BC4 and BC5 compression

1 files changed, 127 insertions, 3 deletions

diff --git a/convtexture.c b/convtexture.c
index a321a9d..b3be7dc 100644
--- a/convtexture.c
+++ b/convtexture.c
@@ -23,6 +23,11 @@ typedef struct {
 	unsigned char indices[4 * 4];
 } Block;
 
+typedef struct {
+	float start, end;
+	unsigned char indices[4 * 4];
+} Block_BC4;
+
 const char* format_names[] = {
 #define x(n) #n,
 	texture_format_xmacro()
@@ -36,6 +41,10 @@ unsigned encode_endpoint(const vec3* v) {
 		((unsigned)(v->b * 31.9999f));
 }
 
+unsigned encode_endpoint_bc4(float v) {
+	return (unsigned)(v * 255.0f);
+}
+
 void get_block(Block* block, Colour* pixels, int stride) {
 	int x, y, i;
 	vec3 cols[4 * 4];
@@ -125,13 +134,97 @@ void compress_block(const Block* block, FILE* f) {
 	fwrite(&start, 1, 2, f);
 	fwrite(&end,   1, 2, f);
 	for (i = 0; i < 4 * 4; i++) {
-		int idx = (4 * 4 - 1) - i;
-		idx = i;
-		indices |= block->indices[i] << (idx * 2);
+		indices |= block->indices[i] << (i * 2);
 	}
 	fwrite(&indices, 1, 4, f);
 }
 
+void get_block_bc4(
+	Block_BC4* block,
+	Colour* pixels,
+	int stride,
+	int channel
+) {
+	int x, y;
+	float palette[8];
+	palette[0] = 0.08f;
+	palette[1] = 0.92f;
+	float biggest = -1.0f;
+	float smallest = 1.0f;
+	for (y = 0; y < 4; y++) {
+		for (x = 0; x < 4; x++) {
+			uint8_t* pixel = (uint8_t*)&pixels[x + y * stride];
+			float v = (float)pixel[channel] / 255.0f;
+			if (v > biggest)  biggest = v;
+			if (v < smallest) smallest = v;
+			if (v > palette[0]) palette[0] = v;
+			if (v < palette[1]) palette[1] = v;
+		}
+	}
+	if (biggest == smallest) {
+		int i, idx = 1;
+		block->end = biggest;
+		if (palette[0] > 0.9f) {
+			idx = 0;
+			block->start = smallest;
+			block->end = 0.0f;
+		} else
+			block->start = 1.0f;
+		assert(block->start > block->end);
+		for (i = 0; i < 16; i++)
+			block->indices[i] = idx;
+	}
+	if (biggest > 0.92f && smallest < 0.08f) {
+		float t = palette[1];
+		palette[1] = palette[0];
+		palette[0] = t;
+		palette[2] = (4.0f * palette[0] + 1.0f * palette[1]) / 5.0f;
+		palette[3] = (3.0f * palette[0] + 2.0f * palette[1]) / 5.0f;
+		palette[4] = (2.0f * palette[0] + 3.0f * palette[1]) / 5.0f;
+		palette[5] = (1.0f * palette[0] + 4.0f * palette[1]) / 5.0f;
+		palette[6] = 0.0f;
+		palette[7] = 1.0f;
+	} else {
+		palette[2] = (6.0f * palette[0] + 1.0f * palette[1]) / 7.0f;
+		palette[3] = (5.0f * palette[0] + 2.0f * palette[1]) / 7.0f;
+		palette[4] = (4.0f * palette[0] + 3.0f * palette[1]) / 7.0f;
+		palette[5] = (3.0f * palette[0] + 4.0f * palette[1]) / 7.0f;
+		palette[6] = (2.0f * palette[0] + 5.0f * palette[1]) / 7.0f;
+		palette[7] = (1.0f * palette[0] + 6.0f * palette[1]) / 7.0f;
+	}
+	block->start = palette[0];
+	block->end = palette[1];
+	for (y = 0; y < 4; y++) {
+		for (x = 0; x < 4; x++) {
+			uint8_t* pixel = (uint8_t*)&pixels[x + y * stride];
+			float v = (float)pixel[channel] / 255.0f;
+			float d = 100.0f;
+			int i;
+			for (i = 0; i < 8; i++) {
+				float d2 = (v - palette[i]);
+				d2 = d2 < 0.0f? -d2: d2;
+				if (d2 < d) {
+					d = d2;
+					block->indices[x + y * 4] = i;
+				}
+			}
+		}
+	}
+}
+
+void compress_block_bc4(const Block_BC4* block, FILE* f) {
+	int i;
+	unsigned start = encode_endpoint_bc4(block->start);
+	unsigned end   = encode_endpoint_bc4(block->end);
+	uint64_t indices = 0;
+	fwrite(&start, 1, 1, f);
+	fwrite(&end,   1, 1, f);
+	for (i = 0; i < 4 * 4; i++) {
+		indices |= ((uint64_t)block->indices[i]) << (i * 3);
+	} /* 48 bits of indices */
+	fwrite(&indices, 1, 6, f);
+}
+
 void compress_bc1(Colour* pixels, int w, int h, FILE* f) {
 	int x, y, cw, ch;
 	Block block;
@@ -145,6 +238,31 @@ void compress_bc1(Colour* pixels, int w, int h, FILE* f) {
 	}
 }
 
+void compress_bc4(Colour* pixels, int w, int h, FILE* f) {
+	int x, y, cw = w / 4, ch = h / 4;
+	Block_BC4 block;
+	for (y = 0; y < ch; y++) {
+		for (x = 0; x < cw; x++) {
+			get_block_bc4(&block, &pixels[x * 4 + y * 4 * w], w, 0);
+			compress_block_bc4(&block, f);
+		}
+	}
+}
+
+void compress_bc5(Colour* pixels, int w, int h, FILE* f) {
+	int x, y, cw = w / 4, ch = h / 4;
+	Block_BC4 a, b;
+	for (y = 0; y < ch; y++) {
+		for (x = 0; x < cw; x++) {
+			Colour* p = &pixels[x * 4 + y * 4 * w];
+			get_block_bc4(&a, p, w, 0);
+			get_block_bc4(&b, p, w, 1);
+			compress_block_bc4(&a, f);
+			compress_block_bc4(&b, f);
+		}
+	}
+}
+
 void convert(Image* image, Texture_Format target, FILE* f) {
 	fwrite("TXTR", 1, 4, f);
 	fwrite(&image->w, 1, 4, f);
@@ -154,6 +272,12 @@ void convert(Image* image, Texture_Format target, FILE* f) {
 		case texture_format_bc1:
 			compress_bc1(image->pixels, image->w, image->h, f);
 			break;
+		case texture_format_bc4:
+			compress_bc4(image->pixels, image->w, image->h, f);
+			break;
+		case texture_format_bc5:
+			compress_bc5(image->pixels, image->w, image->h, f);
+			break;
 		default:
 			print_err("Unsupported target format.\n");
 			pbreak(40);