various small T1 optimizations

This commit is contained in:
Aaron Boxer 2015-06-15 22:38:39 -04:00 committed by Antonin Descampe
parent 6347686385
commit c9e231508b
1 changed files with 30 additions and 40 deletions

View File

@ -324,16 +324,16 @@ OPJ_BYTE opj_t1_getspb(OPJ_UINT32 f) {
}
OPJ_INT16 opj_t1_getnmsedec_sig(OPJ_UINT32 x, OPJ_UINT32 bitpos) {
if (bitpos > T1_NMSEDEC_FRACBITS) {
return lut_nmsedec_sig[(x >> (bitpos - T1_NMSEDEC_FRACBITS)) & ((1 << T1_NMSEDEC_BITS) - 1)];
if (bitpos > 0) {
return lut_nmsedec_sig[(x >> (bitpos)) & ((1 << T1_NMSEDEC_BITS) - 1)];
}
return lut_nmsedec_sig0[x & ((1 << T1_NMSEDEC_BITS) - 1)];
}
OPJ_INT16 opj_t1_getnmsedec_ref(OPJ_UINT32 x, OPJ_UINT32 bitpos) {
if (bitpos > T1_NMSEDEC_FRACBITS) {
return lut_nmsedec_ref[(x >> (bitpos - T1_NMSEDEC_FRACBITS)) & ((1 << T1_NMSEDEC_BITS) - 1)];
if (bitpos > 0) {
return lut_nmsedec_ref[(x >> (bitpos)) & ((1 << T1_NMSEDEC_BITS) - 1)];
}
return lut_nmsedec_ref0[x & ((1 << T1_NMSEDEC_BITS) - 1)];
@ -390,7 +390,7 @@ void opj_t1_enc_sigpass_step( opj_t1_t *t1,
}
if (v) {
v = *datap < 0 ? 1 : 0;
*nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno + T1_NMSEDEC_FRACBITS));
*nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno));
opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc(flag)); /* ESSAI */
if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */
opj_mqc_bypass_enc(mqc, (OPJ_UINT32)v);
@ -624,7 +624,7 @@ void opj_t1_enc_refpass_step( opj_t1_t *t1,
flag = vsc ? (OPJ_UINT32)((*flagsp) & (~(T1_SIG_S | T1_SIG_SE | T1_SIG_SW | T1_SGN_S))) : (OPJ_UINT32)(*flagsp);
if ((flag & (T1_SIG | T1_VISIT)) == T1_SIG) {
*nmsedec += opj_t1_getnmsedec_ref((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno + T1_NMSEDEC_FRACBITS));
*nmsedec += opj_t1_getnmsedec_ref((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno));
v = opj_int_abs(*datap) & one ? 1 : 0;
opj_mqc_setcurctx(mqc, opj_t1_getctxno_mag(flag)); /* ESSAI */
if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */
@ -853,7 +853,7 @@ void opj_t1_enc_clnpass_step(
opj_mqc_encode(mqc, (OPJ_UINT32)v);
if (v) {
LABEL_PARTIAL:
*nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno + T1_NMSEDEC_FRACBITS));
*nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), (OPJ_UINT32)(bpno));
opj_mqc_setcurctx(mqc, opj_t1_getctxno_sc(flag));
v = *datap < 0 ? 1 : 0;
opj_mqc_encode(mqc, (OPJ_UINT32)(v ^ opj_t1_getspb((OPJ_UINT32)flag)));
@ -962,10 +962,10 @@ void opj_t1_enc_clnpass(
|| (MACRO_t1_flags(1 + k + 3,1 + i)
& (~(T1_SIG_S | T1_SIG_SE | T1_SIG_SW | T1_SGN_S))) & (T1_SIG | T1_VISIT | T1_SIG_OTH));
} else {
agg = !(MACRO_t1_flags(1 + k,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 1,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 2,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 3,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH));
agg = !((MACRO_t1_flags(1 + k,1 + i) |
MACRO_t1_flags(1 + k + 1,1 + i) |
MACRO_t1_flags(1 + k + 2,1 + i) |
MACRO_t1_flags(1 + k + 3,1 + i)) & (T1_SIG | T1_VISIT | T1_SIG_OTH));
}
} else {
agg = 0;
@ -1061,10 +1061,10 @@ static void opj_t1_dec_clnpass(
for (i = 0; i < t1->w; ++i) {
OPJ_INT32 *data2 = data1 + i;
opj_flag_t *flags2 = flags1 + i;
agg = !(MACRO_t1_flags(1 + k,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 1,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 2,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH)
|| MACRO_t1_flags(1 + k + 3,1 + i) & (T1_SIG | T1_VISIT | T1_SIG_OTH));
agg = !((MACRO_t1_flags(1 + k, 1 + i) |
MACRO_t1_flags(1 + k + 1, 1 + i) |
MACRO_t1_flags(1 + k + 2, 1 + i) |
MACRO_t1_flags(1 + k + 3, 1 + i)) & (T1_SIG | T1_VISIT | T1_SIG_OTH));
if (agg) {
opj_mqc_setcurctx(mqc, T1_CTXNO_AGG);
if (!opj_mqc_decode(mqc)) {
@ -1075,9 +1075,9 @@ static void opj_t1_dec_clnpass(
runlen = (runlen << 1) | opj_mqc_decode(mqc);
flags2 += (OPJ_UINT32)runlen * t1->flags_stride;
data2 += (OPJ_UINT32)runlen * t1->w;
for (j = k + (OPJ_UINT32)runlen; j < k + 4 && j < t1->h; ++j) {
for (j = (OPJ_UINT32)runlen; j < 4 && j < t1->h; ++j) {
flags2 += t1->flags_stride;
if (agg && (j == k + (OPJ_UINT32)runlen)) {
if (agg && (j == (OPJ_UINT32)runlen)) {
opj_t1_dec_clnpass_step_partial(t1, flags2, data2, orient, oneplushalf);
} else {
opj_t1_dec_clnpass_step(t1, flags2, data2, orient, oneplushalf);
@ -1288,7 +1288,6 @@ OPJ_BOOL opj_t1_decode_cblks( opj_t1_t* t1,
for (cblkno = 0; cblkno < precinct->cw * precinct->ch; ++cblkno) {
opj_tcd_cblk_dec_t* cblk = &precinct->cblks.dec[cblkno];
OPJ_INT32* restrict datap;
/*void* restrict tiledp;*/
OPJ_UINT32 cblk_w, cblk_h;
OPJ_INT32 x, y;
OPJ_UINT32 i, j;
@ -1330,14 +1329,12 @@ OPJ_BOOL opj_t1_decode_cblks( opj_t1_t* t1,
}
}
}
/*tiledp=(void*)&tilec->data[(y * tile_w) + x];*/
if (tccp->qmfbid == 1) {
OPJ_INT32* restrict tiledp = &tilec->data[(OPJ_UINT32)y * tile_w + (OPJ_UINT32)x];
for (j = 0; j < cblk_h; ++j) {
for (i = 0; i < cblk_w; ++i) {
OPJ_INT32 tmp = datap[(j * cblk_w) + i];
((OPJ_INT32*)tiledp)[(j * tile_w) + i] = tmp / 2;
((OPJ_INT32*)tiledp)[(j * tile_w) + i] = tmp/2;
}
}
} else { /* if (tccp->qmfbid == 0) */
@ -1349,18 +1346,11 @@ OPJ_BOOL opj_t1_decode_cblks( opj_t1_t* t1,
*tiledp2 = tmp;
datap++;
tiledp2++;
/*float tmp = datap[(j * cblk_w) + i] * band->stepsize;
((float*)tiledp)[(j * tile_w) + i] = tmp;*/
}
tiledp += tile_w;
}
}
/*opj_free(cblk->data);
opj_free(cblk->segs);*/
/*cblk->segs = 00;*/
} /* cblkno */
/*opj_free(precinct->cblks.dec);*/
} /* precno */
} /* bandno */
} /* resno */
@ -1377,7 +1367,7 @@ OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1,
opj_raw_t *raw = t1->raw; /* RAW component */
opj_mqc_t *mqc = t1->mqc; /* MQC component */
OPJ_INT32 bpno;
OPJ_INT32 bpno_plus_one;
OPJ_UINT32 passtype;
OPJ_UINT32 segno, passno;
OPJ_BYTE type = T1_TYPE_MQ; /* BYPASS mode */
@ -1390,7 +1380,7 @@ OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1,
return OPJ_FALSE;
}
bpno = (OPJ_INT32)(roishift + cblk->numbps - 1);
bpno_plus_one = (OPJ_INT32)(roishift + cblk->numbps);
passtype = 2;
opj_mqc_resetstates(mqc);
@ -1402,7 +1392,7 @@ OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1,
opj_tcd_seg_t *seg = &cblk->segs[segno];
/* BYPASS mode */
type = ((bpno <= ((OPJ_INT32) (cblk->numbps) - 1) - 4) && (passtype < 2) && (cblksty & J2K_CCP_CBLKSTY_LAZY)) ? T1_TYPE_RAW : T1_TYPE_MQ;
type = ((bpno_plus_one <= ((OPJ_INT32) (cblk->numbps)) - 4) && (passtype < 2) && (cblksty & J2K_CCP_CBLKSTY_LAZY)) ? T1_TYPE_RAW : T1_TYPE_MQ;
/* FIXME: slviewer gets here with a null pointer. Why? Partially downloaded and/or corrupt textures? */
if(seg->data == 00){
continue;
@ -1419,28 +1409,28 @@ OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1,
switch (passtype) {
case 0:
if (type == T1_TYPE_RAW) {
opj_t1_dec_sigpass_raw(t1, bpno+1, (OPJ_INT32)orient, (OPJ_INT32)cblksty);
opj_t1_dec_sigpass_raw(t1, bpno_plus_one, (OPJ_INT32)orient, (OPJ_INT32)cblksty);
} else {
if (cblksty & J2K_CCP_CBLKSTY_VSC) {
opj_t1_dec_sigpass_mqc_vsc(t1, bpno+1, (OPJ_INT32)orient);
opj_t1_dec_sigpass_mqc_vsc(t1, bpno_plus_one, (OPJ_INT32)orient);
} else {
opj_t1_dec_sigpass_mqc(t1, bpno+1, (OPJ_INT32)orient);
opj_t1_dec_sigpass_mqc(t1, bpno_plus_one, (OPJ_INT32)orient);
}
}
break;
case 1:
if (type == T1_TYPE_RAW) {
opj_t1_dec_refpass_raw(t1, bpno+1, (OPJ_INT32)cblksty);
opj_t1_dec_refpass_raw(t1, bpno_plus_one, (OPJ_INT32)cblksty);
} else {
if (cblksty & J2K_CCP_CBLKSTY_VSC) {
opj_t1_dec_refpass_mqc_vsc(t1, bpno+1);
opj_t1_dec_refpass_mqc_vsc(t1, bpno_plus_one);
} else {
opj_t1_dec_refpass_mqc(t1, bpno+1);
opj_t1_dec_refpass_mqc(t1, bpno_plus_one);
}
}
break;
case 2:
opj_t1_dec_clnpass(t1, bpno+1, (OPJ_INT32)orient, (OPJ_INT32)cblksty);
opj_t1_dec_clnpass(t1, bpno_plus_one, (OPJ_INT32)orient, (OPJ_INT32)cblksty);
break;
}
@ -1452,11 +1442,11 @@ OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1,
}
if (++passtype == 3) {
passtype = 0;
bpno--;
bpno_plus_one--;
}
}
}
return OPJ_TRUE;
return OPJ_TRUE;
}