We used to split large glyphs into horizontal bands and continue
bisecting them still horizontally if that was not enough. This is
guaranteed to fail when a single scanline cannot fit into the
rendering memory pool. Now we bisect the bands vertically so that
the smallest unit is a column of the band height, which is guranteed
to fit into memory.
* src/smooth/ftgrays.c (gray_convert_glyph): Implement it.
At large sizes almost but not exactly horizontal segments can quickly
drain the rendering pool. This patch at least avoids filling the pool
with trivial cells. Beyond this, we can only increase the pool size.
Reported, analyzed, and tested by Colin Fahey.
* src/smooth/ftgrays.c (gray_set_cell): Do not record trivial cells.
* src/raster/ftrend1.c (ft_raster1_render): Clean up the exit.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Reduce
translations and clean up the exit.
(ft_smooth_render_lcd, ft_smooth_render_lcd): Remove unused `error'..
This mainly just extracts the code for presetting the bitmap metrics
from the monochrome, grayscale, and LCD renderers into a separate
function.
* src/base/ftobjs.c (ft_glyphslot_preset_bitmap): New function that
calculates prespective bitmap metrics for the given rendering mode.
* include/freetype/internal/ftobjs.h (ft_glyphslot_preset_bitmap):
Declare it.
* src/base/ftlcdfil.c (ft_lcd_padding): New helper function that adds
padding to CBox taking into account pecularities of LCD rendering.
* include/freetype/ftlcdfil.h (ft_lcd_padding): Declare it.
* src/raster/ftrend1.c (ft_raster1_render): Reworked to use
`ft_glyphslot_preset_bitmap'.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Ditto.
(ft_smooth_render_lcd, ft_smooth_render_lcd): The pixel_mode setting
is moved to `ft_glyphslot_preset_bitmap'.
Thing are simpler with a NULL-function pointer.
* include/freetype/internal/ftobjs.h (FT_Face_InternalRec): New
pointer to the filter function.
(FT_LibraryRec): Remove unused `lcd_filter'.
(FT_Bitmap_LcdFilterFunc, ft_lcd_filter_fir): Move from here...
* include/freetype/ftlcdfil.h (FT_Bitmap_LcdFilterFunc,
ft_lcd_filter_fir): ... to here.
* src/base/ftobjs.c (ft_open_face_internal): NULL-initialize the
per-face filter.
(FT_Face_Properties): Set it.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Simplify.
* src/base/ftlcdfil.c (ft_lcd_filter_fir, FT_Libary_SetLcdFilter):
Minor.
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.
* src/smooth/ftgrays.c [STANDALONE] (OVERFLOW_ADD_LONG,
OVERFLOW_SUB_LONG, OVERFLOW_MUL_LONG, NEG_LONG): New macros.
[!STANDALONE]: Include FT_INTERNAL_CALC_H.
(gray_render_cubic): Use those macros where appropriate.
Extra bitmap padding for LCD filtering depends on the filter. The
default 5-tap filter needs 2 extra subpixels. The light 3-tap filter
needs only 1 extra subpixel. This space could be already available
due to rounding. In order to optimize the padding, we now expand
CBox for the given filter weights before rounding.
This change breakes current Skia (and Firefox).
* include/freetype/internal/ftobjs.h (FT_LibraryRec)
[FT_CONFIG_OPTION_SUBPIXEL_RENDERING]: Remove `lcd_extra' field.
* src/base/ftlcdfil.c (FT_Library_SetLcdFilterWeights,
FT_Library_SetLcdFilter): Remove `lcd_extra' initializations.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Implement dymanic
LCD padding.
This mode uses fractional advance widths and doesn't scale glyphs
horizontally, only applying vertical scaling and hinting.
At the same time, the behaviour of the `light' auto-hinter gets
restored for backwards compatibility: Both vertical and horizontal
scaling is again based on rounded metrics values (this was changed
in a commit from 2017-03-30 as a side effect). To be more precise,
the behaviour is restored for TrueType fonts only; for other font
formats like Type 1, this is a new feature of the `light' hinting
mode.
* include/freetype/freetype.h (FT_LOAD_TARGET_SLIGHT): New macro.
(FT_RENDER_MODE_SLIGHT): New render mode.
* include/freetype/internal/ftobjs.h (FT_Size_InternalRec): Add
`autohint_mode' and `autohint_metrics' fields.
* src/autofit/afcjk.c (af_cjk_hints_init), src/autofit/aflatin.c
(af_latin_hints_init), src/autofit/aflatin2 (af_latin2_hints_init):
Updated.
* src/autofit/afloader.c (af_loader_embolden_glyph_in_slot): Use
`autohint_metrics'.
(af_loader_load_glyph): s/internal/slot_internal/.
Initialize `autohint_metrics' and `autohint_mode' depending on
current auto-hint mode.
Use `autohint_metrics'.
Updated.
* src/base/ftadvanc.c (LOAD_ADVANCE_FAST_CHECK): Updated.
* src/base/ftobjs.c (FT_Load_Glyph): Updated.
(FT_New_Size): Allocate `internal' object.
* src/pshinter/pshalgo.c (ps_hints_apply): Updated.
* src/smooth/ftsmooth.c (ft_smooth_render): Updated.
We don't need some divisions if a line segments stays within a single
row or a single column of pixels.
* src/smooth/ftgrays.c (gray_render_line) [FT_LONG64]: Make divisions
conditional.
The algorithm calls `gray_set_cell' at the start of each new contour
or when the contours cross the cell boundaries. Double-checking for
that is wasteful.
* src/smooth/ftgrays.c (gray_set_cell): Remove check for a new cell.
(gray_convert_glyph): Remove initialization introduced by 44b172e88.
* src/smooth/ftgrays.c (gray_move_to): Call `gray_set_cell' directly
instead of...
(gray_start_cell): ... this function, which is removed.
(gray_convert_glyph): Make initial y-coordinate invalid.
It turns out that there is significant cost associated with `FT_Span'
creation and calls to `gray_render_span' because it happerns so
frequently. This removes these steps from our internal use but leaves
it alone for `FT_RASTER_FLAG_DIRECT" to preserve API. The speed gain
is about 5%.
* src/smooth/ftgrays.c (gray_render_span): Removed. The code is
migrated to...
(gray_hline): ... here.
Zero coverage is unlikely (1 out of 256) to warrant checking. This
gives 0.5% speed improvement in dendering simple glyphs.
* src/smooth/ftgrays.c (gray_hline, gray_render_span): Remove checks.
This gives 2% speed improvement in rendering simple glyphs.
* src/smooth/ftgrays.c (TPixmap): Reduced pixmap descriptor with a
pointer to its bottom-left and pitch to be used in...
(gray_TWorker): ... here.
(gray_render_span): Move pixmap flow check from here...
(gray_raster_render): .. to here.
This removes unnecessary complexity of span merging and buffering.
Instead, the spans are rendered as they come, speeding up the
rendering by about 5% percents as a result.
* src/smooth/ftgrays.c [FT_MAX_GRAY_SPANS]: Macro removed.
(gray_TWorker): Remove span buffer and related fields.
(gray_sweep, gray_hline): Updated.
* include/freetype/ftimage.h: Remove documentation note about
`FT_MAX_GRAY_SPANS', which was never in `ftoption.h' and is now gone.
Rasterization sub-banding is utilized at large sizes while using
rather small fixed memory pool. Indeed it is possible to make an
educated guess how much memory is necessary at a given size for a
given glyph. It turns out that, for large majority of European glyphs,
you should store about 8 times more boundary pixels than their height.
Or, vice versa, if your memory pool can hold 800 pixels the band
height should be 100 and you should sub-band anything larger than
that. Should you still run out of memory, FreeType bisects the band
but you have wasted some time. This is what has been implemented in
FreeType since the beginning.
It was overlooked, however, that the top band could grow to twice the
default band size leading to unnecessary memory overflows there. This
commit fixes that. Now the bands are distributed more evenly and
cannot exceed the default size.
Now the magic number 8 is really suitable for rather simple European
scripts. For complex Chinese logograms the magic number should be 13
but that is subject for another day.
* src/smooth/ftgrays.c (gray_convert_glyph): Revise sub-banding
protocol.
* src/smooth/ftgrays.c (TArea): Restore original definition as `int'.
(gray_render_line) [FT_LONG64]: Updated.
(gray_convert_glyph): 32-bit band bisection stack should be 32 bands.
(gray_convert_glyph_inner): Trace successes and failures.
This patch restores original `TCoord' definition as `int' so that the
rendering pool is used more efficiently on LP64 platforms (unix).
* src/smooth/ftgrays.c (gray_TWorker, TCell, gray_TBand): Switch some
fields to `TCoord'.
(gray_find_cell, gray_render_scanline, gray_render_line, gray_hline,
gray_sweep, gray_convert_glyph): Updated.
* src/smooth/ftgrays.c (gray_convert_glyph): Do not use volatile
qualifier.
(gray_raster_render): Move span initializations from here.
(gray_sweep): ... to here and remove unused `target' argument.
* src/smooth/ftgrays.c (gray_init_cells): Remove function.
(gray_TWorker): Remove fields that become local variables.
(gray_raster_render): Move rendering buffer declaration from here.
(gray_convert_glyph): ... to here and update accordingly.
Removing the checks from `gray_hline' shaves 1% off rendering speed.
* src/smooth/ftgrays.c [STANDALONE_]: Duplicate `FT_MIN' and `FT_MAX'.
(gray_TWorker): No need to store `clip_box'.
(gray_hline): Remove unnecessary boundary checks.
(gray_convert_glyph): Move boundary checks from here.
(gray_raster_render): ... to here and consolidate.
* src/smooth/ftgrays.c [STANDALONE_]: Duplicate `FT_Outline_Get_CBox'.
(gray_compute_cbox): Remove this function.
(gray_convert_glyph): Update to use `FT_Outline_Get_CBox'.
The convergence of Bézier flatteners is fast with the deviation
from straight line being assymptotically cut 4-fold on each bisection.
This justifies smaller bisection stack size.
* src/smooth/ftgrays.c (gray_TWorker): Remove common `bez_stack'.
(gray_render_conic): Create and use conic `bez_stack'. Move back the
band analysis from...
(gray_conic_to): ... here.
(gray_render_cubic): Create and use cubic `bez_stack'. Move back the
band analysis from...
(gray_cubic_to): ... here.
(gray_move_to): Updated.
A single decrement counter of segments to draw, instead of an array,
contains all the information necessary to decide when to split and
when to draw a conic segment. The number of splits before each draw is
equal to the number of trailing zeros in the counter.
* src/smooth/ftgrays.c (gray_TWorker): Remove `lev_stack'.
(gray_render_conic): Updated to use decrement counter of segments.
* src/smooth/ftgrays.c (gray_render_conic, gray_render_cubic): Move
band clipping from here.
(gray_conic_to, gray_cubic_to): ... to here.
(gray_rander_line, gray_render_scanline): Initialize variables closer
to their use.
Such macro names are reserved for both C and C++.
* include/freetype/ftimage.h, src/raster/ftraster.c,
src/smooth/ftgrays.c, src/smooth/ftgrays.h:
s/_STANDALONE_/STANDALONE_/.
This implementation renders the entire line segment at once without
subdividing it into scanlines. The main speed improvement comes from
reducing the number of divisions to just two per line segment, which
is a bare minimum to calculate cell coverage in a smooth rasterizer.
Notably, the progression from cell to cell does not itself require any
divisions at all. The speed improvement is more noticeable at larger
sizes.
* src/smooth/ftgrays.c (gray_render_line): New implementation.
See
http://lists.nongnu.org/archive/html/freetype-devel/2015-07/msg00008.html
for a rationale.
* src/autofit/afpic.h, src/base/basepic.h, src/cff/cffpic.h,
src/pshinter/pshpic.h, src/psnames/pspic.h, src/raster/rastpic.h,
src/sfnt/sfntpic.h, src/smooth/ftspic.h, src/truetype/ttpic.h
(FT_BEGIN_HEADER, FT_END_HEADER): Move macro calls to not enclose
header files that contain FT_{BEGIN,END}_HEADER macros by
themselves.
* src/autofit/aftypes.h [FT_DEBUG_AUTOFIT]: Include
FT_CONFIG_STANDARD_LIBRARY_H earlier.
* src/truetype/ttpic.h: Include FT_INTERNL_PIC_H.
This change is a result of a discussion thread on freetype-devel
http://lists.nongnu.org/archive/html/freetype-devel/2015-06/msg00041.html
Re-introduce the `freetype2' subdirectory for all FreeType header
files after installation, and rename the `freetype2' subdirectory in
the git repository to `freetype'.
* include/freetype2: Renamed to...
* include/freetype: This.
* CMakeLists.txt (PUBLIC_HEADERS, PUBLIC_CONFIG_HEADERS,
PRIVATE_HEADERS): Updated.
Update creation of `ftconfig.h'.
Install generated `ftconfig.h'.
* Jamfile (HDRMACRO, RefDoc), autogen.sh: Updated.
* builds/amiga/include/config/ftconfig.h, builds/freetype.mk
(PUBLIC_DIR), builds/symbian/bld.inf, builds/toplevel.mk (work),
builds/unix/freetype2.in: Updated.
* builds/unix/freetype-config.in: Updated.
* builds/unix/configure.raw: Don't check for `rmdir'.
* builds/unix/unix-def.in (DELDIR): Use `rm -rf', which is portable
according to the autoconf info manual.
* builds/unix/install.mk (install, uninstall,
distclean_project_unix): Update and simplify.
* builds/wince/*, builds/windows/*: Updated.
* devel/ft2build.h, include/ft2build.h: Updated.
* include/freetype2/config/ftheader.h,
include/freetype2/internal/ftserv.h,
include/freetype2/internal/internal.h: Update all header file
macros.
* src/tools/chktrcmp.py (TRACE_DEF_FILES): Updated.
* docs/*: Updated.
We no longer `pollute' the namespace of possible header file names;
instead we move `ft2build.h' up by one level so that it gets
installed in the default include directory (e.g.,
/usr/local/include). After this commit, only `ft2build.h' stays in
the compiler's include path.
No visible changes for the user who follows the standard FreeType
header inclusion rules.
* include/*: Move to ...
* include/freetype2/*: This directory, except `ft2build.h'.
* CMakeLists.txt (PUBLIC_HEADERS, PUBLIC_CONFIG_HEADERS,
PRIVATE_HEADERS), Jamfile (HDRMACRO, RefDoc), autogen.sh: Updated.
* builds/amiga/include/config/ftconfig.h, builds/freetype.mk
(PUBLIC_DIR), builds/symbian/bld.inf, builds/toplevel.mk (work),
builds/unix/install.mk (install, uninstall),
builds/unix/freetype2.in: Updated.
* builds/unix/freetype-config.in: Updated.
Emit -I directory only if it is not `/usr/include'.
* builds/wince/*, builds/windows/*: Updated.
* devel/ft2build.h, include/ft2build.h: Updated.
* include/freetype2/config/ftheader.h,
include/freetype2/internal/ftserv.h,
include/freetype2/internal/internal.h: Update all header file
macros.
* src/tools/chktrcmp.py (TRACE_DEF_FILES): Updated.
* docs/*: Updated.
This is just very basic stuff and just a little bit tested on
GNU/Linux only. I won't delve into this since I'm not a Jam user.
* Jamfile: Call `HDRMACRO' for `ftserv.h' also.
(DEFINES): Replace with...
(CCFLAGS): ... this.
* src/Jamfile: Don't call `HDRMACRO' for `internal.h'; this is
already handled in the top-level Jamfile.
* src/autofit/Jamfile (DEFINES): Replace with...
(CCFLAGS): ... this.
(_sources): Add missing files.
* src/cache/Jamfile: Don't call `HDRMACRO' for `ftcache.h'; it no
longer contains macro header definitions.
* src/base/Jamfile, src/cff/Jamfile, src/sfnt/Jamfile,
src/truetype/Jamfile (_sources): Add missing files.
Looks like `ras.span_y' could always be used without initialization.
This was never detected by valgrind before because the library-wide
`raster_pool' was used for the worker object and `raster_pool' was
originally zero'ed. But subsequent reuses of it were using `span_y'
uninitialized. With the recent change to not use `render_pool' and
allocate worker and buffer on the stack, valgrind now detects this
uninitialized access.
* src/smooth/ftgrays.c (gray_raster_render): Initialize
`ras.span_y'.
Assure that FreeType's internal include directories are found before
`CPPFLAGS' (which might be set by the user in the environment), and
`CPPFLAGS' before `CFLAGS'.
* builds/freetype.mk (FT_CFLAGS): Don't add `INCLUDE_FLAGS'.
(FT_COMPILE): Make this a special variable for compiling only the
files handled in `freetype.mk'.
(.c.$O): Removed, unused.
* src/*/rules.mk (*_COMPILE): Fix order of include directories.
suggested by Alexei.
* src/pfr/pfrsbit.c (pfr_slot_load_bitmap): Prevent too
negative values in `xpos' and `ypos + ysize'.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Prevent
too negative values in `x_left' and `y_top'. Either negative
values in `width' and `height' are checked.
* src/raster/ftraster.c (ft_black_set_mode): Change return type to
`int' to stay in sync with `FT_Renderer_SetModeFunc' prototype.
* src/smooth/ftgrays.c (gray_raster_set_mode): New dummy function
for orthogonality.
(ft_grays_raster): Use it.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Translate origin
virtually by modifying cbox, actually translate outline if cumulative
shift is not zero.
* include/internal/ftcalc.h (FT_HYPOT): Move macro from here...
* include/internal/ftobjs.h: ... to here, next to required `FT_ABS'.
* src/smooth/ftgrays.c (gray_render_cubic): Use it here.
This large patch changes the header file directory layout from
`include/freetype/...' to `include/...', effectively removing one
level. Since the file `ft2build.h' is also located in `include'
(and it stays there even after installation), all FreeType header
files are now in a single directory.
Applications that use (a) `freetype-config' or FreeType's
`pkg-config' file to get the include directory for the compiler, and
(b) the documented way for header inclusion like
#include <ft2build.h>
#include FT_FREETYPE_H
...
don't need any change to the source code.
* include/freetype/*: Move up to...
* include/*: ... this directory.
* builds/amiga/include/freetype/*: Move up to...
* builds/amiga/include/*: ... this directory.
*/*: Essentially do `s@/freetype/@/@' where appropriate.
* CMakeList.txt: Simplify.
* builds/unix/freetype-config.in, builds/unix/freetype2.in: For
`--cflags', return a single directory.
* builds/unix/install.mk (install): No longer try to remove `cache'
and `internal' subdirectories; instead, remove the `freetype'
subdirectory.
Provide a work-around for an ARM-specific performance bug in GCC.
This speeds up the rasterizer by more than 5%.
Also slightly optimize `set_gray_cell' and `gray_record_cell' (which
also improves performance on other platforms by a tiny bit (<1%).
* src/smooth/ftgrays.c (FT_DIV_MOD): New macro.
Use it where appropriate.
(gray_record_cell, gray_set_cell, gray_move_to,
gray_convert_glyph_inner): Streamline condition handling.
http://cppcheck.sourceforge.net/
Note that the current version heavily chokes on FreeType, delivering
even wrong results. I will report those issues to the cppcheck team
so that a newer version gives improved results hopefully.
*/* Improve variable scopes.
*/* Remove redundant initializations which get overwritten.
* src/base/ftmac.c ,builds/mac/ftmac.c (count_faces_scalable):
Remove unused variable.
* src/base/ftdbgmem.c (ft_mem_table_destroy): `table' can't be zero.
* src/gxvalid/gxvkern.c (gxv_kern_subtable_fmt1_entry_validate):
Remove functionless code.
* src/tools/ftrandom.c (main): Fix memory leak.
This is essentially a mechanical conversion, adding inclusion of
`FT_INTERNAL_DEBUG_H' where necessary, and providing the macros for
stand-alone compiling modes of the rasterizer modules.
To convert the remaining occurrences of FT_Err_XXX and friends it is
necessary to rewrite the code. Note, however, that it doesn't harm
if some cases are not handled since FT_THROW is a no-op.
* src/smooth/ftsmooth.c (ft_smooth_render_generic): Use `FT_Pos'
instead of `FT_UInt' for some variables and update comparisons
accordingly. A detailed analysis can be found in the bug report.
Originally FT_DEFINE_{DRIVER,MODULE,RENDERER}() macros were
designed to declare xxx_pic_{free,init} by themselves.
Because these macros are used at the end of the module
interface (e.g. ttdriver.c) and the wrapper source to build
a module as a single object (e.g. truetype.c) includes
the PIC file (e.g. ttpic.c) before the module interface,
these macros are expanded AFTER xxx_pic_{free,init} body
when the modules are built as single object.
The declaration after the implementation causes the redundant
declaration warnings, so the declarations are moved to module
PIC headers (e.g. ttpic.h). Separating to other header files
are needed for multi build.
* include/freetype/internal/ftdriver.h (FT_DEFINE_DRIVER):
Remove class_##_pic_free and class_##_pic_init declarations.
* include/freetype/internal/ftobjs.h (FT_DEFINE_RENDERER,
FT_DEFINE_MODULE): Ditto.
* src/base/basepic.h: Insert a comment and fix coding style.
* src/autofit/afpic.h: Declare autofit_module_class_pic_{free,
init}.
* src/cff/cffpic.h: Declare cff_driver_class_pic_{free,init}.
* src/pshinter/pshpic.h: Declare pshinter_module_class_pic_{free,
init}.
* src/psnames/pspic.h: Declare psnames_module_class_pic_{free,
init}.
* src/raster/rastpic.h: Declare
ft_raster{1,5}_renderer_class_pic_{free,init}
* src/sfnt/sfntpic.h: Declare sfnt_module_class_pic_{free,init}.
* src/smooth/ftspic.h: Declare
ft_smooth_{,lcd_,lcdv_}renderer_class_pic_{free,init}.
* src/truetype/ttpic.h: Declare tt_driver_class_pic_{free,init}.
error codes with per-module prefix.
* src/autofit/afpic.c: Include `aferrors.h'.
* src/cff/cffpic.c: Include `cfferrs.h'.
* src/pshinter/pshpic.c: Include `pshnterr.h'.
* src/raster/rastpic.c: Include `rasterrs.h'.
* src/sfnt/sfntpic.c: Include `sferrors.h'.
* src/smooth/ftspic.c: Include `ftsmerrs.h'.
* src/truetype/ttpic.c: Include `tterrors.h'.
Werner Lemberg