mewin/libpng
1
0
Fork 0
mirror of https://git.code.sf.net/p/libpng/code.git synced 2025-07-10 18:04:09 +02:00
Code Issues Projects Releases Wiki Activity

Fixed multiple bugs in VSX filter functions

Browse Source 
Now up,sub and avg filter VSX functions pass tests
This commit is contained in:
Vadim Barkov 2017-02-08 13:58:23 +00:00
parent 58e9d5d59d
commit d4bdca45b3
1 changed files with 288 additions and 271 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

195
powerpc/filter_vsx_intrinsics.c
View File

@ -39,24 +39,28 @@
* ( this is taken from ../intel/filter_sse2_intrinsics.c ) * ( this is taken from ../intel/filter_sse2_intrinsics.c )
*/ */
#define declare_common_vars(row_info,row,prev_row) \ #define declare_common_vars(row_info,row,prev_row,offset) \
png_size_t i;\ png_size_t i;\
png_bytep rp = row;\ png_bytep rp = row + offset;\
png_const_bytep pp = prev_row;\ png_const_bytep pp = prev_row;\
png_size_t unaligned_top = 16 - (((png_size_t)row % 16));\ png_size_t unaligned_top = 16 - (((png_size_t)rp % 16));\
png_size_t istop;\ png_size_t istop;\
if(unaligned_top == 16)\ if(unaligned_top == 16)\
unaligned_top = 0;\ unaligned_top = 0;\
istop = row_info->rowbytes - unaligned_top; istop = row_info->rowbytes;\
if((unaligned_top < istop))\
istop -= unaligned_top;\
else{\
unaligned_top = istop;\
istop = 0;\
}
void png_read_filter_row_up_vsx(png_row_infop row_info, png_bytep row, void png_read_filter_row_up_vsx(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row) png_const_bytep prev_row)
{ {
vector unsigned char rp_vec; vector unsigned char rp_vec;
vector unsigned char pp_vec; vector unsigned char pp_vec;
declare_common_vars(row_info,row,prev_row,0)
declare_common_vars(row_info,row,prev_row)
/* Altivec operations require 16-byte aligned data /* Altivec operations require 16-byte aligned data
* but input can be unaligned. So we calculate * but input can be unaligned. So we calculate
@ -83,7 +87,7 @@ void png_read_filter_row_up_vsx(png_row_infop row_info, png_bytep row,
istop -= 16; istop -= 16;
} }
if(istop % 16 > 0) if(istop > 0)
{ {
/* If byte count of row is not divisible by 16 /* If byte count of row is not divisible by 16
* we will process remaining part as usual * we will process remaining part as usual
@ -116,14 +120,12 @@ void png_read_filter_row_up_vsx(png_row_infop row_info, png_bytep row,
#define VEC_AVG_SELECT3_3 (vector unsigned char){16,16,16,16,16,16,16,16,16, 9,10,11,16,16,16,16} #define VEC_AVG_SELECT3_3 (vector unsigned char){16,16,16,16,16,16,16,16,16, 9,10,11,16,16,16,16}
#define VEC_AVG_SELECT4_3 (vector unsigned char){16,16,16,16,16,16,16,16,16,16,16,16,12,13,14,16} #define VEC_AVG_SELECT4_3 (vector unsigned char){16,16,16,16,16,16,16,16,16,16,16,16,12,13,14,16}
#ifdef PNG_USE_ABS #ifdef PNG_USE_ABS
# define vsx_abs(number) abs(number) # define vsx_abs(number) abs(number)
#else #else
# define vsx_abs(number) (number > 0) ? (number) : -(number) # define vsx_abs(number) (number > 0) ? (number) : -(number)
#endif #endif
void png_read_filter_row_sub4_vsx(png_row_infop row_info, png_bytep row, void png_read_filter_row_sub4_vsx(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row) png_const_bytep prev_row)
{ {
@ -133,8 +135,7 @@ void png_read_filter_row_sub4_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char part_vec; vector unsigned char part_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp += bpp;
PNG_UNUSED(pp) PNG_UNUSED(pp)
@ -142,8 +143,7 @@ void png_read_filter_row_sub4_vsx(png_row_infop row_info, png_bytep row,
* but input can be unaligned. So we calculate * but input can be unaligned. So we calculate
* unaligned part as usual. * unaligned part as usual.
*/ */
for (i = 0; i < unaligned_top; i++)
for (i = bpp; i < unaligned_top; i++)
{ {
*rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff); *rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff);
rp++; rp++;
@ -171,14 +171,16 @@ void png_read_filter_row_sub4_vsx(png_row_infop row_info, png_bytep row,
istop -= 16; istop -= 16;
} }
if(istop % 16 > 0) if(istop > 0)
for (i = 0; i < istop % 16; i++) for (i = 0; i < istop % 16; i++)
{ {
*rp = (png_byte)(((int)(*rp) + (int)(*(rp - bpp))) & 0xff); *rp = (png_byte)(((int)(*rp) + (int)(*(rp - bpp))) & 0xff);
rp++; rp++;
} }
} }
void png_read_filter_row_sub3_vsx(png_row_infop row_info, png_bytep row, void png_read_filter_row_sub3_vsx(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row) png_const_bytep prev_row)
{ {
@ -188,16 +190,15 @@ void png_read_filter_row_sub3_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char part_vec; vector unsigned char part_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp += bpp;
PNG_UNUSED(pp) PNG_UNUSED(pp)
/* Altivec operations require 16-byte aligned data /* Altivec operations require 16-byte aligned data
* but input can be unaligned. So we calculate * but input can be unaligned. So we calculate
* unaligned part as usual. * unaligned part as usual.
*/ */
for (i = 0; i < unaligned_top; i++)
for (i = bpp; i < unaligned_top; i++)
{ {
*rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff); *rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff);
rp++; rp++;
@ -232,7 +233,7 @@ void png_read_filter_row_sub3_vsx(png_row_infop row_info, png_bytep row,
*(rp - 1) += *(rp - 1 - 3); *(rp - 1) += *(rp - 1 - 3);
} }
if(istop % 16 > 0) if(istop > 0)
for (i = 0; i < istop % 16; i++) for (i = 0; i < istop % 16; i++)
{ {
*rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff); *rp = (png_byte)(((int)(*rp) + (int)(*(rp-bpp))) & 0xff);
@ -252,7 +253,10 @@ void png_read_filter_row_avg4_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char avg_vec; vector unsigned char avg_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp -= bpp;
if(istop >= bpp)
istop -= bpp;
for (i = 0; i < bpp; i++) for (i = 0; i < bpp; i++)
{ {
@ -266,7 +270,7 @@ void png_read_filter_row_avg4_vsx(png_row_infop row_info, png_bytep row,
* but input can be unaligned. So we calculate * but input can be unaligned. So we calculate
* unaligned part as usual. * unaligned part as usual.
*/ */
for (i = bpp; i < unaligned_top; i++) for (i = 0; i < unaligned_top; i++)
{ {
*rp = (png_byte)(((int)(*rp) + *rp = (png_byte)(((int)(*rp) +
(int)(*pp++ + *(rp-bpp)) / 2 ) & 0xff); (int)(*pp++ + *(rp-bpp)) / 2 ) & 0xff);
@ -315,7 +319,7 @@ void png_read_filter_row_avg4_vsx(png_row_infop row_info, png_bytep row,
istop -= 16; istop -= 16;
} }
if(istop % 16 > 0) if(istop > 0)
for (i = 0; i < istop % 16; i++) for (i = 0; i < istop % 16; i++)
{ {
*rp = (png_byte)(((int)(*rp) + *rp = (png_byte)(((int)(*rp) +
@ -337,7 +341,10 @@ void png_read_filter_row_avg3_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char avg_vec; vector unsigned char avg_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp -= bpp;
if(istop >= bpp)
istop -= bpp;
for (i = 0; i < bpp; i++) for (i = 0; i < bpp; i++)
{ {
@ -351,7 +358,7 @@ void png_read_filter_row_avg3_vsx(png_row_infop row_info, png_bytep row,
* but input can be unaligned. So we calculate * but input can be unaligned. So we calculate
* unaligned part as usual. * unaligned part as usual.
*/ */
for (i = bpp; i < unaligned_top; i++) for (i = 0; i < unaligned_top; i++)
{ {
*rp = (png_byte)(((int)(*rp) + *rp = (png_byte)(((int)(*rp) +
(int)(*pp++ + *(rp-bpp)) / 2 ) & 0xff); (int)(*pp++ + *(rp-bpp)) / 2 ) & 0xff);
@ -393,19 +400,27 @@ void png_read_filter_row_avg3_vsx(png_row_infop row_info, png_bytep row,
avg_vec = vec_sub(avg_vec, vec_and(vec_xor(rp_part_vec,pp_part_vec),vec_splat_u8(1))); avg_vec = vec_sub(avg_vec, vec_and(vec_xor(rp_part_vec,pp_part_vec),vec_splat_u8(1)));
rp_vec = vec_add(rp_vec,avg_vec); rp_vec = vec_add(rp_vec,avg_vec);
rp_part_vec = vec_perm(rp_vec,zero_vec,VEC_SELECT4_3);
pp_part_vec = vec_perm(pp_vec,zero_vec,VEC_AVG_SELECT4_3);
avg_vec = vec_avg(rp_part_vec,pp_part_vec);
avg_vec = vec_sub(avg_vec, vec_and(vec_xor(rp_part_vec,pp_part_vec),vec_splat_u8(1)));
rp_vec = vec_add(rp_vec,avg_vec);
vec_st(rp_vec,0,rp); vec_st(rp_vec,0,rp);
rp += 16; rp += 15;
pp += 16; pp += 15;
istop -= 16; istop -= 16;
/* Since 16 % bpp = 16 % 3 = 1, last element of array must /* Since 16 % bpp = 16 % 3 = 1, last element of array must
* be proceeded manually * be proceeded manually
*/ */
*(rp - 1) += ((int)(*(pp-1) + *(rp-1-bpp)) / 2 ) & 0xff; *rp = (png_byte)(((int)(*rp) +
(int)(*pp++ + *(rp-bpp)) / 2 ) & 0xff);
rp++;
} }
if(istop % 16 > 0) if(istop > 0)
for (i = 0; i < istop % 16; i++) for (i = 0; i < istop % 16; i++)
{ {
*rp = (png_byte)(((int)(*rp) + *rp = (png_byte)(((int)(*rp) +
@ -413,7 +428,6 @@ void png_read_filter_row_avg3_vsx(png_row_infop row_info, png_bytep row,
rp++; rp++;
} }
} }
/* Bytewise c ? t : e. */ /* Bytewise c ? t : e. */
@ -443,10 +457,14 @@ void png_read_filter_row_paeth4_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char rp_vec; vector unsigned char rp_vec;
vector unsigned char pp_vec; vector unsigned char pp_vec;
vector unsigned char a_vec,b_vec,c_vec,nearest_vec; vector unsigned char a_vec,b_vec,c_vec,nearest_vec;
vector signed char pa_vec,pb_vec,pc_vec,smallest_vec; vector signed char pa_vec,pb_vec,pc_vec;
vector unsigned char pa_vec_abs,pb_vec_abs,pc_vec_abs,smallest_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp -= bpp;
if(istop >= bpp)
istop -= bpp;
/* Process the first pixel in the row completely (this is the same as 'up' /* Process the first pixel in the row completely (this is the same as 'up'
* because there is only one candidate predictor for the first row). * because there is only one candidate predictor for the first row).
@ -458,12 +476,12 @@ void png_read_filter_row_paeth4_vsx(png_row_infop row_info, png_bytep row,
pp++; pp++;
} }
for(i = bpp; i < unaligned_top ; i++) for(i = 0; i < unaligned_top ; i++)
{ {
vsx_paeth_process(rp,pp,a,b,c,pa,pb,pc,bpp) vsx_paeth_process(rp,pp,a,b,c,pa,pb,pc,bpp)
} }
while( istop > 16) while( istop >= 16)
{ {
for(i = 0; i < bpp ; i++) for(i = 0; i < bpp ; i++)
{ {
@ -481,14 +499,14 @@ void png_read_filter_row_paeth4_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
@ -498,16 +516,15 @@ void png_read_filter_row_paeth4_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT3_4); a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT3_4);
@ -516,16 +533,15 @@ void png_read_filter_row_paeth4_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
vec_st(rp_vec,0,rp); vec_st(rp_vec,0,rp);
@ -551,10 +567,14 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
vector unsigned char rp_vec; vector unsigned char rp_vec;
vector unsigned char pp_vec; vector unsigned char pp_vec;
vector unsigned char a_vec,b_vec,c_vec,nearest_vec; vector unsigned char a_vec,b_vec,c_vec,nearest_vec;
vector signed char pa_vec,pb_vec,pc_vec,smallest_vec; vector signed char pa_vec,pb_vec,pc_vec;
vector unsigned char pa_vec_abs,pb_vec_abs,pc_vec_abs,smallest_vec;
vector unsigned char zero_vec = {0}; vector unsigned char zero_vec = {0};
declare_common_vars(row_info,row,prev_row) declare_common_vars(row_info,row,prev_row,bpp)
rp -= bpp;
if(istop >= bpp)
istop -= bpp;
/* Process the first pixel in the row completely (this is the same as 'up' /* Process the first pixel in the row completely (this is the same as 'up'
* because there is only one candidate predictor for the first row). * because there is only one candidate predictor for the first row).
@ -566,12 +586,12 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
pp++; pp++;
} }
for(i = bpp; i < unaligned_top ; i++) for(i = 0; i < unaligned_top ; i++)
{ {
vsx_paeth_process(rp,pp,a,b,c,pa,pb,pc,bpp) vsx_paeth_process(rp,pp,a,b,c,pa,pb,pc,bpp)
} }
while( istop > 16) while( istop >= 16)
{ {
for(i = 0; i < bpp ; i++) for(i = 0; i < bpp ; i++)
{ {
@ -589,14 +609,14 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
@ -606,16 +626,15 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT3_3); a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT3_3);
@ -624,16 +643,15 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT4_3); a_vec = vec_perm(rp_vec , zero_vec , VEC_SELECT4_3);
@ -642,24 +660,23 @@ void png_read_filter_row_paeth3_vsx(png_row_infop row_info, png_bytep row,
pa_vec = (vector signed char) vec_sub(b_vec,c_vec); pa_vec = (vector signed char) vec_sub(b_vec,c_vec);
pb_vec = (vector signed char) vec_sub(a_vec , c_vec); pb_vec = (vector signed char) vec_sub(a_vec , c_vec);
pc_vec = (vector signed char) vec_add(pa_vec,pb_vec); pc_vec = (vector signed char) vec_add(pa_vec,pb_vec);
pa_vec = vec_abs(pa_vec); pa_vec_abs = (vector unsigned char)vec_abs(pa_vec);
pb_vec = vec_abs(pb_vec); pb_vec_abs = (vector unsigned char)vec_abs(pb_vec);
pc_vec = vec_abs(pc_vec); pc_vec_abs = (vector unsigned char)vec_abs(pc_vec);
smallest_vec = vec_min(pc_vec, vec_min(pa_vec,pb_vec)); smallest_vec = vec_min(pc_vec_abs, vec_min(pa_vec_abs,pb_vec_abs));
nearest_vec = if_then_else( nearest_vec = if_then_else(
vec_cmpeq(pa_vec,smallest_vec), vec_cmpeq(pa_vec_abs,smallest_vec),
a_vec, a_vec,
if_then_else(vec_cmpeq(pb_vec,smallest_vec),b_vec,c_vec) if_then_else(vec_cmpeq(pb_vec_abs,smallest_vec),b_vec,c_vec)
); );
rp_vec = vec_add(rp_vec, nearest_vec); rp_vec = vec_add(rp_vec, nearest_vec);
vec_st(rp_vec,0,rp); vec_st(rp_vec,0,rp);
rp += 16-1; rp += 15;
pp += 16-1; pp += 15;
istop -= 16; istop -= 16;
/* Since 16 % bpp = 16 % 3 = 1, last element of array must /* Since 16 % bpp = 16 % 3 = 1, last element of array must
* be proceeded manually * be proceeded manually
*/ */