lavfi/gradfun: fix rounding in MMX code.

Current code divides before increasing precision.

Also reduce upper bound for strength from 255 to 64.  This will prevent
an overflow in the SSSE3 and MMX filter_line code: delta is expressed as
an u16 being shifted by 2 to the left. If it overflows, having a
strength not above 64 will make sure that m is set to 0 (making the
m*m*delta >> 14 expression void).

A value above 64 should not make any sense unless gradfun is used as
a blur filter.

Signed-off-by: Anton Khirnov <anton@khirnov.net>
This commit is contained in:
Clément Bœsch 2012-12-07 00:36:29 +01:00 committed by Anton Khirnov
parent 8b9a153ef3
commit 2d66fc543b
3 changed files with 3 additions and 3 deletions

View File

@ -1142,7 +1142,7 @@ The filter takes two optional parameters, separated by ':':
@var{strength} is the maximum amount by which the filter will change
any one pixel. Also the threshold for detecting nearly flat
regions. Acceptable values range from .51 to 255, default value is
regions. Acceptable values range from .51 to 64, default value is
1.2, out-of-range values will be clipped to the valid range.
@var{radius} is the neighborhood to fit the gradient to. A larger

View File

@ -128,7 +128,7 @@ static av_cold int init(AVFilterContext *ctx, const char *args)
if (args)
sscanf(args, "%f:%d", &thresh, &radius);
thresh = av_clipf(thresh, 0.51, 255);
thresh = av_clipf(thresh, 0.51, 64);
gf->thresh = (1 << 15) / thresh;
gf->radius = av_clip((radius + 1) & ~1, 4, 32);

View File

@ -62,8 +62,8 @@ static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc,
"pminsw %%mm7, %%mm2 \n" // m = -max(0, 127-m)
"pmullw %%mm2, %%mm2 \n"
"paddw %%mm4, %%mm0 \n" // pix += dither
"pmulhw %%mm2, %%mm1 \n"
"psllw $2, %%mm1 \n" // m = m*m*delta >> 14
"pmulhw %%mm2, %%mm1 \n"
"paddw %%mm1, %%mm0 \n" // pix += m
"psraw $7, %%mm0 \n"
"packuswb %%mm0, %%mm0 \n"