cheng/libpng
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[libpng15] Optimized png_combine_row() when rows are aligned This gains a

Browse Source 
small percentage for 16-bit and 32-bit pixels in the typical case where the
output row buffers are appropriately aligned. The optimization was not
previously possible because the png_struct buffer was always misaligned.
This commit is contained in:
John Bowler 2011-10-07 18:36:37 -05:00 committed by Glenn Randers-Pehrson
parent ac8375d000
commit 8be28e3eb6
3 changed files with 24 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ANNOUNCE
View File

@ -64,6 +64,10 @@ Version 1.5.6beta05 [October 7, 2011]
a bug in the code that attempted to align it; the code needs to subtract a bug in the code that attempted to align it; the code needs to subtract
one from the pointer to take account of the filter byte prepended to one from the pointer to take account of the filter byte prepended to
each row. each row.
Optimized png_combine_row() when rows are aligned This gains a small
percentage for 16-bit and 32-bit pixels in the typical case where the
output row buffers are appropriately aligned. The optimization was not
previously possible because the png_struct buffer was always misaligned.
Send comments/corrections/commendations to png-mng-implement at lists.sf.net: Send comments/corrections/commendations to png-mng-implement at lists.sf.net:
(subscription required; visit (subscription required; visit

4
CHANGES
View File

@ -3625,6 +3625,10 @@ Version 1.5.6beta05 [October 7, 2011]
a bug in the code that attempted to align it; the code needs to subtract a bug in the code that attempted to align it; the code needs to subtract
one from the pointer to take account of the filter byte prepended to one from the pointer to take account of the filter byte prepended to
each row. each row.
Optimized png_combine_row() when rows are aligned This gains a small
percentage for 16-bit and 32-bit pixels in the typical case where the
output row buffers are appropriately aligned. The optimization was not
previously possible because the png_struct buffer was always misaligned.
Send comments/corrections/commendations to png-mng-implement at lists.sf.net Send comments/corrections/commendations to png-mng-implement at lists.sf.net
(subscription required; visit (subscription required; visit

27
pngrutil.c
View File

@ -2931,12 +2931,13 @@ png_combine_row(png_structp png_ptr, png_bytep dp, int display)
/* This is a common optimization for 2 and 4 byte pixels, for other /* This is a common optimization for 2 and 4 byte pixels, for other
* values rely on the toolchain memcpy being optimized. * values rely on the toolchain memcpy being optimized.
*
* TBD: this should use png_isaligned, but currently something isn't
* aligned (NOTE: to be investigated in a really serious fashion.)
*/ */
else if (pixel_depth == 2) else if (pixel_depth == sizeof (png_uint_16) &&
png_isaligned(sp, png_uint_16) && png_isaligned(dp, png_uint_16))
{ {
png_uint_16p dp16 = (png_uint_16p)dp;
png_uint_16p sp16 = (png_uint_16p)sp;
do do
{ {
m >>= 1; m >>= 1;
@ -2945,16 +2946,20 @@ png_combine_row(png_structp png_ptr, png_bytep dp, int display)
m = mask; m = mask;
if (m & 1) if (m & 1)
dp[0] = sp[0], dp[1] = sp[1]; *dp16 = *sp16;
dp += 2; ++dp16;
sp += 2; ++sp16;
} }
while (--row_width > 0); while (--row_width > 0);
} }
else if (pixel_depth == 4) /* as above, not optimal */ else if (pixel_depth == sizeof (png_uint_32) &&
png_isaligned(sp, png_uint_32) && png_isaligned(dp, png_uint_32))
{ {
png_uint_32p dp32 = (png_uint_32p)dp;
png_uint_32p sp32 = (png_uint_32p)sp;
do do
{ {
m >>= 1; m >>= 1;
@ -2963,10 +2968,10 @@ png_combine_row(png_structp png_ptr, png_bytep dp, int display)
m = mask; m = mask;
if (m & 1) if (m & 1)
dp[0] = sp[0], dp[1] = sp[1], dp[2] = sp[2], dp[3] = sp[3]; *dp32 = *sp32;
dp += 4; ++dp32;
sp += 4; ++sp32;
} }
while (--row_width > 0); while (--row_width > 0);
} }