walkero
/
openjpeg
1
0
Fork 0
Code Issues Pull Requests Projects Releases 1 Wiki Activity

T1: (failed) optimization attempt for low bit planes in refpass (#932) 

In the lowest bit plane in 6 to 8% of the cases, the 4 samples in a column
in the zig-zag pattern are in context 16 of the refpass. So let's add a
dedicated function for this.

However the (best) timings show that it is actually slower. With this special
case: 50205 ms vs 49570 ms originally on MAPA_005.jp2 reencoded with default options.
This commit is contained in:
Even Rouault 2017-05-22 13:27:27 +02:00
parent 83d7a6d4a4
commit dcf013db28
1 changed files with 42 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
src/lib/openjp2/t1.c
View File

@ -817,6 +817,26 @@ static INLINE void opj_t1_dec_refpass_step_mqc(
}
} /* VSC and BYPASS by Antonin */
static INLINE void opj_t1_dec_refpass_step_mqc_4_at_context16(
opj_t1_t *t1,
OPJ_INT32 *datap,
OPJ_INT32 poshalf,
OPJ_INT32 neghalf,
OPJ_UINT32 w)
{
OPJ_INT32 i, v, t;
opj_mqc_t *mqc = t1->mqc; /* MQC component */
opj_mqc_setcurctx(mqc, T1_CTXNO_MAG + 2);
for (i = 0; i < 4; i++) {
v = opj_mqc_decode(mqc);
t = v ? poshalf : neghalf;
*datap += *datap < 0 ? -t : t;
datap += w;
}
}
static INLINE void opj_t1_dec_refpass_step_mqc_vsc(
opj_t1_t *t1,
opj_flag_t *flagsp,
@ -922,7 +942,28 @@ static void opj_t1_dec_refpass_raw(
OPJ_INT32 *data2 = data1 + i; \
opj_flag_t *flags2 = flags1 + i; \
opj_colflag_t *colflags2 = colflags1 + i; \
if( *colflags2 == 0 ) continue; \
if ((*colflags2 & (T1_COLFLAG_SIG_ROW_0 | T1_COLFLAG_VISIT_ROW_0 | \
T1_COLFLAG_SIG_ROW_1 | T1_COLFLAG_VISIT_ROW_1 | \
T1_COLFLAG_SIG_ROW_2 | T1_COLFLAG_VISIT_ROW_2 | \
T1_COLFLAG_SIG_ROW_3 | T1_COLFLAG_VISIT_ROW_3 )) == 0) \
{ \
/* No sample in this column is eligible to refpass */ \
continue; \
} \
if ((*colflags2 & (T1_COLFLAG_SIG_ROW_0 | T1_COLFLAG_VISIT_ROW_0 | T1_COLFLAG_REFINE_ROW_0 | \
T1_COLFLAG_SIG_ROW_1 | T1_COLFLAG_VISIT_ROW_1 | T1_COLFLAG_REFINE_ROW_1 | \
T1_COLFLAG_SIG_ROW_2 | T1_COLFLAG_VISIT_ROW_2 | T1_COLFLAG_REFINE_ROW_2 | \
T1_COLFLAG_SIG_ROW_3 | T1_COLFLAG_VISIT_ROW_3 | T1_COLFLAG_REFINE_ROW_3 )) \
== (T1_COLFLAG_SIG_ROW_0 | T1_COLFLAG_REFINE_ROW_0 | \
T1_COLFLAG_SIG_ROW_1 | T1_COLFLAG_REFINE_ROW_1 | \
T1_COLFLAG_SIG_ROW_2 | T1_COLFLAG_REFINE_ROW_2 | \
T1_COLFLAG_SIG_ROW_3 | T1_COLFLAG_REFINE_ROW_3 )) \
{ \
/* In the lowest bit planes this can be 6-8% of cases. */ \
/* See https://github.com/uclouvain/openjpeg/issues/932 */ \
opj_t1_dec_refpass_step_mqc_4_at_context16(t1, data2, poshalf, neghalf, w); \
continue; \
} \
flags2 += flags_stride; \
opj_t1_dec_refpass_step_mqc(t1, flags2, colflags2, data2, poshalf, neghalf, 0U); \
data2 += w; \