map 0.0 -> 0, 1.0 -> 255.
+			// If the width is 256, 255 is right (with offset.)  We aim for the texel centers.
+			float texel_mult = 255.0f / (float)clutVfb->bufferWidth;
+			const float f[4] = { texel_mult, 0.0f, 0.0f, 0.0f };
+			pD3Ddevice->SetPixelShaderConstantF(CONST_PS_DEPAL_OFFSET, f, 1);
+		} else {
+			const float f[4] = { 1.0f, 0.0f, 0.0f, 0.0f };
+			pD3Ddevice->SetPixelShaderConstantF(CONST_PS_DEPAL_OFFSET, f, 1);
+		}
+
+		if (clutVfb) {
+			pD3Ddevice->SetTexture(1, fbo_get_color_texture(clutVfb->fbo_dx9));
+		} else {
+			pD3Ddevice->SetTexture(1, clutTexture);
+		}
 		pD3Ddevice->SetSamplerState(1, D3DSAMP_MINFILTER, D3DTEXF_POINT);
 		pD3Ddevice->SetSamplerState(1, D3DSAMP_MAGFILTER, D3DTEXF_POINT);
 		pD3Ddevice->SetSamplerState(1, D3DSAMP_MIPFILTER, D3DTEXF_NONE);

GPU/GLES/TextureCache.cpp

@@ -1068,7 +1068,19 @@ void TextureCache::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuf
 		depal = depalShaderCache_->GetDepalettizeShader(clutFormat, framebuffer->drawnFormat);
 	}
 	if (depal) {
-		GLuint clutTexture = depalShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBuf_);
+		GLuint clutTexture = 0;
+
+		VirtualFramebuffer *clutVfb = nullptr;
+		for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
+			auto clutFramebuffer = fbCache_[i];
+			if (clutFramebuffer->fb_address == clutRenderAddress_) {
+				clutVfb = clutFramebuffer;
+			}
+		}
+
+		if (!clutVfb) {
+			clutTexture = depalShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBuf_);
+		}
 		FBO *depalFBO = framebufferManager_->GetTempFBO(framebuffer->renderWidth, framebuffer->renderHeight, FBO_8888);
 		fbo_bind_as_render_target(depalFBO);
 		shaderManager_->DirtyLastShader();
@@ -1077,8 +1089,25 @@ void TextureCache::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuf
 		shaderApply.ApplyBounds(gstate_c.vertBounds, gstate_c.curTextureXOffset, gstate_c.curTextureYOffset);
 		shaderApply.Use();
 
+		if (depal->u_offset != -1) {
+			if (clutVfb) {
+				// We scale by the width of the CLUT - to map 0.0 -> 0, 1.0 -> 255.
+				// If the width is 256, 255 is right (with offset.)  We aim for the texel centers.
+				float texel_mult = 255.0f / (float)clutVfb->bufferWidth;
+				glUniform2f(depal->u_offset, texel_mult, 0.0f);
+			} else {
+				glUniform2f(depal->u_offset, 1.0f, 0.0f);
+			}
+		}
+
 		glActiveTexture(GL_TEXTURE3);
-		glBindTexture(GL_TEXTURE_2D, clutTexture);
+		if (clutVfb) {
+			fbo_bind_color_as_texture(clutVfb->fbo, 0);
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+		} else {
+			glBindTexture(GL_TEXTURE_2D, clutTexture);
+		}
 		glActiveTexture(GL_TEXTURE0);
 
 		framebufferManager_->BindFramebufferColor(GL_TEXTURE0, gstate.getFrameBufRawAddress(), framebuffer, BINDFBCOLOR_SKIP_COPY);