__builtin_shuffle() was introduced in gcc-4.7. The lowest
gcc to enable vector operation is delayed from 4.6 to 4.7.
* src/sfnt/pngshim.c (premultiply_data): Fix cpp-macro to
enable the vector operation, to change the lowest gcc version
from 4.6 to 4.7.
Improve the code by 5d3ff05615 ,
issues are found by Behdad Esfahbod and Werner Lemberg.
* src/cache/ftcbasic.c (FTC_ImageCache_Lookup): Replace
a subtraction to check higher bit by a bit operation,
and cpp-conditionalize for appropriate systems. Add better
documentation to the comment.
(FTC_ImageCache_LookupScaler): Ditto.
(FTC_SBitCache_Lookup): Ditto.
(FTC_SBitCache_LookupScaler): Ditto.
* src/autofit/afshaper.c: Include FT_ADVANCE_H, to use
FT_Get_Advance() in it.
* src/sfnt/ttcmap.c: Include FT_SERVICE_POSTSCRIPT_CMAPS_H
to use PS_Unicodes in it, also include `ttpost.h' to use
tt_face_get_ps_name() in it.
LastResort.dfont has a marginal resource ID 0xFFFF for sfnt
resource. Inside Macintosh: More Macintosh Toolbox, `Resource IDs'
(1-46), tells that some IDs are reserved and should not be used.
FreeType2 just uses resource ID to sort the fragmented resource.
To accept the marginal fonts, the checking is removed.
* src/base/ftrfork.c (FT_Raccess_Get_DataOffsets): Remove res_id
validity check, fix a trace message format.
The documentation of `FT_Bitmap_Convert' says that multiple calls do
proper reallocation of the target FT_Bitmap object. However, this
failed for the sequence
non-empty bitmap
empty bitmap
non-empty bitmap
Reason was that `FT_Bitmap_Convert' only reallocated the bitmap
buffer if it became too small; it didn't make the buffer smaller.
For an empty bitmap following a non-empty one, only the buffer
dimension got set to zero, without deallocation. If the next call
was a non-empty buffer again, an assertion in `ft_mem_qrealloc' was
triggered.
* src/base/ftbitmap.c (FT_Bitmap_Convert): Always reallocate target
buffer to the correct size.
* docs/CHANGES: Document it.
* src/bdf/bdfdrivr.c (BDF_Face_Init): Use `SIZE' values if
`POINT_SIZE', `RESOLUTION_X', or `RESOLUTION_Y' properties are
missing.
* docs/CHANGES: Document it.
* src/sfnt/pngshim.c (premultiply_data): Use vectors instead of
scalars.
(vector_shuffle): New macro to take of a different built-in function
name on clang.
This reduces the overhead of `premultiply_data' by 60%.
* src/sfnt/pngshim.c (premultiply_data): Provide code which uses
gcc's (and clang's) `vector_byte' attribute to process 4 pixels at a
time.
Requested by Behdad.
Modern bitmap-only SFNTs like `NotoColorEmoji.ttf' don't contain
entries in the bitmap strike(s) for empty glyphs. Instead, they
rely that a space glyph gets created from the font's metrics data.
This commit makes FreeType behave accordingly.
* include/freetype/fterrdef.h (FT_Err_Missing_Bitmap): New error
code.
* src/sfnt/ttsbit.c (tt_sbit_decoder_load_image): Change error codes
to make a distinction between a missing bitmap in a composite and a
simple missing bitmap.
* src/truetype/ttgload.c (TT_Load_Glyph): For a missing bitmap (in a
bitmap-only font), synthesize an empty bitmap glyph if metrics are
available.
We now use a separate preprocessor macro to handle both definition
and declaration of the glyph name arrays.
* src/psnames/psmodule.c (DEFINE_PS_TABLE_DATA): New macro.
* src/tools/glnames.py (StringTable::dump,
StringTable::dump_sublist): Use `DEFINE_PS_TABLE_DATA'.
(dump_encoding): Ditto.
(main): Use `wb' mode for writing the output file, which works on
Windows also.
* src/psnames/pstables.h: Regenerated.
This is a new technology for LCD-optimized rendering. It capitalizes
on the fact that each color channel grid is shifted by a third of a
pixel. Therefore it is logical to render 3 separate monochrome
bitmaps shifting the outline by 1/3 pixel, and then combine them.
Importantly, the resulting output does not require additional LCD
filtering.
* src/smooth/ftsmooth.c (ft_smooth_render_generic)
[!FT_CONFIG_OPTION_SUBPIXEL_RENDERING]: Implement new LCD-optimized
rendering.
* include/freetype/ftlcdfil.h, include/freetype/freetype.h,
include/freetype/config/ftoption.h, devel/ftoption.h: Updated
documentation.
The reserved `flags' field got a value in OpenType version 1.8.2;
unfortunately, the public `FT_Var_Axis' structure misses the
corresponding element. Since we can't add a new field, we add an
access function.
* src/base/ftmm.c (FT_Get_Var_Axis_Flags): New function.
* include/freetype/ftmm.h (FT_VAR_AXIS_FLAG_HIDDEN): New macro.
Updated.
* src/truetype/ttgxvar.c (TT_Get_MM_Var): Increase allocated memory
of `mmvar' to hold axis flags.
Fill the axis flags array.
* docs/CHANGES: Updated.
Phantom points are now saved outside v40 backwards compatibility
mode. This fixes the jumping glyphs when switching between v35 and
v40 monochrome mode.
* src/truetype/ttgload.c (TT_Hint_Glyph): Fix inversed bool logic.
This fixes weird behavior of instructions that resulted in rendering
differences between v35 and v40 in monochrome mode, e.g., in
`timesbi.ttf'.
* src/truetype/ttinterp.c (Ins_GETINFO)
[TT_SUPPORT_SUBPIXEL_HINTING_MINIMAL]: Check
`subpixel_hinting_lean'.
* src/sfnt/sfobjs.c (sfnt_load_face): The units per EM value has now
(tighter) limits.
* src/truetype/ttgload.c (load_truetype_glyph): The new OpenType
version explicitly allows all negative values for the number of
contours if we have a composite glyph (this is for better backwards
compatibility I guess), but it still recommends value -1.
While clang's sanitizer recommends a cast to unsigned for safe
negation (to handle -INT_MIN), both MSVC and Visualc emit warning
C4146 if an unsigned value gets negated.
* include/freetype/internal/ftcalc.h (NEG_LONG, NEG_INT32),
src/base/ftcalc.c (FT_MOVE_SIGN): Replace negation with a
subtraction.
Werner Lemberg