X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=effect_chain.h;h=7e0cd9f97b9b6465ea46b38908f7213aa1f2664c;hp=071f9ebe51b33b511facbba7d7c2a68d8c542863;hb=e8499e3e9892a74c7882af4be14ccdc1e3d92c2b;hpb=2d3b64d51be3e8f38a8c224f558de4ceb222e799

diff --git a/effect_chain.h b/effect_chain.h
index 071f9eb..7e0cd9f 100644
--- a/effect_chain.h
+++ b/effect_chain.h
@@ -274,6 +274,12 @@ public:
 	                      const YCbCrFormat &ycbcr_format,
 			      YCbCrOutputSplitting output_splitting = YCBCR_OUTPUT_INTERLEAVED);
 
+	// Change Y'CbCr output format. (This can be done also after finalize()).
+	// Note that you are not allowed to change subsampling parameters;
+	// however, you can change the color space parameters, ie.,
+	// luma_coefficients, full_range and num_levels.
+	void change_ycbcr_output_format(const YCbCrFormat &ycbcr_format);
+
 	// Set number of output bits, to scale the dither.
 	// 8 is the right value for most outputs.
 	// The default, 0, is a special value that means no dither.
@@ -299,16 +305,16 @@ public:
 	// no matter what you specify.
 	//
 	// Of special interest is GL_SRGB8_ALPHA8, which stores sRGB-encoded RGB
-	// and linear alpha; this is half the memory bandwidth og GL_RGBA16F,
+	// and linear alpha; this is half the memory bandwidth of GL_RGBA16F,
 	// while retaining reasonable precision for typical image data. It will,
 	// however, cause some gamut clipping if your colorspace is far from sRGB,
 	// as it cannot represent values outside [0,1]. NOTE: If you construct
-	// a chain where you end up bouncing pixels in non-linear light this
-	// will not do the wrong thing. However, it's hard to see how this
-	// could happen in a non-contrived chain; few effects ever need texture
-	// bounce or resizing without also combining multiple pixels, which
-	// really needs linear light and thus triggers a conversion before the
-	// bounce.
+	// a chain where you end up bouncing pixels in non-linear light
+	// (gamma different from GAMMA_LINEAR), this will be the wrong thing.
+	// However, it's hard to see how this could happen in a non-contrived
+	// chain; few effects ever need texture bounce or resizing without also
+	// combining multiple pixels, which really needs linear light and thus
+	// triggers a conversion before the bounce.
 	//
 	// If you don't need alpha (or can do with very little of it), GL_RGB10_A2
 	// is even better, as it has two more bits for each color component. There
@@ -318,7 +324,10 @@ public:
 	// the sRGB curve, and reduces maximum error (in sRGB distance) by almost an
 	// order of magnitude, well below what you can get from 8-bit true sRGB.
 	// (Note that this strategy avoids the problem with bounced non-linear data
-	// above, since the square root is turned off in that case.)
+	// above, since the square root is turned off in that case.) However, texture
+	// filtering will happen on the transformed values, so if you have heavy
+	// downscaling or the likes (e.g. mipmaps), you could get subtly bad results.
+	// You'll need to see which of the two that works the best for you in practice.
 	void set_intermediate_format(
 		GLenum intermediate_format,
 		FramebufferTransformation transformation = NO_FRAMEBUFFER_TRANSFORMATION)
@@ -338,7 +347,6 @@ public:
 	void reset_phase_timing();
 	void print_phase_timing();
 
-	//void render(unsigned char *src, unsigned char *dst);
 	void render_to_screen()
 	{
 		render_to_fbo(0, 0, 0);