[PATCH renderproto] Add floating point transforms

[walter harms]

Am 15.08.2014 17:55, schrieb Keith Packard:
> Fixed point coordinates don't provide reasonable precision for
> transformation operations; the resulting transforms are often off by
> several pixels.
> 
> Allow clients to represent the transformation using either fixed point
> or floating point.
> 
> Signed-off-by: Keith Packard <keithp at keithp.com>
> ---
>  configure.ac    |  2 +-
>  render.h        |  5 ++++-
>  renderproto.h   | 26 ++++++++++++++++++++++++++
>  renderproto.txt | 42 ++++++++++++++++++++++++++++++++++++++++--
>  4 files changed, 71 insertions(+), 4 deletions(-)
> 
> diff --git a/configure.ac b/configure.ac
> index 8154a3a..43d054f 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -1,5 +1,5 @@
>  AC_PREREQ([2.60])
> -AC_INIT([RenderProto], [0.11.1], [https://bugs.freedesktop.org/enter_bug.cgi?product=xorg])
> +AC_INIT([RenderProto], [0.12.0], [https://bugs.freedesktop.org/enter_bug.cgi?product=xorg])
>  AM_INIT_AUTOMAKE([foreign dist-bzip2])
>  
>  # Require xorg-macros: XORG_DEFAULT_OPTIONS
> diff --git a/render.h b/render.h
> index 7ecde3f..0f04cc6 100644
> --- a/render.h
> +++ b/render.h
> @@ -77,7 +77,10 @@ typedef XID		PictFormat;
>  #define X_RenderCreateLinearGradient        34
>  #define X_RenderCreateRadialGradient        35
>  #define X_RenderCreateConicalGradient       36
> -#define RenderNumberRequests		    (X_RenderCreateConicalGradient+1)
> +/* 0.12 */
> +#define X_RenderSetPictureTransformFloat    37
> +
> +#define RenderNumberRequests		    (X_RenderSetPictureTransformFloat+1)
>  
>  #define BadPictFormat			    0
>  #define BadPicture			    1
> diff --git a/renderproto.h b/renderproto.h
> index b7b72c1..4fe8da6 100644
> --- a/renderproto.h
> +++ b/renderproto.h
> @@ -640,6 +640,32 @@ typedef struct {
>  
>  #define sz_xRenderCreateConicalGradientReq                 24
>  
> +/* 0.12 and higher */
> +
> +typedef struct {
> +    float	matrix11;
> +    float	matrix12;
> +    float	matrix13;
> +    float	matrix21;
> +    float	matrix22;
> +    float	matrix23;
> +    float	matrix31;
> +    float	matrix32;
> +    float	matrix33;
> +} xRenderTransformFloat;


Would it make sense to use double for internal calculation ?
Most systems provide 64bit (or more) fpu-registers these days.

re,
 wh

> +#define sz_xRenderTransformFloat 36
> +
> +typedef struct {
> +    CARD8		        reqType;
> +    CARD8		        renderReqType;
> +    CARD16		        length B16;
> +    Picture		        picture B32;
> +    xRenderTransformFloat	transform;
> +} xRenderSetPictureTransformFloatReq;
> +
> +#define sz_xRenderSetPictureTransformFloatReq	    44
> +
>  #undef Window
>  #undef Drawable
>  #undef Font
> diff --git a/renderproto.txt b/renderproto.txt
> index 9f1f231..fe5a233 100644
> --- a/renderproto.txt
> +++ b/renderproto.txt
> @@ -1,6 +1,6 @@
>  			The X Rendering Extension
> -			      Version 0.11
> -			       2009-07-15
> +			      Version 0.12
> +			       2014-08-13
>  			      Keith Packard
>  			    keithp at keithp.com
>  
> @@ -12,6 +12,10 @@ Rendering geometric figures is accomplished by client-side tessellation into
>  either triangles or trapezoids.  Text is drawn by loading glyphs into the
>  server and rendering sets of them.
>  
> +1.1 Version 0.12
> +
> +Version 0.12 adds 32-bit floating point transforms.
> +
>  2. Acknowledgments
>  
>  This extension was the work of many people, in particular:
> @@ -252,6 +256,15 @@ ANIMCURSORELT	[
>  			cursor:		CURSOR
>  			delay:		CARD32
>  		]
> +6.1 Protocol Types added in Version 0.12
> +
> +FLOAT		32-bit IEEE 754 single precision float
> +TRANSFORMFLOAT	[
> +			p11, p12, p13:	FLOAT
> +			p21, p22, p23:	FLOAT
> +			p31, p32, p33:	FLOAT
> +		]
> +
>  7. Standard PictFormats
>  
>  The server must support a Direct PictFormat with 8 bits each of red, green,
> @@ -1143,6 +1156,29 @@ CreateConicalGradient
>  	The colors are non premultiplied.
>  
>  	
> +14.1 Extension requests added in Version 0.12
> +
> +SetPictureTransformFloat
> +
> +	picture:	PICTURE
> +	transform:	TRANSFORMFLOAT
> +
> +	Errors:
> +		Alloc, Value, Picture
> +
> +	This request changes the projective transformation used to
> +	map coordinates when 'picture' is used as the source or
> +	mask in any compositing operation.  The transform
> +	maps from destination pixel geometry back to the source pixel
> +	geometry.
> +
> +	Setting a transform via this request replaces any transform
> +	set by SetPictureTransform. Similarly, setting a transform via
> +	SetPictureTransform replaces any transform set with this
> +	request.
> +
> +	The matrix must be invertable, else a Value error is generated.
> +
>  15. Extension Versioning
>  
>  The Render extension was developed in parallel with the implementation to
> @@ -1206,3 +1242,5 @@ what each version before 1.0 implemented:
>  	0.11:
>  		Blend mode operators
>  
> +	0.12:
> +		Floating point transforms