From cbb62ea555221dd023085564207c634779d45314 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Tue, 30 Nov 2021 22:05:03 +0100
Subject: [PATCH 01/14] Remove QOI_RUN_16, add new QOI_GDIFF_16 op

---
 qoi.h | 165 +++++++++++++++++++++++++++++++---------------------------
 1 file changed, 89 insertions(+), 76 deletions(-)

diff --git a/qoi.h b/qoi.h
index 0aec7287..a16262f6 100644
--- a/qoi.h
+++ b/qoi.h
@@ -122,24 +122,14 @@ The possible chunks are:
 6-bit index into the color index array: 0..63
 
 
- - QOI_RUN_8 -------------
+ - QOI_RUN ---------------
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
-|----------+--------------|
-|  0  1  0 |     run      |
+|-------+-----------------|
+|  0  1 |       run       |
 
-3-bit tag b010
-5-bit run-length repeating the previous pixel: 1..32
-
-
- - QOI_RUN_16 --------------------------------------
-|         Byte[0]         |         Byte[1]         |
-|  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
-|----------+----------------------------------------|
-|  0  1  1 |                 run                    |
-
-3-bit tag b011
-13-bit run-length repeating the previous pixel: 33..8224
+2-bit tag b01
+6-bit run-length repeating the previous pixel: 1..64
 
 
  - QOI_DIFF_8 ------------
@@ -160,14 +150,33 @@ so "1 - 2" will result in 255, while "255 + 1" will result in 0.
  - QOI_DIFF_16 -------------------------------------
 |         Byte[0]         |         Byte[1]         |
 |  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
-|----------+--------------|------------ +-----------|
-|  1  1  0 |   red diff   |  green diff | blue diff |
+|-------------+-----------|------------ +-----------|
+|  1  1  0  0 |  red diff |  green diff | blue diff |
 
-3-bit tag b110
-5-bit   red channel difference from the previous pixel between -16..15
+4-bit tag b1100
+4-bit   red channel difference from the previous pixel between -8..7
 4-bit green channel difference from the previous pixel between -8..7
 4-bit  blue channel difference from the previous pixel between -8..7
 
+The difference to the current channel values are using a wraparound operation, 
+so "5 - 8" will result in 253, while "250 + 7" will result in 1.
+
+
+ - QOI_GDIFF_16 ------------------------------------
+|         Byte[0]         |         Byte[1]         |
+|  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
+|-------------+--------+-------------------+--------|
+|  1  1  0  1 |  dr-dg |    green diff     |  db-dg |
+
+4-bit tag b1101
+3-bit   red channel difference minus green channel difference -4..3
+6-bit green channel difference from the previous pixel -32..31
+3-bit  blue channel difference minus green channel difference -4..3
+
+The green channel is used to indicate the general direction of change and gets
+a few more bits. dr and db base their diffs off of the green channel diff. E.g.
+  dr = (last_px.r - cur_px.r) - (last_px.g - cur_px.g)
+
 The difference to the current channel values are using a wraparound operation, 
 so "10 - 13" will result in 253, while "250 + 7" will result in 1.
 
@@ -309,17 +318,16 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
 	#define QOI_FREE(p)    free(p)
 #endif
 
-#define QOI_INDEX   0x00 // 00xxxxxx
-#define QOI_RUN_8   0x40 // 010xxxxx
-#define QOI_RUN_16  0x60 // 011xxxxx
-#define QOI_DIFF_8  0x80 // 10xxxxxx
-#define QOI_DIFF_16 0xc0 // 110xxxxx
-#define QOI_DIFF_24 0xe0 // 1110xxxx
-#define QOI_COLOR   0xf0 // 1111xxxx
+#define QOI_INDEX     0x00 // 00xxxxxx
+#define QOI_RUN       0x40 // 01xxxxxx
+#define QOI_DIFF_8    0x80 // 10xxxxxx
+#define QOI_DIFF_16   0xc0 // 1100xxxx
+#define QOI_GDIFF_16  0xd0 // 1101xxxx
+#define QOI_DIFF_24   0xe0 // 1110xxxx
+#define QOI_COLOR     0xf0 // 1111xxxx
 
-#define QOI_MASK_2  0xc0 // 11000000
-#define QOI_MASK_3  0xe0 // 11100000
-#define QOI_MASK_4  0xf0 // 11110000
+#define QOI_MASK_2    0xc0 // 11000000
+#define QOI_MASK_4    0xf0 // 11110000
 
 #define QOI_COLOR_HASH(C) (C.rgba.r ^ C.rgba.g ^ C.rgba.b ^ C.rgba.a)
 #define QOI_MAGIC \
@@ -403,17 +411,9 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 
 		if ( 
 			run > 0 && 
-			(run == 0x2020 || px.v != px_prev.v || px_pos == px_end)
+			(run == 64 || px.v != px_prev.v || px_pos == px_end)
 		) {
-			if (run < 33) {
-				run -= 1;
-				bytes[p++] = QOI_RUN_8 | run;
-			}
-			else {
-				run -= 33;
-				bytes[p++] = QOI_RUN_16 | run >> 8;
-				bytes[p++] = run;
-			}
+			bytes[p++] = QOI_RUN | (run - 1);
 			run = 0;
 		}
 
@@ -426,39 +426,49 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 			else {
 				index[index_pos] = px;
 
-				int vr = px.rgba.r - px_prev.rgba.r;
-				int vg = px.rgba.g - px_prev.rgba.g;
-				int vb = px.rgba.b - px_prev.rgba.b;
-				int va = px.rgba.a - px_prev.rgba.a;
-				
+				char vr = px.rgba.r - px_prev.rgba.r;
+				char vg = px.rgba.g - px_prev.rgba.g;
+				char vb = px.rgba.b - px_prev.rgba.b;
+				char va = px.rgba.a - px_prev.rgba.a;
+
+				char vg_r = vr - vg;
+				char vg_b = vb - vg;
+
 				if (
+					va == 0 && 
+					vr > -3 && vr < 2 &&
+					vg > -3 && vg < 2 && 
+					vb > -3 && vb < 2
+				) {
+					bytes[p++] = QOI_DIFF_8 | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2);
+				}
+				else if (
+					va == 0 && 
+					vr > -9 && vr < 8 && 
+					vg > -9 && vg < 8 && 
+					vb > -9 && vb < 8
+				) {
+					bytes[p++] = QOI_DIFF_16   | (vr + 8);
+					bytes[p++] = (vg + 8) << 4 | (vb + 8);
+				}
+				else if (
+					va == 0 && 
+					vg_r >  -5 && vg_r <  4 && 
+					vg   > -33 && vg   < 32 && 
+					vg_b >  -5 && vg_b <  4
+				) {
+					bytes[p++] = QOI_GDIFF_16   | (vg_r + 4) << 1 | (vg + 32) >> 5;
+					bytes[p++] = (vg + 32) << 3 | (vg_b + 4);
+				}
+				else if (
 					vr > -17 && vr < 16 && 
 					vg > -17 && vg < 16 && 
 					vb > -17 && vb < 16 && 
 					va > -17 && va < 16
 				) {
-					if (
-						va == 0 && 
-						vr > -3 && vr < 2 &&
-						vg > -3 && vg < 2 && 
-						vb > -3 && vb < 2
-					) {
-						bytes[p++] = QOI_DIFF_8 | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2);
-					}
-					else if (
-						va == 0 && 
-						vr > -17 && vr < 16 && 
-						vg >  -9 && vg <  8 && 
-						vb >  -9 && vb <  8
-					) {
-						bytes[p++] = QOI_DIFF_16   | (vr + 16);
-						bytes[p++] = (vg + 8) << 4 | (vb +  8);
-					}
-					else {
-						bytes[p++] = QOI_DIFF_24    | (vr + 16) >> 1;
-						bytes[p++] = (vr + 16) << 7 | (vg + 16) << 2 | (vb + 16) >> 3;
-						bytes[p++] = (vb + 16) << 5 | (va + 16);
-					}
+					bytes[p++] = QOI_DIFF_24    | (vr + 16) >> 1;
+					bytes[p++] = (vr + 16) << 7 | (vg + 16) << 2 | (vb + 16) >> 3;
+					bytes[p++] = (vb + 16) << 5 | (va + 16);
 				}
 				else {
 					bytes[p++] = QOI_COLOR | (vr ? 8 : 0) | (vg ? 4 : 0) | (vb ? 2 : 0) | (va ? 1 : 0);
@@ -531,23 +541,26 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 			if ((b1 & QOI_MASK_2) == QOI_INDEX) {
 				px = index[b1 ^ QOI_INDEX];
 			}
-			else if ((b1 & QOI_MASK_3) == QOI_RUN_8) {
-				run = (b1 & 0x1f);
-			}
-			else if ((b1 & QOI_MASK_3) == QOI_RUN_16) {
-				int b2 = bytes[p++];
-				run = (((b1 & 0x1f) << 8) | (b2)) + 32;
+			else if ((b1 & QOI_MASK_2) == QOI_RUN) {
+				run = (b1 & 0x3f);
 			}
 			else if ((b1 & QOI_MASK_2) == QOI_DIFF_8) {
 				px.rgba.r += ((b1 >> 4) & 0x03) - 2;
 				px.rgba.g += ((b1 >> 2) & 0x03) - 2;
 				px.rgba.b += ( b1       & 0x03) - 2;
 			}
-			else if ((b1 & QOI_MASK_3) == QOI_DIFF_16) {
+			else if ((b1 & QOI_MASK_4) == QOI_DIFF_16) {
 				int b2 = bytes[p++];
-				px.rgba.r += (b1 & 0x1f) - 16;
-				px.rgba.g += (b2 >> 4)   -  8;
-				px.rgba.b += (b2 & 0x0f) -  8;
+				px.rgba.r += (b1 & 0x0f) - 8;
+				px.rgba.g += (b2 >> 4)   - 8;
+				px.rgba.b += (b2 & 0x0f) - 8;
+			}
+			else if ((b1 & QOI_MASK_4) == QOI_GDIFF_16) {
+				int b2 = bytes[p++];
+				int vg = ((b1 & 0x01) << 5 | (b2 & 0xf8) >> 3) - 32;
+				px.rgba.r += vg - 4 + ((b1 & 0x0e) >> 1);
+				px.rgba.g += vg;
+				px.rgba.b += vg - 4 +  (b2 & 0x07);
 			}
 			else if ((b1 & QOI_MASK_4) == QOI_DIFF_24) {
 				int b2 = bytes[p++];

From f45f47c9f0ecc53df2ece721e8df6164bf87e005 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Mon, 6 Dec 2021 13:35:54 +0100
Subject: [PATCH 02/14] Recursive traversal; compression ratio; grand total;
 options to disable some features

---
 qoibench.c | 351 +++++++++++++++++++++++++++++++++++------------------
 1 file changed, 231 insertions(+), 120 deletions(-)

diff --git a/qoibench.c b/qoibench.c
index f48e2361..b60514ed 100644
--- a/qoibench.c
+++ b/qoibench.c
@@ -195,8 +195,12 @@ void png_decode_callback(png_structp png, png_bytep data, png_size_t length) {
 	read_data->pos += length;
 }
 
+void png_warning_callback(png_structp png_ptr, png_const_charp warning_msg) {
+	// Ingore warnings about sRGB profiles and such.
+}
+
 void *libpng_decode(void *data, int size, int *out_w, int *out_h) {	
-	png_structp png = png_create_read_struct(PNG_LIBPNG_VER_STRING, NULL, NULL, NULL);
+	png_structp png = png_create_read_struct(PNG_LIBPNG_VER_STRING, NULL, NULL, png_warning_callback);
 	if (!png) {
 		ERROR("png_create_read_struct");
 	}
@@ -308,6 +312,16 @@ void *fload(const char *path, int *out_size) {
 // -----------------------------------------------------------------------------
 // benchmark runner
 
+
+int opt_runs = 1;
+int opt_nopng = 0;
+int opt_nowarmup = 0;
+int opt_noverify = 0;
+int opt_nodecode = 0;
+int opt_noencode = 0;
+int opt_norecurse = 0;
+
+
 typedef struct {
 	uint64_t size;
 	uint64_t encode_time;
@@ -315,6 +329,8 @@ typedef struct {
 } benchmark_lib_result_t;
 
 typedef struct {
+	int count;
+	uint64_t raw_size;
 	uint64_t px;
 	int w;
 	int h;
@@ -324,12 +340,59 @@ typedef struct {
 } benchmark_result_t;
 
 
+void benchmark_print_result(benchmark_result_t res) {
+	res.px /= res.count;
+	res.raw_size /= res.count;
+	res.libpng.encode_time /= res.count;
+	res.libpng.decode_time /= res.count;
+	res.libpng.size /= res.count;
+	res.stbi.encode_time /= res.count;
+	res.stbi.decode_time /= res.count;
+	res.stbi.size /= res.count;
+	res.qoi.encode_time /= res.count;
+	res.qoi.decode_time /= res.count;
+	res.qoi.size /= res.count;
+
+	double px = res.px;
+	printf("        decode ms   encode ms   decode mpps   encode mpps   size kb    rate\n");
+	if (!opt_nopng) {
+		printf(
+			"libpng:  %8.1f    %8.1f      %8.2f      %8.2f  %8d   %4.1f%%\n", 
+			(double)res.libpng.decode_time/1000000.0, 
+			(double)res.libpng.encode_time/1000000.0, 
+			(res.libpng.decode_time > 0 ? px / ((double)res.libpng.decode_time/1000.0) : 0),
+			(res.libpng.encode_time > 0 ? px / ((double)res.libpng.encode_time/1000.0) : 0),
+			res.libpng.size/1024,
+			((double)res.libpng.size/(double)res.raw_size) * 100.0
+		);
+		printf(
+			"stbi:    %8.1f    %8.1f      %8.2f      %8.2f  %8d   %4.1f%%\n", 
+			(double)res.stbi.decode_time/1000000.0,
+			(double)res.stbi.encode_time/1000000.0,
+			(res.stbi.decode_time > 0 ? px / ((double)res.stbi.decode_time/1000.0) : 0),
+			(res.stbi.encode_time > 0 ? px / ((double)res.stbi.encode_time/1000.0) : 0),
+			res.stbi.size/1024,
+			((double)res.stbi.size/(double)res.raw_size) * 100.0
+		);
+	}
+	printf(
+		"qoi:     %8.1f    %8.1f      %8.2f      %8.2f  %8d   %4.1f%%\n", 
+		(double)res.qoi.decode_time/1000000.0,
+		(double)res.qoi.encode_time/1000000.0,
+		(res.qoi.decode_time > 0 ? px / ((double)res.qoi.decode_time/1000.0) : 0),
+		(res.qoi.encode_time > 0 ? px / ((double)res.qoi.encode_time/1000.0) : 0),
+		res.qoi.size/1024,
+		((double)res.qoi.size/(double)res.raw_size) * 100.0
+	);
+	printf("\n");
+}
+
 // Run __VA_ARGS__ a number of times and meassure the time taken. The first
 // run is ignored.
-#define BENCHMARK_FN(RUNS, AVG_TIME, ...) \
+#define BENCHMARK_FN(NOWARMUP, RUNS, AVG_TIME, ...) \
 	do { \
 		uint64_t time = 0; \
-		for (int i = 0; i <= RUNS; i++) { \
+		for (int i = NOWARMUP; i <= RUNS; i++) { \
 			uint64_t time_start = ns(); \
 			__VA_ARGS__ \
 			uint64_t time_end = ns(); \
@@ -341,7 +404,7 @@ typedef struct {
 	} while (0)
 
 
-benchmark_result_t benchmark_image(const char *path, int runs) {
+benchmark_result_t benchmark_image(const char *path) {
 	int encoded_png_size;
 	int encoded_qoi_size;
 	int w;
@@ -358,10 +421,22 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
 		}, &encoded_qoi_size);
 
 	if (!pixels || !encoded_qoi || !encoded_png) {
-		ERROR("Error decoding %s\n", path);
+		ERROR("Error decoding %s", path);
+	}
+
+	// Verify QOI Output
+	if (!opt_noverify) {
+		qoi_desc dc;
+		void *pixels_qoi = qoi_decode(encoded_qoi, encoded_qoi_size, &dc, 4);
+		if (memcmp(pixels, pixels_qoi, w * h * 4) != 0) {
+			ERROR("QOI roundtrip pixel missmatch for %s", path);
+		}
+		free(pixels_qoi);
 	}
 
 	benchmark_result_t res = {0};
+	res.count = 1;
+	res.raw_size = w * h * 4;
 	res.px = w * h;
 	res.w = w;
 	res.h = h;
@@ -369,51 +444,59 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
 
 	// Decoding
 
-	BENCHMARK_FN(runs, res.libpng.decode_time, {
-		int dec_w, dec_h;
-		void *dec_p = libpng_decode(encoded_png, encoded_png_size, &dec_w, &dec_h);
-		free(dec_p);
-	});
+	if (!opt_nodecode) {
+		if (!opt_nopng) {
+			BENCHMARK_FN(opt_nowarmup, opt_runs, res.libpng.decode_time, {
+				int dec_w, dec_h;
+				void *dec_p = libpng_decode(encoded_png, encoded_png_size, &dec_w, &dec_h);
+				free(dec_p);
+			});
 
-	BENCHMARK_FN(runs, res.stbi.decode_time, {
-		int dec_w, dec_h, dec_channels;
-		void *dec_p = stbi_load_from_memory(encoded_png, encoded_png_size, &dec_w, &dec_h, &dec_channels, 4);
-		free(dec_p);
-	});
+			BENCHMARK_FN(opt_nowarmup, opt_runs, res.stbi.decode_time, {
+				int dec_w, dec_h, dec_channels;
+				void *dec_p = stbi_load_from_memory(encoded_png, encoded_png_size, &dec_w, &dec_h, &dec_channels, 4);
+				free(dec_p);
+			});
+		}
 
-	BENCHMARK_FN(runs, res.qoi.decode_time, {
-		qoi_desc desc;
-		void *dec_p = qoi_decode(encoded_qoi, encoded_qoi_size, &desc, 4);
-		free(dec_p);
-	});
+		BENCHMARK_FN(opt_nowarmup, opt_runs, res.qoi.decode_time, {
+			qoi_desc desc;
+			void *dec_p = qoi_decode(encoded_qoi, encoded_qoi_size, &desc, 4);
+			free(dec_p);
+		});
+	}
 
 
 	// Encoding
 
-	BENCHMARK_FN(runs, res.libpng.encode_time, {
-		int enc_size;
-		void *enc_p = libpng_encode(pixels, w, h, &enc_size);
-		res.libpng.size = enc_size;
-		free(enc_p);
-	});
+	if (!opt_noencode) {
+		if (!opt_nopng) {
+			BENCHMARK_FN(opt_nowarmup, opt_runs, res.libpng.encode_time, {
+				int enc_size;
+				void *enc_p = libpng_encode(pixels, w, h, &enc_size);
+				res.libpng.size = enc_size;
+				free(enc_p);
+			});
 
-	BENCHMARK_FN(runs, res.stbi.encode_time, {
-		int enc_size = 0;
-		stbi_write_png_to_func(stbi_write_callback, &enc_size, w, h, 4, pixels, 0);
-		res.stbi.size = enc_size;
-	});
+			BENCHMARK_FN(opt_nowarmup, opt_runs, res.stbi.encode_time, {
+				int enc_size = 0;
+				stbi_write_png_to_func(stbi_write_callback, &enc_size, w, h, 4, pixels, 0);
+				res.stbi.size = enc_size;
+			});
+		}
 
-	BENCHMARK_FN(runs, res.qoi.encode_time, {
-		int enc_size;
-		void *enc_p = qoi_encode(pixels, &(qoi_desc){
-			.width = w,
-			.height = h, 
-			.channels = 4,
-			.colorspace = QOI_SRGB
-		}, &enc_size);
-		res.qoi.size = enc_size;
-		free(enc_p);
-	});
+		BENCHMARK_FN(opt_nowarmup, opt_runs, res.qoi.encode_time, {
+			int enc_size;
+			void *enc_p = qoi_encode(pixels, &(qoi_desc){
+				.width = w,
+				.height = h, 
+				.channels = 4,
+				.colorspace = QOI_SRGB
+			}, &enc_size);
+			res.qoi.size = enc_size;
+			free(enc_p);
+		});
+	}
 
 	free(pixels);
 	free(encoded_png);
@@ -422,102 +505,130 @@ benchmark_result_t benchmark_image(const char *path, int runs) {
 	return res;
 }
 
-void benchmark_print_result(const char *head, benchmark_result_t res) {
-	double px = res.px;
-	printf("## %s size: %dx%d\n", head, res.w, res.h);
-	printf("        decode ms   encode ms   decode mpps   encode mpps   size kb\n");
-	printf(
-		"libpng:  %8.1f    %8.1f      %8.2f      %8.2f  %8d\n", 
-		(double)res.libpng.decode_time/1000000.0, 
-		(double)res.libpng.encode_time/1000000.0, 
-		(res.libpng.decode_time > 0 ? px / ((double)res.libpng.decode_time/1000.0) : 0),
-		(res.libpng.encode_time > 0 ? px / ((double)res.libpng.encode_time/1000.0) : 0),
-		res.libpng.size/1024
-	);
-	printf(
-		"stbi:    %8.1f    %8.1f      %8.2f      %8.2f  %8d\n", 
-		(double)res.stbi.decode_time/1000000.0,
-		(double)res.stbi.encode_time/1000000.0,
-		(res.stbi.decode_time > 0 ? px / ((double)res.stbi.decode_time/1000.0) : 0),
-		(res.stbi.encode_time > 0 ? px / ((double)res.stbi.encode_time/1000.0) : 0),
-		res.stbi.size/1024
-	);
-	printf(
-		"qoi:     %8.1f    %8.1f      %8.2f      %8.2f  %8d\n", 
-		(double)res.qoi.decode_time/1000000.0,
-		(double)res.qoi.encode_time/1000000.0,
-		(res.qoi.decode_time > 0 ? px / ((double)res.qoi.decode_time/1000.0) : 0),
-		(res.qoi.encode_time > 0 ? px / ((double)res.qoi.encode_time/1000.0) : 0),
-		res.qoi.size/1024
-	);
-	printf("\n");
-}
+void benchmark_directory(const char *path, benchmark_result_t *grand_total) {
+	DIR *dir = opendir(path);
+	if (!dir) {
+		ERROR("Couldn't open directory %s", path);
+	}
 
-int main(int argc, char **argv) {
-	if (argc < 3) {
-		printf("Usage: qoibench <iterations> <directory>\n");
-		printf("Example: qoibench 10 images/textures/\n");
-		exit(1);
+	struct dirent *file;
+
+	if (!opt_norecurse) {
+		for (int i = 0; (file = readdir(dir)) != NULL; i++) {
+			if (
+				file->d_type & DT_DIR &&
+				strcmp(file->d_name, ".") != 0 &&
+				strcmp(file->d_name, "..") != 0
+			) {
+				char subpath[1024];
+				snprintf(subpath, 1024, "%s/%s", path, file->d_name);
+				benchmark_directory(subpath, grand_total);
+			}
+		}
+		rewinddir(dir);
 	}
 
 	float total_percentage = 0;
 	int total_size = 0;
 
-	benchmark_result_t totals = {0};
-
-	int runs = atoi(argv[1]);
-	DIR *dir = opendir(argv[2]);
-	if (runs <=0) {
-		runs = 1;
-	}
-
-	if (!dir) {
-		ERROR("Couldn't open directory %s", argv[2]);
-	}
-
-	printf("## Benchmarking %s/*.png -- %d runs\n\n", argv[2], runs);
-	struct dirent *file;
-	int count = 0;
-	for (int i = 0; dir && (file = readdir(dir)) != NULL; i++) {
+	benchmark_result_t dir_total = {0};
+	
+	int has_shown_heaad = 0;
+	for (int i = 0; (file = readdir(dir)) != NULL; i++) {
 		if (strcmp(file->d_name + strlen(file->d_name) - 4, ".png") != 0) {
 			continue;
 		}
-		count++;
 
-		char *file_path = malloc(strlen(file->d_name) + strlen(argv[2])+8);
-		sprintf(file_path, "%s/%s", argv[2], file->d_name);
+		if (!has_shown_heaad) {
+			has_shown_heaad = 1;
+			printf("## Benchmarking %s/*.png -- %d runs\n\n", path, opt_runs);
+		}
+
+		char *file_path = malloc(strlen(file->d_name) + strlen(path)+8);
+		sprintf(file_path, "%s/%s", path, file->d_name);
 		
-		benchmark_result_t res = benchmark_image(file_path, runs);
-		benchmark_print_result(file_path, res);
+		benchmark_result_t res = benchmark_image(file_path);
+
+		printf("## %s size: %dx%d\n", file_path, res.w, res.h);
+		benchmark_print_result(res);
 
 		free(file_path);
-
 		
-		totals.px += res.px;
-		totals.libpng.encode_time += res.libpng.encode_time;
-		totals.libpng.decode_time += res.libpng.decode_time;
-		totals.libpng.size += res.libpng.size;
-		totals.stbi.encode_time += res.stbi.encode_time;
-		totals.stbi.decode_time += res.stbi.decode_time;
-		totals.stbi.size += res.stbi.size;
-		totals.qoi.encode_time += res.qoi.encode_time;
-		totals.qoi.decode_time += res.qoi.decode_time;
-		totals.qoi.size += res.qoi.size;
+		dir_total.count++;
+		dir_total.raw_size += res.raw_size;
+		dir_total.px += res.px;
+		dir_total.libpng.encode_time += res.libpng.encode_time;
+		dir_total.libpng.decode_time += res.libpng.decode_time;
+		dir_total.libpng.size += res.libpng.size;
+		dir_total.stbi.encode_time += res.stbi.encode_time;
+		dir_total.stbi.decode_time += res.stbi.decode_time;
+		dir_total.stbi.size += res.stbi.size;
+		dir_total.qoi.encode_time += res.qoi.encode_time;
+		dir_total.qoi.decode_time += res.qoi.decode_time;
+		dir_total.qoi.size += res.qoi.size;
+
+		grand_total->count++;
+		grand_total->raw_size += res.raw_size;
+		grand_total->px += res.px;
+		grand_total->libpng.encode_time += res.libpng.encode_time;
+		grand_total->libpng.decode_time += res.libpng.decode_time;
+		grand_total->libpng.size += res.libpng.size;
+		grand_total->stbi.encode_time += res.stbi.encode_time;
+		grand_total->stbi.decode_time += res.stbi.decode_time;
+		grand_total->stbi.size += res.stbi.size;
+		grand_total->qoi.encode_time += res.qoi.encode_time;
+		grand_total->qoi.decode_time += res.qoi.decode_time;
+		grand_total->qoi.size += res.qoi.size;
 	}
 	closedir(dir);
 
-	totals.px /= count;
-	totals.libpng.encode_time /= count;
-	totals.libpng.decode_time /= count;
-	totals.libpng.size /= count;
-	totals.stbi.encode_time /= count;
-	totals.stbi.decode_time /= count;
-	totals.stbi.size /= count;
-	totals.qoi.encode_time /= count;
-	totals.qoi.decode_time /= count;
-	totals.qoi.size /= count;
+	if (dir_total.count > 0) {
+		printf("## Total for %s\n", path);
+		benchmark_print_result(dir_total);
+	}
+}
 
-	benchmark_print_result("Totals (AVG)", totals);
+int main(int argc, char **argv) {
+	if (argc < 3) {
+		printf("Usage: qoibench <iterations> <directory> [options]\n");
+		printf("Options:\n");
+		printf("    --nowarmup ... don't perform a warmup run\n");
+		printf("    --nopng ...... don't run png encode/decode\n");
+		printf("    --noverify ... don't verify qoi roundtrip\n");
+		printf("    --noencode ... don't run encoders\n");
+		printf("    --nodecode ... don't run decoders\n");
+		printf("    --norecurse .. don't descend into directories\n");
+		printf("Examples\n");
+		printf("    qoibench 10 images/textures/\n");
+		printf("    qoibench 1 images/textures/ --nopng --nowarmup\n");
+		exit(1);
+	}
+
+	for (int i = 3; i < argc; i++) {
+		if (strcmp(argv[i], "--nowarmup") == 0) { opt_nowarmup = 1; }
+		else if (strcmp(argv[i], "--nopng") == 0) { opt_nopng = 1; }
+		else if (strcmp(argv[i], "--noverify") == 0) { opt_noverify = 1; }
+		else if (strcmp(argv[i], "--noencode") == 0) { opt_noencode = 1; }
+		else if (strcmp(argv[i], "--nodecode") == 0) { opt_nodecode = 1; }
+		else if (strcmp(argv[i], "--norecurse") == 0) { opt_norecurse = 1; }
+		else { ERROR("Unknown option %s", argv[i]); }
+	}
+
+	opt_runs = atoi(argv[1]);
+	if (opt_runs <=0) {
+		ERROR("Invalid number of runs %d", opt_runs);
+	}
+
+	benchmark_result_t grand_total = {0};
+	benchmark_directory(argv[2], &grand_total);
+
+	if (grand_total.count > 0) {
+		printf("# Grand total for %s\n", argv[2]);
+		benchmark_print_result(grand_total);
+	}
+	else {
+		printf("No images found in %s\n", argv[2]);
+	}
 
 	return 0;
 }

From 66d12eb078b16b3c41fcaf2257e829ca3f499ae1 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Mon, 6 Dec 2021 19:55:39 +0100
Subject: [PATCH 03/14] Add option to only print directory totals

---
 qoibench.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/qoibench.c b/qoibench.c
index b60514ed..ba401a4a 100644
--- a/qoibench.c
+++ b/qoibench.c
@@ -320,6 +320,7 @@ int opt_noverify = 0;
 int opt_nodecode = 0;
 int opt_noencode = 0;
 int opt_norecurse = 0;
+int opt_onlytotals = 0;
 
 
 typedef struct {
@@ -411,6 +412,7 @@ benchmark_result_t benchmark_image(const char *path) {
 	int h;
 
 	// Load the encoded PNG, encoded QOI and raw pixels into memory
+
 	void *pixels = (void *)stbi_load(path, &w, &h, NULL, 4);
 	void *encoded_png = fload(path, &encoded_png_size);
 	void *encoded_qoi = qoi_encode(pixels, &(qoi_desc){
@@ -425,6 +427,7 @@ benchmark_result_t benchmark_image(const char *path) {
 	}
 
 	// Verify QOI Output
+
 	if (!opt_noverify) {
 		qoi_desc dc;
 		void *pixels_qoi = qoi_decode(encoded_qoi, encoded_qoi_size, &dc, 4);
@@ -434,6 +437,8 @@ benchmark_result_t benchmark_image(const char *path) {
 		free(pixels_qoi);
 	}
 
+
+
 	benchmark_result_t res = {0};
 	res.count = 1;
 	res.raw_size = w * h * 4;
@@ -549,8 +554,10 @@ void benchmark_directory(const char *path, benchmark_result_t *grand_total) {
 		
 		benchmark_result_t res = benchmark_image(file_path);
 
-		printf("## %s size: %dx%d\n", file_path, res.w, res.h);
-		benchmark_print_result(res);
+		if (!opt_onlytotals) {
+			printf("## %s size: %dx%d\n", file_path, res.w, res.h);
+			benchmark_print_result(res);
+		}
 
 		free(file_path);
 		
@@ -598,6 +605,7 @@ int main(int argc, char **argv) {
 		printf("    --noencode ... don't run encoders\n");
 		printf("    --nodecode ... don't run decoders\n");
 		printf("    --norecurse .. don't descend into directories\n");
+		printf("    --onlytotals . don't print individual image results\n");
 		printf("Examples\n");
 		printf("    qoibench 10 images/textures/\n");
 		printf("    qoibench 1 images/textures/ --nopng --nowarmup\n");
@@ -611,6 +619,7 @@ int main(int argc, char **argv) {
 		else if (strcmp(argv[i], "--noencode") == 0) { opt_noencode = 1; }
 		else if (strcmp(argv[i], "--nodecode") == 0) { opt_nodecode = 1; }
 		else if (strcmp(argv[i], "--norecurse") == 0) { opt_norecurse = 1; }
+		else if (strcmp(argv[i], "--onlytotals") == 0) { opt_onlytotals = 1; }
 		else { ERROR("Unknown option %s", argv[i]); }
 	}
 

From 28954f7a9a82e4a23ec6926c26617c9657839971 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Mon, 6 Dec 2021 21:23:13 +0100
Subject: [PATCH 04/14] Remove QOI_DIFF_16 and QOI_DIFF_24; better tagging of
 remaining ops; better hash function

---
 qoi.h | 282 +++++++++++++++++++++++++---------------------------------
 1 file changed, 121 insertions(+), 161 deletions(-)

diff --git a/qoi.h b/qoi.h
index a16262f6..8bd81264 100644
--- a/qoi.h
+++ b/qoi.h
@@ -102,17 +102,23 @@ pixel value. Pixels are either encoded as
 
 A running array[64] of previously seen pixel values is maintained by the encoder
 and decoder. Each pixel that is seen by the encoder and decoder is put into this
-array at the position (r^g^b^a) % 64. In the encoder, if the pixel value at this
-index matches the current pixel, this index position is written to the stream.
+array at the position (r * 3 + g * 5 + b * 7) % 64. In the encoder, if the pixel
+value at this index matches the current pixel, this index position is written to
+the stream.
 
-Each chunk starts with a 2, 3 or 4 bit tag, followed by a number of data bits. 
+Each chunk starts with a 8 or 2 bit tag, followed by a number of data bits. 
 The bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned.
 All values encoded in these data bits have the most significant bit (MSB) on the
 left.
 
+The 8-bit tags have precedence over the 2-bit tags. A decoder must check the 
+8-bit tags first.
+
+
 The possible chunks are:
 
- - QOI_INDEX -------------
+
+ - QOI_OP_INDEX ----------
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
 |-------+-----------------|
@@ -122,23 +128,13 @@ The possible chunks are:
 6-bit index into the color index array: 0..63
 
 
- - QOI_RUN ---------------
-|         Byte[0]         |
-|  7  6  5  4  3  2  1  0 |
-|-------+-----------------|
-|  0  1 |       run       |
-
-2-bit tag b01
-6-bit run-length repeating the previous pixel: 1..64
-
-
- - QOI_DIFF_8 ------------
+ - QOI_OP_DIFF -----------
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
 |-------+-----+-----+-----|
-|  1  0 |  dr |  db |  bg |
+|  0  1 |  dr |  dg |  db |
 
-2-bit tag b10
+2-bit tag b01
 2-bit   red channel difference from the previous pixel between -2..1
 2-bit green channel difference from the previous pixel between -2..1
 2-bit  blue channel difference from the previous pixel between -2..1
@@ -147,31 +143,16 @@ The difference to the current channel values are using a wraparound operation,
 so "1 - 2" will result in 255, while "255 + 1" will result in 0.
 
 
- - QOI_DIFF_16 -------------------------------------
+ - QOI_OP_LUMA -------------------------------------
 |         Byte[0]         |         Byte[1]         |
 |  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
-|-------------+-----------|------------ +-----------|
-|  1  1  0  0 |  red diff |  green diff | blue diff |
+|-------+-----------------+-------------+-----------|
+|  1  0 |  green diff     |   dr - dg   |  db - dg  |
 
-4-bit tag b1100
-4-bit   red channel difference from the previous pixel between -8..7
-4-bit green channel difference from the previous pixel between -8..7
-4-bit  blue channel difference from the previous pixel between -8..7
-
-The difference to the current channel values are using a wraparound operation, 
-so "5 - 8" will result in 253, while "250 + 7" will result in 1.
-
-
- - QOI_GDIFF_16 ------------------------------------
-|         Byte[0]         |         Byte[1]         |
-|  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
-|-------------+--------+-------------------+--------|
-|  1  1  0  1 |  dr-dg |    green diff     |  db-dg |
-
-4-bit tag b1101
-3-bit   red channel difference minus green channel difference -4..3
+2-bit tag b10
 6-bit green channel difference from the previous pixel -32..31
-3-bit  blue channel difference minus green channel difference -4..3
+4-bit   red channel difference minus green channel difference -8..7
+4-bit  blue channel difference minus green channel difference -8..7
 
 The green channel is used to indicate the general direction of change and gets
 a few more bits. dr and db base their diffs off of the green channel diff. E.g.
@@ -181,42 +162,46 @@ The difference to the current channel values are using a wraparound operation,
 so "10 - 13" will result in 253, while "250 + 7" will result in 1.
 
 
- - QOI_DIFF_24 ---------------------------------------------------------------
-|         Byte[0]         |         Byte[1]         |         Byte[2]         |
-|  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
-|-------------+----------------+--------------+----------------+--------------|
-|  1  1  1  0 |   red diff     |   green diff |    blue diff   |  alpha diff  |
-
-4-bit tag b1110
-5-bit   red channel difference from the previous pixel between -16..15
-5-bit green channel difference from the previous pixel between -16..15
-5-bit  blue channel difference from the previous pixel between -16..15
-5-bit alpha channel difference from the previous pixel between -16..15
-
-The difference to the current channel values are using a wraparound operation, 
-so "10 - 13" will result in 253, while "250 + 7" will result in 1.
-
-
- - QOI_COLOR -------------
+ - QOI_OP_RUN ------------
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
-|-------------+--+--+--+--|
-|  1  1  1  1 |hr|hg|hb|ha|
+|-------+-----------------|
+|  1  1 |       run       |
 
-4-bit tag b1111
-1-bit   red byte follows
-1-bit green byte follows
-1-bit  blue byte follows
-1-bit alpha byte follows
-
-For each set bit hr, hg, hb and ha another byte follows in this order. If such a
-byte follows, it will replace the current color channel value with the value of
-this byte.
+2-bit tag b01
+6-bit run-length repeating the previous pixel: 1..62
+Note that the run-lengths 63 and 64 (b111110 and b111111) are illegal as they
+are occupied by the QOI_OP_RGB and QOI_OP_RGBA tags.
 
 
-The byte stream is padded at the end with 4 zero bytes. Size the longest chunk
-we can encounter is 5 bytes (QOI_COLOR with RGBA set), with this padding we just
-have to check for an overrun once per decode loop iteration.
+ - QOI_OP_RGB ------------------------------------------
+|         Byte[0]         | Byte[1] | Byte[2] | Byte[3] |
+|  7  6  5  4  3  2  1  0 | 7 .. 0  | 7 .. 0  | 7 .. 0  |
+|-------------------------+---------+---------+---------|
+|  1  1  1  1  1  1  1  0 |   red   |  green  |  blue   |
+
+8-bit tag b11111110
+8-bit   red channel value
+8-bit green channel value
+8-bit  blue channel value
+
+
+- QOI_OP_RGBA ----------------------------------------------------
+|         Byte[0]         | Byte[1] | Byte[2] | Byte[3] | Byte[4] |
+|  7  6  5  4  3  2  1  0 | 7 .. 0  | 7 .. 0  | 7 .. 0  | 7 .. 0  |
+|-------------------------+---------+---------+---------+---------|
+|  1  1  1  1  1  1  1  1 |   red   |  green  |  blue   |  alpha  |
+
+8-bit tag b11111111
+8-bit   red channel value
+8-bit green channel value
+8-bit  blue channel value
+8-bit alpha channel value
+
+
+The byte stream is padded at the end with 4 zero bytes. Size the longest legal 
+chunk is 5 bytes (QOI_OP_RGBA), with this padding it is possible to check for an
+overrun only once per decode loop iteration.
 
 */
 
@@ -318,18 +303,16 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
 	#define QOI_FREE(p)    free(p)
 #endif
 
-#define QOI_INDEX     0x00 // 00xxxxxx
-#define QOI_RUN       0x40 // 01xxxxxx
-#define QOI_DIFF_8    0x80 // 10xxxxxx
-#define QOI_DIFF_16   0xc0 // 1100xxxx
-#define QOI_GDIFF_16  0xd0 // 1101xxxx
-#define QOI_DIFF_24   0xe0 // 1110xxxx
-#define QOI_COLOR     0xf0 // 1111xxxx
+#define QOI_OP_INDEX  0x00 // 00xxxxxx
+#define QOI_OP_DIFF   0x40 // 01xxxxxx
+#define QOI_OP_LUMA   0x80 // 10xxxxxx
+#define QOI_OP_RUN    0xc0 // 11xxxxxx
+#define QOI_OP_RGB    0xfe // 11111110
+#define QOI_OP_RGBA   0xff // 11111111
 
 #define QOI_MASK_2    0xc0 // 11000000
-#define QOI_MASK_4    0xf0 // 11110000
 
-#define QOI_COLOR_HASH(C) (C.rgba.r ^ C.rgba.g ^ C.rgba.b ^ C.rgba.a)
+#define QOI_COLOR_HASH(C) (C.rgba.r * 3 + C.rgba.g * 5 + C.rgba.b * 7)
 #define QOI_MAGIC \
 	(((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \
 	 ((unsigned int)'i') <<  8 | ((unsigned int)'f'))
@@ -400,9 +383,9 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 			px = *(qoi_rgba_t *)(pixels + px_pos);
 		}
 		else {
-			px.rgba.r = pixels[px_pos];
-			px.rgba.g = pixels[px_pos+1];
-			px.rgba.b = pixels[px_pos+2];
+			px.rgba.r = pixels[px_pos + 0];
+			px.rgba.g = pixels[px_pos + 1];
+			px.rgba.b = pixels[px_pos + 2];
 		}
 
 		if (px.v == px_prev.v) {
@@ -411,9 +394,9 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 
 		if ( 
 			run > 0 && 
-			(run == 64 || px.v != px_prev.v || px_pos == px_end)
+			(run == 62 || px.v != px_prev.v || px_pos == px_end)
 		) {
-			bytes[p++] = QOI_RUN | (run - 1);
+			bytes[p++] = QOI_OP_RUN | (run - 1);
 			run = 0;
 		}
 
@@ -421,61 +404,47 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 			int index_pos = QOI_COLOR_HASH(px) % 64;
 
 			if (index[index_pos].v == px.v) {
-				bytes[p++] = QOI_INDEX | index_pos;
+				bytes[p++] = QOI_OP_INDEX | index_pos;
 			}
 			else {
 				index[index_pos] = px;
 
-				char vr = px.rgba.r - px_prev.rgba.r;
-				char vg = px.rgba.g - px_prev.rgba.g;
-				char vb = px.rgba.b - px_prev.rgba.b;
-				char va = px.rgba.a - px_prev.rgba.a;
+				if (px.rgba.a == px_prev.rgba.a) {
+					char vr = px.rgba.r - px_prev.rgba.r;
+					char vg = px.rgba.g - px_prev.rgba.g;
+					char vb = px.rgba.b - px_prev.rgba.b;
 
-				char vg_r = vr - vg;
-				char vg_b = vb - vg;
+					char vg_r = vr - vg;
+					char vg_b = vb - vg;
 
-				if (
-					va == 0 && 
-					vr > -3 && vr < 2 &&
-					vg > -3 && vg < 2 && 
-					vb > -3 && vb < 2
-				) {
-					bytes[p++] = QOI_DIFF_8 | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2);
-				}
-				else if (
-					va == 0 && 
-					vr > -9 && vr < 8 && 
-					vg > -9 && vg < 8 && 
-					vb > -9 && vb < 8
-				) {
-					bytes[p++] = QOI_DIFF_16   | (vr + 8);
-					bytes[p++] = (vg + 8) << 4 | (vb + 8);
-				}
-				else if (
-					va == 0 && 
-					vg_r >  -5 && vg_r <  4 && 
-					vg   > -33 && vg   < 32 && 
-					vg_b >  -5 && vg_b <  4
-				) {
-					bytes[p++] = QOI_GDIFF_16   | (vg_r + 4) << 1 | (vg + 32) >> 5;
-					bytes[p++] = (vg + 32) << 3 | (vg_b + 4);
-				}
-				else if (
-					vr > -17 && vr < 16 && 
-					vg > -17 && vg < 16 && 
-					vb > -17 && vb < 16 && 
-					va > -17 && va < 16
-				) {
-					bytes[p++] = QOI_DIFF_24    | (vr + 16) >> 1;
-					bytes[p++] = (vr + 16) << 7 | (vg + 16) << 2 | (vb + 16) >> 3;
-					bytes[p++] = (vb + 16) << 5 | (va + 16);
+					if (
+						vr > -3 && vr < 2 &&
+						vg > -3 && vg < 2 && 
+						vb > -3 && vb < 2
+					) {
+						bytes[p++] = QOI_OP_DIFF | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2);
+					}
+					else if (
+						vg_r >  -9 && vg_r <  8 && 
+						vg   > -33 && vg   < 32 && 
+						vg_b >  -9 && vg_b <  8
+					) {
+						bytes[p++] = QOI_OP_LUMA     | (vg   + 32);
+						bytes[p++] = (vg_r + 8) << 4 | (vg_b +  8);
+					}
+					else {
+						bytes[p++] = QOI_OP_RGB;
+						bytes[p++] = px.rgba.r;
+						bytes[p++] = px.rgba.g;
+						bytes[p++] = px.rgba.b;
+					}
 				}
 				else {
-					bytes[p++] = QOI_COLOR | (vr ? 8 : 0) | (vg ? 4 : 0) | (vb ? 2 : 0) | (va ? 1 : 0);
-					if (vr) { bytes[p++] = px.rgba.r; }
-					if (vg) { bytes[p++] = px.rgba.g; }
-					if (vb) { bytes[p++] = px.rgba.b; }
-					if (va) { bytes[p++] = px.rgba.a; }
+					bytes[p++] = QOI_OP_RGBA;
+					bytes[p++] = px.rgba.r;
+					bytes[p++] = px.rgba.g;
+					bytes[p++] = px.rgba.b;
+					bytes[p++] = px.rgba.a;
 				}
 			}
 		}
@@ -538,43 +507,34 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 		else if (p < chunks_len) {
 			int b1 = bytes[p++];
 
-			if ((b1 & QOI_MASK_2) == QOI_INDEX) {
-				px = index[b1 ^ QOI_INDEX];
+			if (b1 == QOI_OP_RGB) {
+				px.rgba.r = bytes[p++];
+				px.rgba.g = bytes[p++];
+				px.rgba.b = bytes[p++];
 			}
-			else if ((b1 & QOI_MASK_2) == QOI_RUN) {
-				run = (b1 & 0x3f);
+			else if (b1 == QOI_OP_RGBA) {
+				px.rgba.r = bytes[p++];
+				px.rgba.g = bytes[p++];
+				px.rgba.b = bytes[p++];
+				px.rgba.a = bytes[p++];
 			}
-			else if ((b1 & QOI_MASK_2) == QOI_DIFF_8) {
+			else if ((b1 & QOI_MASK_2) == QOI_OP_INDEX) {
+				px = index[b1 ^ QOI_OP_INDEX];
+			}
+			else if ((b1 & QOI_MASK_2) == QOI_OP_DIFF) {
 				px.rgba.r += ((b1 >> 4) & 0x03) - 2;
 				px.rgba.g += ((b1 >> 2) & 0x03) - 2;
 				px.rgba.b += ( b1       & 0x03) - 2;
 			}
-			else if ((b1 & QOI_MASK_4) == QOI_DIFF_16) {
+			else if ((b1 & QOI_MASK_2) == QOI_OP_LUMA) {
 				int b2 = bytes[p++];
-				px.rgba.r += (b1 & 0x0f) - 8;
-				px.rgba.g += (b2 >> 4)   - 8;
-				px.rgba.b += (b2 & 0x0f) - 8;
-			}
-			else if ((b1 & QOI_MASK_4) == QOI_GDIFF_16) {
-				int b2 = bytes[p++];
-				int vg = ((b1 & 0x01) << 5 | (b2 & 0xf8) >> 3) - 32;
-				px.rgba.r += vg - 4 + ((b1 & 0x0e) >> 1);
+				int vg = (b1 & 0x3f) - 32;
+				px.rgba.r += vg - 8 + ((b2 >> 4) & 0x0f);
 				px.rgba.g += vg;
-				px.rgba.b += vg - 4 +  (b2 & 0x07);
+				px.rgba.b += vg - 8 +  (b2       & 0x0f);
 			}
-			else if ((b1 & QOI_MASK_4) == QOI_DIFF_24) {
-				int b2 = bytes[p++];
-				int b3 = bytes[p++];
-				px.rgba.r += (((b1 & 0x0f) << 1) | (b2 >> 7)) - 16;
-				px.rgba.g +=  ((b2 & 0x7c) >> 2) - 16;
-				px.rgba.b += (((b2 & 0x03) << 3) | ((b3 & 0xe0) >> 5)) - 16;
-				px.rgba.a +=   (b3 & 0x1f) - 16;
-			}
-			else if ((b1 & QOI_MASK_4) == QOI_COLOR) {
-				if (b1 & 8) { px.rgba.r = bytes[p++]; }
-				if (b1 & 4) { px.rgba.g = bytes[p++]; }
-				if (b1 & 2) { px.rgba.b = bytes[p++]; }
-				if (b1 & 1) { px.rgba.a = bytes[p++]; }
+			else if ((b1 & QOI_MASK_2) == QOI_OP_RUN) {
+				run = (b1 & 0x3f);
 			}
 
 			index[QOI_COLOR_HASH(px) % 64] = px;
@@ -584,9 +544,9 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 			*(qoi_rgba_t*)(pixels + px_pos) = px;
 		}
 		else {
-			pixels[px_pos] = px.rgba.r;
-			pixels[px_pos+1] = px.rgba.g;
-			pixels[px_pos+2] = px.rgba.b;
+			pixels[px_pos + 0] = px.rgba.r;
+			pixels[px_pos + 1] = px.rgba.g;
+			pixels[px_pos + 2] = px.rgba.b;
 		}
 	}
 

From eb292694320595abdf975f4be408e320af81159e Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 11:25:24 +0100
Subject: [PATCH 05/14] Fix typo in documentation

---
 qoi.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/qoi.h b/qoi.h
index 8bd81264..2c339f46 100644
--- a/qoi.h
+++ b/qoi.h
@@ -168,7 +168,7 @@ so "10 - 13" will result in 253, while "250 + 7" will result in 1.
 |-------+-----------------|
 |  1  1 |       run       |
 
-2-bit tag b01
+2-bit tag b11
 6-bit run-length repeating the previous pixel: 1..62
 Note that the run-lengths 63 and 64 (b111110 and b111111) are illegal as they
 are occupied by the QOI_OP_RGB and QOI_OP_RGBA tags.

From d6b1ec673ae4df22964646efb7f109ec92c531fc Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 11:30:58 +0100
Subject: [PATCH 06/14] Add alpha channel to QOI_HASH

---
 qoi.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/qoi.h b/qoi.h
index 2c339f46..d0884c9f 100644
--- a/qoi.h
+++ b/qoi.h
@@ -312,7 +312,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
 
 #define QOI_MASK_2    0xc0 // 11000000
 
-#define QOI_COLOR_HASH(C) (C.rgba.r * 3 + C.rgba.g * 5 + C.rgba.b * 7)
+#define QOI_COLOR_HASH(C) (C.rgba.r * 3 + C.rgba.g * 5 + C.rgba.b * 7 + C.rgba.a * 11)
 #define QOI_MAGIC \
 	(((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \
 	 ((unsigned int)'i') <<  8 | ((unsigned int)'f'))

From 947941fbd0a237906d04f7b0433735a52c3e9944 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 14:14:51 +0100
Subject: [PATCH 07/14] Change colorspace header to an enum to avoid confusion

---
 qoi.h | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

diff --git a/qoi.h b/qoi.h
index d0884c9f..e4067ae9 100644
--- a/qoi.h
+++ b/qoi.h
@@ -87,10 +87,7 @@ struct qoi_header_t {
 	uint32_t width;      // image width in pixels (BE)
 	uint32_t height;     // image height in pixels (BE)
 	uint8_t  channels;   // must be 3 (RGB) or 4 (RGBA)
-	uint8_t  colorspace; // a bitmap 0000rgba where
-	                     //   - a zero bit indicates sRGBA, 
-	                     //   - a one bit indicates linear (user interpreted)
-	                     //   colorspace for each channel
+	uint8_t  colorspace; // 0 = sRGB with linear alpha, 1 = all channels linear
 };
 
 The decoder and encoder start with {r: 0, g: 0, b: 0, a: 255} as the previous
@@ -220,15 +217,15 @@ extern "C" {
 // describes either the input format (for qoi_write, qoi_encode), or is filled
 // with the description read from the file header (for qoi_read, qoi_decode).
 
-// The colorspace in this qoi_desc is a bitmap with 0000rgba where a 0-bit 
-// indicates sRGB and a 1-bit indicates linear colorspace for each channel. You 
-// may use one of the predefined constants: QOI_SRGB, QOI_SRGB_LINEAR_ALPHA or 
-// QOI_LINEAR. The colorspace is purely informative. It will be saved to the
-// file header, but does not affect en-/decoding in any way.
+// The colorspace in this qoi_desc is an enum where 
+//   0 = sRGB, i.e. gamma scaled RGB channels and a linear alpha channel
+//   1 = all channels are linear
+// You may use the the constants QOI_SRGB or QOI_LINEAR. The colorspace is 
+// purely informative. It will be saved to the file header, but does not affect
+// en-/decoding in any way.
 
-#define QOI_SRGB 0x00
-#define QOI_SRGB_LINEAR_ALPHA 0x01
-#define QOI_LINEAR 0x0f
+#define QOI_SRGB   0
+#define QOI_LINEAR 1
 
 typedef struct {
 	unsigned int width;

From 6c83cf2e0c0e0ae44fd3d186b804e64390b1269b Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 15:29:46 +0100
Subject: [PATCH 08/14] Increase padding to 8 zero-bytes

---
 qoi.h | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/qoi.h b/qoi.h
index e4067ae9..27b0fa79 100644
--- a/qoi.h
+++ b/qoi.h
@@ -80,7 +80,8 @@ you can define QOI_MALLOC and QOI_FREE before including this library.
 
 -- Data Format
 
-A QOI file has a 14 byte header, followed by any number of data "chunks".
+A QOI file has a 14 byte header, followed by any number of data "chunks" and 8
+zero bytes to mark the end of the data stream.
 
 struct qoi_header_t {
 	char     magic[4];   // magic bytes "qoif"
@@ -196,9 +197,11 @@ are occupied by the QOI_OP_RGB and QOI_OP_RGBA tags.
 8-bit alpha channel value
 
 
-The byte stream is padded at the end with 4 zero bytes. Size the longest legal 
+The byte stream is padded at the end with 8 zero bytes. Since the longest legal 
 chunk is 5 bytes (QOI_OP_RGBA), with this padding it is possible to check for an
-overrun only once per decode loop iteration.
+overrun only once per decode loop iteration. These 0x00 bytes also mark the end
+of the data stream, as an encoder should never produce 8 consecutive zero bytes
+within the stream.
 
 */
 
@@ -314,7 +317,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
 	(((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \
 	 ((unsigned int)'i') <<  8 | ((unsigned int)'f'))
 #define QOI_HEADER_SIZE 14
-#define QOI_PADDING 4
+#define QOI_PADDING 8
 
 typedef union {
 	struct { unsigned char r, g, b, a; } rgba;

From 92f7ebd3f83ac20988a6c97ef20158bb653ed8df Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 15:30:56 +0100
Subject: [PATCH 09/14] Fix qoi_desc colorspace check

---
 qoi.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/qoi.h b/qoi.h
index 27b0fa79..f3f543a9 100644
--- a/qoi.h
+++ b/qoi.h
@@ -344,7 +344,7 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 		data == NULL || out_len == NULL || desc == NULL ||
 		desc->width == 0 || desc->height == 0 ||
 		desc->channels < 3 || desc->channels > 4 ||
-		(desc->colorspace & 0xf0) != 0
+		desc->colorspace > 2
 	) {
 		return NULL;
 	}
@@ -480,6 +480,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 	if (
 		desc->width == 0 || desc->height == 0 || 
 		desc->channels < 3 || desc->channels > 4 ||
+		desc->colorspace > 2 ||
 		header_magic != QOI_MAGIC
 	) {
 		return NULL;

From 6a73cc65c5fdb87d63a5b1f8c655df5c38d45392 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 15:32:43 +0100
Subject: [PATCH 10/14] Wording, whitespace

---
 qoi.h | 36 ++++++++++++++++++------------------
 1 file changed, 18 insertions(+), 18 deletions(-)

diff --git a/qoi.h b/qoi.h
index f3f543a9..a25d6cdf 100644
--- a/qoi.h
+++ b/qoi.h
@@ -95,22 +95,21 @@ The decoder and encoder start with {r: 0, g: 0, b: 0, a: 255} as the previous
 pixel value. Pixels are either encoded as
  - a run of the previous pixel
  - an index into a previously seen pixel
- - a difference to the previous pixel value in r,g,b,a
- - full r,g,b,a values
+ - a difference to the previous pixel value in r,g,b
+ - full r,g,b or r,g,b,a values
 
 A running array[64] of previously seen pixel values is maintained by the encoder
 and decoder. Each pixel that is seen by the encoder and decoder is put into this
-array at the position (r * 3 + g * 5 + b * 7) % 64. In the encoder, if the pixel
-value at this index matches the current pixel, this index position is written to
-the stream.
+array at the position (r * 3 + g * 5 + b * 7 + a * 11) % 64. In the encoder, if
+the pixel value at this index matches the current pixel, this index position is 
+written to the stream as QOI_OP_INDEX.
 
-Each chunk starts with a 8 or 2 bit tag, followed by a number of data bits. 
-The bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned.
-All values encoded in these data bits have the most significant bit (MSB) on the
-left.
+Each chunk starts with a 2 or 8 bit tag, followed by a number of data bits. The 
+bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned. All 
+values encoded in these data bits have the most significant bit on the left.
 
-The 8-bit tags have precedence over the 2-bit tags. A decoder must check the 
-8-bit tags first.
+The 8-bit tags have precedence over the 2-bit tags. A decoder must check for the
+presence of an 8-bit tag first.
 
 
 The possible chunks are:
@@ -216,15 +215,16 @@ within the stream.
 extern "C" {
 #endif
 
-// A pointer to qoi_desc struct has to be supplied to all of qoi's functions. It
-// describes either the input format (for qoi_write, qoi_encode), or is filled
-// with the description read from the file header (for qoi_read, qoi_decode).
+// A pointer to a qoi_desc struct has to be supplied to all of qoi's functions. 
+// It describes either the input format (for qoi_write and qoi_encode), or is 
+// filled with the description read from the file header (for qoi_read and
+// qoi_decode).
 
 // The colorspace in this qoi_desc is an enum where 
 //   0 = sRGB, i.e. gamma scaled RGB channels and a linear alpha channel
 //   1 = all channels are linear
-// You may use the the constants QOI_SRGB or QOI_LINEAR. The colorspace is 
-// purely informative. It will be saved to the file header, but does not affect
+// You may use the constants QOI_SRGB or QOI_LINEAR. The colorspace is purely 
+// informative. It will be saved to the file header, but does not affect
 // en-/decoding in any way.
 
 #define QOI_SRGB   0
@@ -312,7 +312,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels);
 
 #define QOI_MASK_2    0xc0 // 11000000
 
-#define QOI_COLOR_HASH(C) (C.rgba.r * 3 + C.rgba.g * 5 + C.rgba.b * 7 + C.rgba.a * 11)
+#define QOI_COLOR_HASH(C) (C.rgba.r*3 + C.rgba.g*5 + C.rgba.b*7 + C.rgba.a*11)
 #define QOI_MAGIC \
 	(((unsigned int)'q') << 24 | ((unsigned int)'o') << 16 | \
 	 ((unsigned int)'i') <<  8 | ((unsigned int)'f'))
@@ -422,7 +422,7 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 						vg > -3 && vg < 2 && 
 						vb > -3 && vb < 2
 					) {
-						bytes[p++] = QOI_OP_DIFF | ((vr + 2) << 4) | (vg + 2) << 2 | (vb + 2);
+						bytes[p++] = QOI_OP_DIFF | (vr + 2) << 4 | (vg + 2) << 2 | (vb + 2);
 					}
 					else if (
 						vg_r >  -9 && vg_r <  8 && 

From 21031685199533e136bcbd6ace54c1b665884fb7 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Wed, 8 Dec 2021 15:45:18 +0100
Subject: [PATCH 11/14] Minor encoding throughput improvement

---
 qoi.h | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/qoi.h b/qoi.h
index a25d6cdf..07b4561d 100644
--- a/qoi.h
+++ b/qoi.h
@@ -390,17 +390,17 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 
 		if (px.v == px_prev.v) {
 			run++;
+			if (run == 62 || px_pos == px_end) {
+				bytes[p++] = QOI_OP_RUN | (run - 1);
+				run = 0;
+			}
 		}
+		else  {
+			if (run > 0) {
+				bytes[p++] = QOI_OP_RUN | (run - 1);
+				run = 0;
+			}
 
-		if ( 
-			run > 0 && 
-			(run == 62 || px.v != px_prev.v || px_pos == px_end)
-		) {
-			bytes[p++] = QOI_OP_RUN | (run - 1);
-			run = 0;
-		}
-
-		if (px.v != px_prev.v) {
 			int index_pos = QOI_COLOR_HASH(px) % 64;
 
 			if (index[index_pos].v == px.v) {
@@ -520,7 +520,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 				px.rgba.a = bytes[p++];
 			}
 			else if ((b1 & QOI_MASK_2) == QOI_OP_INDEX) {
-				px = index[b1 ^ QOI_OP_INDEX];
+				px = index[b1];
 			}
 			else if ((b1 & QOI_MASK_2) == QOI_OP_DIFF) {
 				px.rgba.r += ((b1 >> 4) & 0x03) - 2;

From 075ab8fe42f3ca4a295b787c862308de01013e78 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Fri, 10 Dec 2021 20:09:52 +0100
Subject: [PATCH 12/14] Zero-initialize previous pixel color

---
 qoi.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/qoi.h b/qoi.h
index 07b4561d..5a5ee919 100644
--- a/qoi.h
+++ b/qoi.h
@@ -91,7 +91,7 @@ struct qoi_header_t {
 	uint8_t  colorspace; // 0 = sRGB with linear alpha, 1 = all channels linear
 };
 
-The decoder and encoder start with {r: 0, g: 0, b: 0, a: 255} as the previous
+The decoder and encoder start with {r: 0, g: 0, b: 0, a: 0} as the previous
 pixel value. Pixels are either encoded as
  - a run of the previous pixel
  - an index into a previously seen pixel
@@ -371,7 +371,7 @@ void *qoi_encode(const void *data, const qoi_desc *desc, int *out_len) {
 	qoi_rgba_t index[64] = {0};
 
 	int run = 0;
-	qoi_rgba_t px_prev = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 255}};
+	qoi_rgba_t px_prev = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 0}};
 	qoi_rgba_t px = px_prev;
 	
 	int px_len = desc->width * desc->height * desc->channels;
@@ -496,7 +496,7 @@ void *qoi_decode(const void *data, int size, qoi_desc *desc, int channels) {
 		return NULL;
 	}
 
-	qoi_rgba_t px = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 255}};
+	qoi_rgba_t px = {.rgba = {.r = 0, .g = 0, .b = 0, .a = 0}};
 	qoi_rgba_t index[64] = {0};
 
 	int run = 0;

From 0ad304d7615a57d80eb94f9eeef5097f8f8496bf Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Fri, 10 Dec 2021 20:27:35 +0100
Subject: [PATCH 13/14] Be more specific with the documentation of the file
 format

---
 qoi.h | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/qoi.h b/qoi.h
index 5a5ee919..8eed5803 100644
--- a/qoi.h
+++ b/qoi.h
@@ -81,36 +81,44 @@ you can define QOI_MALLOC and QOI_FREE before including this library.
 -- Data Format
 
 A QOI file has a 14 byte header, followed by any number of data "chunks" and 8
-zero bytes to mark the end of the data stream.
+zero-bytes to mark the end of the data stream.
 
 struct qoi_header_t {
 	char     magic[4];   // magic bytes "qoif"
 	uint32_t width;      // image width in pixels (BE)
 	uint32_t height;     // image height in pixels (BE)
-	uint8_t  channels;   // must be 3 (RGB) or 4 (RGBA)
+	uint8_t  channels;   // 3 = RGB, 4 = RGBA
 	uint8_t  colorspace; // 0 = sRGB with linear alpha, 1 = all channels linear
 };
 
 The decoder and encoder start with {r: 0, g: 0, b: 0, a: 0} as the previous
 pixel value. Pixels are either encoded as
  - a run of the previous pixel
- - an index into a previously seen pixel
+ - an index into an array of previously seen pixels
  - a difference to the previous pixel value in r,g,b
  - full r,g,b or r,g,b,a values
 
-A running array[64] of previously seen pixel values is maintained by the encoder
-and decoder. Each pixel that is seen by the encoder and decoder is put into this
-array at the position (r * 3 + g * 5 + b * 7 + a * 11) % 64. In the encoder, if
-the pixel value at this index matches the current pixel, this index position is 
-written to the stream as QOI_OP_INDEX.
+The color channels are assumed to not be premultiplied with the alpha channel 
+(“un-premultiplied alpha”).
 
-Each chunk starts with a 2 or 8 bit tag, followed by a number of data bits. The 
+A running array[64] (zero-initialized) of previously seen pixel values is 
+maintained by the encoder and decoder. Each pixel that is seen by the encoder
+and decoder is put into this array at the position formed by a hash function of
+the color value. In the encoder, if the pixel value at the index matches the
+current pixel, this index position is written to the stream as QOI_OP_INDEX. 
+The hash function for the index is:
+
+  index_position = (r * 3 + g * 5 + b * 7 + a * 11) % 64
+
+Each chunk starts with a 2- or 8-bit tag, followed by a number of data bits. The 
 bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned. All 
 values encoded in these data bits have the most significant bit on the left.
 
 The 8-bit tags have precedence over the 2-bit tags. A decoder must check for the
 presence of an 8-bit tag first.
 
+The byte stream is padded with 8 zero-bytes at the end.
+
 
 The possible chunks are:
 
@@ -139,8 +147,10 @@ The possible chunks are:
 The difference to the current channel values are using a wraparound operation, 
 so "1 - 2" will result in 255, while "255 + 1" will result in 0.
 
+Values are stored as unsigned integers with a bias of 2. E.g. -2 is stored as 
+0 (b00). 1 is stored as 3 (b11).
+
 
- - QOI_OP_LUMA -------------------------------------
 |         Byte[0]         |         Byte[1]         |
 |  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
 |-------+-----------------+-------------+-----------|
@@ -151,13 +161,18 @@ so "1 - 2" will result in 255, while "255 + 1" will result in 0.
 4-bit   red channel difference minus green channel difference -8..7
 4-bit  blue channel difference minus green channel difference -8..7
 
-The green channel is used to indicate the general direction of change and gets
-a few more bits. dr and db base their diffs off of the green channel diff. E.g.
-  dr = (last_px.r - cur_px.r) - (last_px.g - cur_px.g)
+The green channel is used to indicate the general direction of change and is 
+encoded in 6 bits. The red and green channels (dr and db) base their diffs off
+of the green channel difference and are encoded in 4 bits. I.e.:
+  dr_dg = (last_px.r - cur_px.r) - (last_px.g - cur_px.g)
+  db_dg = (last_px.b - cur_px.b) - (last_px.g - cur_px.g)
 
 The difference to the current channel values are using a wraparound operation, 
 so "10 - 13" will result in 253, while "250 + 7" will result in 1.
 
+Values are stored as unsigned integers with a bias of 32 for the green channel 
+and a bias of 8 for the red and blue channel.
+
 
  - QOI_OP_RUN ------------
 |         Byte[0]         |
@@ -167,8 +182,10 @@ so "10 - 13" will result in 253, while "250 + 7" will result in 1.
 
 2-bit tag b11
 6-bit run-length repeating the previous pixel: 1..62
-Note that the run-lengths 63 and 64 (b111110 and b111111) are illegal as they
-are occupied by the QOI_OP_RGB and QOI_OP_RGBA tags.
+
+The run-length is stored with a bias of 1. Note that the run-lengths 63 and 64 
+(b111110 and b111111) are illegal as they are occupied by the QOI_OP_RGB and 
+QOI_OP_RGBA tags.
 
 
  - QOI_OP_RGB ------------------------------------------

From c2edcd3d7a164d81ed149073602e0ac3842442a4 Mon Sep 17 00:00:00 2001
From: Dominic Szablewski <dominic@phoboslab.org>
Date: Sat, 11 Dec 2021 12:54:01 +0100
Subject: [PATCH 14/14] Cosmetics

---
 qoi.h | 23 ++++++++++++-----------
 1 file changed, 12 insertions(+), 11 deletions(-)

diff --git a/qoi.h b/qoi.h
index 8eed5803..bee6cfb2 100644
--- a/qoi.h
+++ b/qoi.h
@@ -123,22 +123,22 @@ The byte stream is padded with 8 zero-bytes at the end.
 The possible chunks are:
 
 
- - QOI_OP_INDEX ----------
+.- QOI_OP_INDEX ----------.
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
 |-------+-----------------|
 |  0  0 |     index       |
-
+`-------------------------`
 2-bit tag b00
 6-bit index into the color index array: 0..63
 
 
- - QOI_OP_DIFF -----------
+.- QOI_OP_DIFF -----------. 
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
 |-------+-----+-----+-----|
 |  0  1 |  dr |  dg |  db |
-
+`-------------------------`
 2-bit tag b01
 2-bit   red channel difference from the previous pixel between -2..1
 2-bit green channel difference from the previous pixel between -2..1
@@ -151,11 +151,12 @@ Values are stored as unsigned integers with a bias of 2. E.g. -2 is stored as
 0 (b00). 1 is stored as 3 (b11).
 
 
+.- QOI_OP_LUMA -------------------------------------. 
 |         Byte[0]         |         Byte[1]         |
 |  7  6  5  4  3  2  1  0 |  7  6  5  4  3  2  1  0 |
 |-------+-----------------+-------------+-----------|
 |  1  0 |  green diff     |   dr - dg   |  db - dg  |
-
+`---------------------------------------------------`
 2-bit tag b10
 6-bit green channel difference from the previous pixel -32..31
 4-bit   red channel difference minus green channel difference -8..7
@@ -174,12 +175,12 @@ Values are stored as unsigned integers with a bias of 32 for the green channel
 and a bias of 8 for the red and blue channel.
 
 
- - QOI_OP_RUN ------------
+.- QOI_OP_RUN ------------.
 |         Byte[0]         |
 |  7  6  5  4  3  2  1  0 |
 |-------+-----------------|
 |  1  1 |       run       |
-
+`-------------------------`
 2-bit tag b11
 6-bit run-length repeating the previous pixel: 1..62
 
@@ -188,24 +189,24 @@ The run-length is stored with a bias of 1. Note that the run-lengths 63 and 64
 QOI_OP_RGBA tags.
 
 
- - QOI_OP_RGB ------------------------------------------
+.- QOI_OP_RGB ------------------------------------------.
 |         Byte[0]         | Byte[1] | Byte[2] | Byte[3] |
 |  7  6  5  4  3  2  1  0 | 7 .. 0  | 7 .. 0  | 7 .. 0  |
 |-------------------------+---------+---------+---------|
 |  1  1  1  1  1  1  1  0 |   red   |  green  |  blue   |
-
+`-------------------------------------------------------`
 8-bit tag b11111110
 8-bit   red channel value
 8-bit green channel value
 8-bit  blue channel value
 
 
-- QOI_OP_RGBA ----------------------------------------------------
+.- QOI_OP_RGBA ---------------------------------------------------.
 |         Byte[0]         | Byte[1] | Byte[2] | Byte[3] | Byte[4] |
 |  7  6  5  4  3  2  1  0 | 7 .. 0  | 7 .. 0  | 7 .. 0  | 7 .. 0  |
 |-------------------------+---------+---------+---------+---------|
 |  1  1  1  1  1  1  1  1 |   red   |  green  |  blue   |  alpha  |
-
+`-----------------------------------------------------------------`
 8-bit tag b11111111
 8-bit   red channel value
 8-bit green channel value