webm/theora/yuv2rgb: Fix YUV conversion issues and add BGRP WEBM handling. (Now clang-format friendly.)

This should fix the various issues with colours in Ogg Theora and WEBM playback.
(A reference project is attached to PR #26051, which this commit should be part of.)

This version of the commit, rather than moving x->RGBA handling into libsimplewebm, uses a colourspace field added to libsimplewebm by a PR there.
Thus, the commit that precedes this should be the synchronization & cleanup commit for that.

Also, this version is now clang-format friendly. I hope.

modules/theora/video_stream_theora.cpp

@@ -94,15 +94,15 @@ void VideoStreamPlaybackTheora::video_write(void) {
 
 		if (px_fmt == TH_PF_444) {
 
-			yuv444_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[1].data, (uint8_t *)yuv[2].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2, 0);
+			yuv444_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[1].data, (uint8_t *)yuv[2].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2);
 
 		} else if (px_fmt == TH_PF_422) {
 
-			yuv422_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[1].data, (uint8_t *)yuv[2].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2, 0);
+			yuv422_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[1].data, (uint8_t *)yuv[2].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2);
 
 		} else if (px_fmt == TH_PF_420) {
 
-			yuv420_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[2].data, (uint8_t *)yuv[1].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2, 0);
+			yuv420_2_rgb8888((uint8_t *)dst, (uint8_t *)yuv[0].data, (uint8_t *)yuv[1].data, (uint8_t *)yuv[2].data, size.x, size.y, yuv[0].stride, yuv[1].stride, size.x << 2);
 		};
 
 		format = Image::FORMAT_RGBA8;

modules/webm/video_stream_webm.cpp

@@ -32,6 +32,7 @@
 
 #include "OpusVorbisDecoder.hpp"
 #include "VPXDecoder.hpp"
+#include <vpx/vpx_image.h>
 
 #include "mkvparser/mkvparser.h"
 
@@ -314,19 +315,37 @@ void VideoStreamPlaybackWebm::update(float p_delta) {
 						PoolVector<uint8_t>::Write w = frame_data.write();
 						bool converted = false;
 
-						if (image.chromaShiftW == 1 && image.chromaShiftH == 1) {
+						if (image.chromaShiftW == 0 && image.chromaShiftH == 0 && image.cs == VPX_CS_SRGB) {
+
+							uint8_t *wp = w.ptr();
+							unsigned char *rRow = image.planes[2];
+							unsigned char *gRow = image.planes[0];
+							unsigned char *bRow = image.planes[1];
+							for (size_t i = 0; i < image.h; i++) {
+								for (size_t j = 0; j < image.w; j++) {
+									*wp++ = rRow[j];
+									*wp++ = gRow[j];
+									*wp++ = bRow[j];
+									*wp++ = 255;
+								}
+								rRow += image.linesize[2];
+								gRow += image.linesize[0];
+								bRow += image.linesize[1];
+							}
+							converted = true;
+						} else if (image.chromaShiftW == 1 && image.chromaShiftH == 1) {
 
-							yuv420_2_rgb8888(w.ptr(), image.planes[0], image.planes[2], image.planes[1], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2, 0);
+							yuv420_2_rgb8888(w.ptr(), image.planes[0], image.planes[1], image.planes[2], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2);
 							// 								libyuv::I420ToARGB(image.planes[0], image.linesize[0], image.planes[2], image.linesize[2], image.planes[1], image.linesize[1], w.ptr(), image.w << 2, image.w, image.h);
 							converted = true;
 						} else if (image.chromaShiftW == 1 && image.chromaShiftH == 0) {
 
-							yuv422_2_rgb8888(w.ptr(), image.planes[0], image.planes[2], image.planes[1], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2, 0);
+							yuv422_2_rgb8888(w.ptr(), image.planes[0], image.planes[1], image.planes[2], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2);
 							// 								libyuv::I422ToARGB(image.planes[0], image.linesize[0], image.planes[2], image.linesize[2], image.planes[1], image.linesize[1], w.ptr(), image.w << 2, image.w, image.h);
 							converted = true;
 						} else if (image.chromaShiftW == 0 && image.chromaShiftH == 0) {
 
-							yuv444_2_rgb8888(w.ptr(), image.planes[0], image.planes[2], image.planes[1], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2, 0);
+							yuv444_2_rgb8888(w.ptr(), image.planes[0], image.planes[1], image.planes[2], image.w, image.h, image.linesize[0], image.linesize[1], image.w << 2);
 							// 								libyuv::I444ToARGB(image.planes[0], image.linesize[0], image.planes[2], image.linesize[2], image.planes[1], image.linesize[1], w.ptr(), image.w << 2, image.w, image.h);
 							converted = true;
 						} else if (image.chromaShiftW == 2 && image.chromaShiftH == 0) {

thirdparty/misc/yuv2rgb.h

@@ -24,6 +24,14 @@ does not infringe any patents that apply in your area before you
 ship it.
 */
 
+/*
+ * Please note that this version has been modified for various reasons:
+ * 1. Using the Godot core typedefs
+ * 2. At some point or another the code relied on the byte order of a uint32_t, this has been fixed
+ * 3. Output has been reordered to struct { uint8_t r, g, b, a; } precisely in accordance with the function names
+ * 4. Removing unused 'dither' parameter
+ */
+
 #ifndef YUV2RGB_H
 #define YUV2RGB_H
 
@@ -803,6 +811,8 @@ static const uint32_t tables[256*3] = {
 	0xE6365800U
 };
 
+/* -- Common -- */
+
 #define FLAGS 0x40080100
 #define READUV(U,V) (tables[256 + (U)] + tables[512 + (V)])
 #define READY(Y)    tables[Y]
@@ -820,12 +830,14 @@ do {                             \
 
 #define STORE(Y,DSTPTR)         \
 do {                            \
-    *(DSTPTR)++ = (Y);          \
-    *(DSTPTR)++ = (Y)>>22;      \
     *(DSTPTR)++ = (Y)>>11;      \
-    *(DSTPTR)++ = 255;            \
+    *(DSTPTR)++ = (Y)>>22;       \
+    *(DSTPTR)++ = (Y);          \
+	*(DSTPTR)++ = 255;           \
 } while (0 == 1)
 
+/* -- End Common -- */
+
 static void yuv422_2_rgb8888(uint8_t  *dst_ptr,
 		const uint8_t  *y_ptr,
 		const uint8_t  *u_ptr,
@@ -834,8 +846,7 @@ static void yuv422_2_rgb8888(uint8_t  *dst_ptr,
 		      int32_t   height,
 		      int32_t   y_span,
 		      int32_t   uv_span,
-		      int32_t   dst_span,
-		      int32_t   dither)
+		      int32_t   dst_span)
 {
     height -= 1;
     while (height > 0)
@@ -909,35 +920,7 @@ static void yuv422_2_rgb8888(uint8_t  *dst_ptr,
     }
 }
 
-
-#undef FLAGS
-#undef READUV
-#undef READY
-#undef FIXUP
-#undef STORE
-
-
-#define FLAGS 0x40080100
-#define READUV(U,V) (tables[256 + (U)] + tables[512 + (V)])
-#define READY(Y)    tables[Y]
-#define FIXUP(Y)                 \
-do {                             \
-    int tmp = (Y) & FLAGS;       \
-    if (tmp != 0)                \
-    {                            \
-	tmp  -= tmp>>8;          \
-	(Y)  |= tmp;             \
-	tmp   = FLAGS & ~(Y>>1); \
-	(Y)  += tmp>>8;          \
-    }                            \
-} while (0 == 1)
-
-#define STORE(Y,DSTPTR)     \
-do {                        \
-    (DSTPTR) = 0xFF000000 | (Y & 0xFF) | (0xFF00 & (Y>>14)) | (0xFF0000 & (Y<<5));\
-} while (0 == 1)
-
-static void yuv420_2_rgb8888(uint8_t  *dst_ptr_,
+static void yuv420_2_rgb8888(uint8_t  *dst_ptr,
 		const uint8_t  *y_ptr,
 		const uint8_t  *u_ptr,
 		const uint8_t  *v_ptr,
@@ -945,12 +928,9 @@ static void yuv420_2_rgb8888(uint8_t  *dst_ptr_,
 		      int32_t   height,
 		      int32_t   y_span,
 		      int32_t   uv_span,
-		      int32_t   dst_span,
-		      int32_t   dither)
+		      int32_t   dst_span)
 {
-    uint32_t *dst_ptr = (uint32_t *)(void *)dst_ptr_;
-    dst_span >>= 2;
-
+    /* The 'dst_ptr as uint32_t' thing is not endianness-aware, so that's been removed. */
     height -= 1;
     while (height > 0)
     {
@@ -960,36 +940,38 @@ static void yuv420_2_rgb8888(uint8_t  *dst_ptr_,
 	{
 	    /* Do 2 column pairs */
 	    uint32_t uv, y0, y1;
+        uint8_t * dst_ptr_1span = dst_ptr + dst_span;
 
 	    uv  = READUV(*u_ptr++,*v_ptr++);
 	    y1  = uv + READY(y_ptr[y_span]);
 	    y0  = uv + READY(*y_ptr++);
 	    FIXUP(y1);
 	    FIXUP(y0);
-	    STORE(y1, dst_ptr[dst_span]);
-	    STORE(y0, *dst_ptr++);
+	    STORE(y1, dst_ptr_1span);
+	    STORE(y0, dst_ptr);
 	    y1  = uv + READY(y_ptr[y_span]);
 	    y0  = uv + READY(*y_ptr++);
 	    FIXUP(y1);
 	    FIXUP(y0);
-	    STORE(y1, dst_ptr[dst_span]);
-	    STORE(y0, *dst_ptr++);
+	    STORE(y1, dst_ptr_1span);
+	    STORE(y0, dst_ptr);
 	    height += (2<<16);
 	}
 	if ((height>>16) == 0)
 	{
 	    /* Trailing column pair */
 	    uint32_t uv, y0, y1;
+        uint8_t * dst_ptr_1span = dst_ptr + dst_span;
 
 	    uv = READUV(*u_ptr,*v_ptr);
 	    y1 = uv + READY(y_ptr[y_span]);
 	    y0 = uv + READY(*y_ptr++);
 	    FIXUP(y1);
 	    FIXUP(y0);
-	    STORE(y0, dst_ptr[dst_span]);
-	    STORE(y1, *dst_ptr++);
+	    STORE(y0, dst_ptr_1span);
+	    STORE(y1, dst_ptr);
 	}
-	dst_ptr += dst_span*2-width;
+	dst_ptr += (dst_span * 2) - (width * 4);
 	y_ptr   += y_span*2-width;
 	u_ptr   += uv_span-(width>>1);
 	v_ptr   += uv_span-(width>>1);
@@ -1011,8 +993,8 @@ static void yuv420_2_rgb8888(uint8_t  *dst_ptr_,
 	    y0  = uv + READY(*y_ptr++);
 	    FIXUP(y1);
 	    FIXUP(y0);
-	    STORE(y1, *dst_ptr++);
-	    STORE(y0, *dst_ptr++);
+	    STORE(y1, dst_ptr);
+	    STORE(y0, dst_ptr);
 	    height += (2<<16);
 	}
 	if ((height>>16) == 0)
@@ -1023,42 +1005,11 @@ static void yuv420_2_rgb8888(uint8_t  *dst_ptr_,
 	    uv = READUV(*u_ptr++,*v_ptr++);
 	    y0 = uv + READY(*y_ptr++);
 	    FIXUP(y0);
-	    STORE(y0, *dst_ptr++);
+	    STORE(y0, dst_ptr);
 	}
     }
 }
 
-
-
-#undef FLAGS
-#undef READUV
-#undef READY
-#undef FIXUP
-#undef STORE
-
-#define FLAGS 0x40080100
-#define READUV(U,V) (tables[256 + (U)] + tables[512 + (V)])
-#define READY(Y)    tables[Y]
-#define FIXUP(Y)                 \
-do {                             \
-    int tmp = (Y) & FLAGS;       \
-    if (tmp != 0)                \
-    {                            \
-	tmp  -= tmp>>8;          \
-	(Y)  |= tmp;             \
-	tmp   = FLAGS & ~(Y>>1); \
-	(Y)  += tmp>>8;          \
-    }                            \
-} while (0 == 1)
-
-#define STORE(Y,DSTPTR)         \
-do {                            \
-    *(DSTPTR)++ = (Y);          \
-    *(DSTPTR)++ = (Y)>>22;      \
-    *(DSTPTR)++ = (Y)>>11;      \
-	*(DSTPTR)++ = 255;           \
-} while (0 == 1)
-
 static void yuv444_2_rgb8888(uint8_t  *dst_ptr,
 		const uint8_t  *y_ptr,
 		const uint8_t  *u_ptr,
@@ -1067,8 +1018,7 @@ static void yuv444_2_rgb8888(uint8_t  *dst_ptr,
 		      int32_t   height,
 		      int32_t   y_span,
 		      int32_t   uv_span,
-		      int32_t   dst_span,
-		      int32_t   dither)
+		      int32_t   dst_span)
 {
     height -= 1;
     while (height > 0)
@@ -1143,4 +1093,11 @@ static void yuv444_2_rgb8888(uint8_t  *dst_ptr,
 	height -= 1;
     }
 }
+
+#undef FLAGS
+#undef READUV
+#undef READY
+#undef FIXUP
+#undef STORE
+
 #endif // YUV2RGB_H