Merge pull request #18304 from hrydgard/manhunt-color-ramp-fix

Extend the Test Drive color ramp smoother to detect up to 3 ramps in a texture

GPU/Common/FragmentShaderGenerator.cpp

@@ -717,6 +717,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 				// lookup with the filtered value once.
 				p.F("  vec4 t = ").SampleTexture2D("tex", "uv").C(";\n");
 				p.C("  uint depalShift = (u_depal_mask_shift_off_fmt >> 0x8u) & 0xFFu;\n");
+				p.C("  uint depalOffset = ((u_depal_mask_shift_off_fmt >> 0x10u) & 0xFFu) << 0x4u;\n");
 				p.C("  uint depalFmt = (u_depal_mask_shift_off_fmt >> 0x18u) & 0x3u;\n");
 				p.C("  float index0 = t.r;\n");
 				p.C("  float factor = 31.0 / 256.0;\n");
@@ -727,7 +728,8 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 				p.C("    if (depalShift == 0x5u) { index0 = t.g; }\n");
 				p.C("    else if (depalShift == 0xAu) { index0 = t.b; }\n");
 				p.C("  }\n");
-				p.F("  t = ").SampleTexture2D("pal", "vec2(index0 * factor * 0.5, 0.0)").C(";\n");  // 0.5 for 512-entry CLUT.
+				p.C("  float offset = float(depalOffset) / 256.0;\n");
+				p.F("  t = ").SampleTexture2D("pal", "vec2((index0 * factor + offset) * 0.5 + 0.5 / 512.0, 0.0)").C(";\n");  // 0.5 for 512-entry CLUT.
 				break;
 			case ShaderDepalMode::NORMAL:
 				if (doTextureProjection) {

GPU/Common/TextureCacheCommon.cpp

@@ -2194,10 +2194,11 @@ static bool CanDepalettize(GETextureFormat texFormat, GEBufferFormat bufferForma
 // If the palette is detected as a smooth ramp, we can interpolate for higher color precision.
 // But we only do it if the mask/shift exactly matches a color channel, else something different might be going
 // on and we definitely don't want to interpolate.
-// Great enhancement for Test Drive.
-static bool CanUseSmoothDepal(const GPUgstate &gstate, GEBufferFormat framebufferFormat, int rampLength) {
-	if (gstate.getClutIndexStartPos() == 0 &&
-		gstate.getClutIndexMask() < rampLength) {
+// Great enhancement for Test Drive and Manhunt 2.
+static bool CanUseSmoothDepal(const GPUgstate &gstate, GEBufferFormat framebufferFormat, const ClutTexture &clutTexture) {
+	for (int i = 0; i < ClutTexture::MAX_RAMPS; i++) {
+		if (gstate.getClutIndexStartPos() == clutTexture.rampStarts[i] &&
+			gstate.getClutIndexMask() < clutTexture.rampLengths[i]) {
 			switch (framebufferFormat) {
 			case GE_FORMAT_565:
 				if (gstate.getClutIndexShift() == 0 || gstate.getClutIndexShift() == 11) {
@@ -2216,6 +2217,7 @@ static bool CanUseSmoothDepal(const GPUgstate &gstate, GEBufferFormat framebuffe
 				break;
 			}
 		}
+	}
 	return false;
 }
 
@@ -2253,7 +2255,7 @@ void TextureCacheCommon::ApplyTextureFramebuffer(VirtualFramebuffer *framebuffer
 	if (need_depalettize) {
 		if (clutRenderAddress_ == 0xFFFFFFFF) {
 			clutTexture = textureShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBufRaw_);
-			smoothedDepal = CanUseSmoothDepal(gstate, framebuffer->fb_format, clutTexture.rampLength);
+			smoothedDepal = CanUseSmoothDepal(gstate, framebuffer->fb_format, clutTexture);
 		} else {
 			// The CLUT texture is dynamic, it's the framebuffer pointed to by clutRenderAddress.
 			// Instead of texturing directly from that, we copy to a temporary CLUT texture.

GPU/Common/TextureShaderCommon.cpp

@@ -95,21 +95,27 @@ ClutTexture TextureShaderCache::GetClutTexture(GEPaletteFormat clutFormat, const
 		break;
 	}
 
+
+	for (int i = 0; i < 3; i++) {
+		tex->rampLengths[i] = 0;
+		tex->rampStarts[i] = 0;
+	}
+	// Quick check for how many continuously growing entries we have at the start.
+	// Bilinearly filtering CLUTs only really makes sense for this kind of ramp.
+	int i = 0;
+	for (int j = 0; j < ClutTexture::MAX_RAMPS; j++) {
+		tex->rampStarts[j] = i;
 		int lastR = 0;
 		int lastG = 0;
 		int lastB = 0;
 		int lastA = 0;
-
-	int rampLength = 0;
-	// Quick check for how many continuously growing entries we have at the start.
-	// Bilinearly filtering CLUTs only really makes sense for this kind of ramp.
-	for (int i = 0; i < maxClutEntries; i++) {
-		rampLength = i;
+		for (; i < maxClutEntries; i++) {
 			int r = desc.initData[0][i * 4];
 			int g = desc.initData[0][i * 4 + 1];
 			int b = desc.initData[0][i * 4 + 2];
 			int a = desc.initData[0][i * 4 + 3];
 			if (r < lastR || g < lastG || b < lastB || a < lastA) {
+				lastR = r; lastG = g; lastB = b; lastA = a;
 				break;
 			} else {
 				lastR = r;
@@ -118,10 +124,14 @@ ClutTexture TextureShaderCache::GetClutTexture(GEPaletteFormat clutFormat, const
 				lastA = a;
 			}
 		}
+		tex->rampLengths[j] = i - tex->rampStarts[j];
+		if (i >= maxClutEntries) {
+			break;
+		}
+	}
 
 	tex->texture = draw_->CreateTexture(desc);
 	tex->lastFrame = gpuStats.numFlips;
-	tex->rampLength = rampLength;
 
 	texCache_[clutId] = tex;
 	return *tex;

GPU/Common/TextureShaderCommon.h

@@ -29,11 +29,14 @@
 #include "GPU/Common/ShaderCommon.h"
 #include "GPU/Common/DepalettizeShaderCommon.h"
 
+
 class ClutTexture {
 public:
+	enum { MAX_RAMPS = 3 };
 	Draw::Texture *texture;
 	int lastFrame;
-	int rampLength;
+	int rampLengths[MAX_RAMPS];
+	int rampStarts[MAX_RAMPS];
 };
 
 // For CLUT depal shaders, and other pre-bind texture shaders.