X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;ds=sidebyside;f=src%2Flib%2Fopenjp2%2Ft1.c;h=52e466eb974afd1fc89f0ed180b95973ee8bc532;hb=c06632c6f6d0a3e106374fe462869b131366c2d0;hp=3615a0e6690c6cfd3dd927459b3fa5bb7d4e10c2;hpb=f6551f822fe020843299bd807ec6989abd070b2c;p=openjpeg.git diff --git a/src/lib/openjp2/t1.c b/src/lib/openjp2/t1.c index 3615a0e6..52e466eb 100644 --- a/src/lib/openjp2/t1.c +++ b/src/lib/openjp2/t1.c @@ -38,7 +38,20 @@ * POSSIBILITY OF SUCH DAMAGE. */ +#define OPJ_SKIP_POISON #include "opj_includes.h" + +#ifdef __SSE__ +#include +#endif +#ifdef __SSE2__ +#include +#endif + +#if defined(__GNUC__) +#pragma GCC poison malloc calloc realloc free +#endif + #include "t1_luts.h" /** @defgroup T1 T1 - Implementation of the tier-1 coding */ @@ -48,6 +61,13 @@ #define opj_t1_setcurctx(curctx, ctxno) curctx = &(mqc)->ctxs[(OPJ_UINT32)(ctxno)] +/* Macros to deal with signed integer with just MSB bit set for + * negative values (smr = signed magnitude representation) */ +#define opj_smr_abs(x) (((OPJ_UINT32)(x)) & 0x7FFFFFFFU) +#define opj_smr_sign(x) (((OPJ_UINT32)(x)) >> 31) +#define opj_to_smr(x) ((x) >= 0 ? (OPJ_UINT32)(x) : ((OPJ_UINT32)(-x) | 0x80000000U)) + + /** @name Local static functions */ /*@{*/ @@ -164,18 +184,18 @@ static OPJ_FLOAT64 opj_t1_getwmsedec( const OPJ_FLOAT64 * mct_norms, OPJ_UINT32 mct_numcomps); -static void opj_t1_encode_cblk(opj_t1_t *t1, - opj_tcd_cblk_enc_t* cblk, - OPJ_UINT32 orient, - OPJ_UINT32 compno, - OPJ_UINT32 level, - OPJ_UINT32 qmfbid, - OPJ_FLOAT64 stepsize, - OPJ_UINT32 cblksty, - OPJ_UINT32 numcomps, - opj_tcd_tile_t * tile, - const OPJ_FLOAT64 * mct_norms, - OPJ_UINT32 mct_numcomps); +/** Return "cumwmsedec" that should be used to increase tile->distotile */ +static double opj_t1_encode_cblk(opj_t1_t *t1, + opj_tcd_cblk_enc_t* cblk, + OPJ_UINT32 orient, + OPJ_UINT32 compno, + OPJ_UINT32 level, + OPJ_UINT32 qmfbid, + OPJ_FLOAT64 stepsize, + OPJ_UINT32 cblksty, + OPJ_UINT32 numcomps, + const OPJ_FLOAT64 * mct_norms, + OPJ_UINT32 mct_numcomps); /** Decode 1 code-block @@ -184,6 +204,9 @@ Decode 1 code-block @param orient @param roishift Region of interest shifting value @param cblksty Code-block style +@param p_manager the event manager +@param p_manager_mutex mutex for the event manager +@param check_pterm whether PTERM correct termination should be checked */ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, opj_tcd_cblk_dec_t* cblk, @@ -194,6 +217,27 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, opj_mutex_t* p_manager_mutex, OPJ_BOOL check_pterm); +/** +Decode 1 HT code-block +@param t1 T1 handle +@param cblk Code-block coding parameters +@param orient +@param roishift Region of interest shifting value +@param cblksty Code-block style +@param p_manager the event manager +@param p_manager_mutex mutex for the event manager +@param check_pterm whether PTERM correct termination should be checked +*/ +OPJ_BOOL opj_t1_ht_decode_cblk(opj_t1_t *t1, + opj_tcd_cblk_dec_t* cblk, + OPJ_UINT32 orient, + OPJ_UINT32 roishift, + OPJ_UINT32 cblksty, + opj_event_mgr_t *p_manager, + opj_mutex_t* p_manager_mutex, + OPJ_BOOL check_pterm); + + static OPJ_BOOL opj_t1_allocate_buffers(opj_t1_t *t1, OPJ_UINT32 w, OPJ_UINT32 h); @@ -313,61 +357,53 @@ static INLINE void opj_t1_update_flags(opj_flag_t *flagsp, OPJ_UINT32 ci, /** Encode significant pass */ -static INLINE void opj_t1_enc_sigpass_step(opj_t1_t *t1, - opj_flag_t *flagsp, - OPJ_INT32 *datap, - OPJ_INT32 bpno, - OPJ_INT32 one, - OPJ_INT32 *nmsedec, - OPJ_BYTE type, - OPJ_UINT32 ci, - OPJ_UINT32 vsc) -{ - OPJ_UINT32 v; - - opj_mqc_t *mqc = &(t1->mqc); /* MQC component */ - - OPJ_UINT32 const flags = *flagsp; - - if ((flags & ((T1_SIGMA_THIS | T1_PI_THIS) << (ci * 3U))) == 0U && - (flags & (T1_SIGMA_NEIGHBOURS << (ci * 3U))) != 0U) { - OPJ_UINT32 ctxt1 = opj_t1_getctxno_zc(mqc, flags >> (ci * 3U)); - v = opj_int_abs(*datap) & one ? 1 : 0; -#ifdef DEBUG_ENC_SIG - fprintf(stderr, " ctxt1=%d\n", ctxt1); -#endif - opj_mqc_setcurctx(mqc, ctxt1); - if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ - opj_mqc_bypass_enc(mqc, v); - } else { - opj_mqc_encode(mqc, v); - } - if (v) { - OPJ_UINT32 lu = opj_t1_getctxtno_sc_or_spb_index( - *flagsp, - flagsp[-1], flagsp[1], - ci); - OPJ_UINT32 ctxt2 = opj_t1_getctxno_sc(lu); - v = *datap < 0 ? 1U : 0U; - *nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), - (OPJ_UINT32)bpno); -#ifdef DEBUG_ENC_SIG - fprintf(stderr, " ctxt2=%d\n", ctxt2); -#endif - opj_mqc_setcurctx(mqc, ctxt2); - if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ - opj_mqc_bypass_enc(mqc, v); - } else { - OPJ_UINT32 spb = opj_t1_getspb(lu); -#ifdef DEBUG_ENC_SIG - fprintf(stderr, " spb=%d\n", spb); -#endif - opj_mqc_encode(mqc, v ^ spb); - } - opj_t1_update_flags(flagsp, ci, v, t1->w + 2, vsc); - } - *flagsp |= T1_PI_THIS << (ci * 3U); - } +#define opj_t1_enc_sigpass_step_macro(mqc, curctx, a, c, ct, flagspIn, datapIn, bpno, one, nmsedec, type, ciIn, vscIn) \ +{ \ + OPJ_UINT32 v; \ + const OPJ_UINT32 ci = (ciIn); \ + const OPJ_UINT32 vsc = (vscIn); \ + const OPJ_INT32* l_datap = (datapIn); \ + opj_flag_t* flagsp = (flagspIn); \ + OPJ_UINT32 const flags = *flagsp; \ + if ((flags & ((T1_SIGMA_THIS | T1_PI_THIS) << (ci * 3U))) == 0U && \ + (flags & (T1_SIGMA_NEIGHBOURS << (ci * 3U))) != 0U) { \ + OPJ_UINT32 ctxt1 = opj_t1_getctxno_zc(mqc, flags >> (ci * 3U)); \ + v = (opj_smr_abs(*l_datap) & (OPJ_UINT32)one) ? 1 : 0; \ +/* #ifdef DEBUG_ENC_SIG */ \ +/* fprintf(stderr, " ctxt1=%d\n", ctxt1); */ \ +/* #endif */ \ + opj_t1_setcurctx(curctx, ctxt1); \ + if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ \ + opj_mqc_bypass_enc_macro(mqc, c, ct, v); \ + } else { \ + opj_mqc_encode_macro(mqc, curctx, a, c, ct, v); \ + } \ + if (v) { \ + OPJ_UINT32 lu = opj_t1_getctxtno_sc_or_spb_index( \ + *flagsp, \ + flagsp[-1], flagsp[1], \ + ci); \ + OPJ_UINT32 ctxt2 = opj_t1_getctxno_sc(lu); \ + v = opj_smr_sign(*l_datap); \ + *nmsedec += opj_t1_getnmsedec_sig(opj_smr_abs(*l_datap), \ + (OPJ_UINT32)bpno); \ +/* #ifdef DEBUG_ENC_SIG */ \ +/* fprintf(stderr, " ctxt2=%d\n", ctxt2); */ \ +/* #endif */ \ + opj_t1_setcurctx(curctx, ctxt2); \ + if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ \ + opj_mqc_bypass_enc_macro(mqc, c, ct, v); \ + } else { \ + OPJ_UINT32 spb = opj_t1_getspb(lu); \ +/* #ifdef DEBUG_ENC_SIG */ \ +/* fprintf(stderr, " spb=%d\n", spb); */ \ +/* #endif */ \ + opj_mqc_encode_macro(mqc, curctx, a, c, ct, v ^ spb); \ + } \ + opj_t1_update_flags(flagsp, ci, v, t1->w + 2, vsc); \ + } \ + *flagsp |= T1_PI_THIS << (ci * 3U); \ + } \ } static INLINE void opj_t1_dec_sigpass_step_raw( @@ -448,63 +484,64 @@ static void opj_t1_enc_sigpass(opj_t1_t *t1, OPJ_INT32 const one = 1 << (bpno + T1_NMSEDEC_FRACBITS); opj_flag_t* f = &T1_FLAGS(0, 0); OPJ_UINT32 const extra = 2; + opj_mqc_t* mqc = &(t1->mqc); + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); + const OPJ_INT32* datap = t1->data; *nmsedec = 0; #ifdef DEBUG_ENC_SIG fprintf(stderr, "enc_sigpass: bpno=%d\n", bpno); #endif - for (k = 0; k < (t1->h & ~3U); k += 4) { + for (k = 0; k < (t1->h & ~3U); k += 4, f += extra) { + const OPJ_UINT32 w = t1->w; #ifdef DEBUG_ENC_SIG fprintf(stderr, " k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < w; ++i, ++f, datap += 4) { #ifdef DEBUG_ENC_SIG fprintf(stderr, " i=%d\n", i); #endif if (*f == 0U) { /* Nothing to do for any of the 4 data points */ - f++; continue; } - opj_t1_enc_sigpass_step( - t1, + opj_t1_enc_sigpass_step_macro( + mqc, curctx, a, c, ct, f, - &t1->data[((k + 0) * t1->data_stride) + i], + &datap[0], bpno, one, nmsedec, type, 0, cblksty & J2K_CCP_CBLKSTY_VSC); - opj_t1_enc_sigpass_step( - t1, + opj_t1_enc_sigpass_step_macro( + mqc, curctx, a, c, ct, f, - &t1->data[((k + 1) * t1->data_stride) + i], + &datap[1], bpno, one, nmsedec, type, 1, 0); - opj_t1_enc_sigpass_step( - t1, + opj_t1_enc_sigpass_step_macro( + mqc, curctx, a, c, ct, f, - &t1->data[((k + 2) * t1->data_stride) + i], + &datap[2], bpno, one, nmsedec, type, 2, 0); - opj_t1_enc_sigpass_step( - t1, + opj_t1_enc_sigpass_step_macro( + mqc, curctx, a, c, ct, f, - &t1->data[((k + 3) * t1->data_stride) + i], + &datap[3], bpno, one, nmsedec, type, 3, 0); - ++f; } - f += extra; } if (k < t1->h) { @@ -512,20 +549,20 @@ static void opj_t1_enc_sigpass(opj_t1_t *t1, #ifdef DEBUG_ENC_SIG fprintf(stderr, " k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < t1->w; ++i, ++f) { #ifdef DEBUG_ENC_SIG fprintf(stderr, " i=%d\n", i); #endif if (*f == 0U) { /* Nothing to do for any of the 4 data points */ - f++; + datap += (t1->h - k); continue; } - for (j = k; j < t1->h; ++j) { - opj_t1_enc_sigpass_step( - t1, + for (j = k; j < t1->h; ++j, ++datap) { + opj_t1_enc_sigpass_step_macro( + mqc, curctx, a, c, ct, f, - &t1->data[(j * t1->data_stride) + i], + &datap[0], bpno, one, nmsedec, @@ -533,9 +570,10 @@ static void opj_t1_enc_sigpass(opj_t1_t *t1, j - k, (j == k && (cblksty & J2K_CCP_CBLKSTY_VSC) != 0)); } - ++f; } } + + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); } static void opj_t1_dec_sigpass_raw( @@ -610,7 +648,7 @@ static void opj_t1_dec_sigpass_raw( register opj_flag_t *flagsp = &t1->flags[(flags_stride) + 1]; \ const OPJ_UINT32 l_w = w; \ opj_mqc_t* mqc = &(t1->mqc); \ - DOWNLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ register OPJ_UINT32 v; \ one = 1 << bpno; \ half = one >> 1; \ @@ -635,7 +673,7 @@ static void opj_t1_dec_sigpass_raw( } \ } \ } \ - UPLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ if( k < h ) { \ for (i = 0; i < l_w; ++i, ++data, ++flagsp) { \ for (j = 0; j < h - k; ++j) { \ @@ -699,38 +737,27 @@ static void opj_t1_dec_sigpass_mqc( /** Encode refinement pass step */ -static INLINE void opj_t1_enc_refpass_step(opj_t1_t *t1, - opj_flag_t *flagsp, - OPJ_INT32 *datap, - OPJ_INT32 bpno, - OPJ_INT32 one, - OPJ_INT32 *nmsedec, - OPJ_BYTE type, - OPJ_UINT32 ci) -{ - OPJ_UINT32 v; - - opj_mqc_t *mqc = &(t1->mqc); /* MQC component */ - - OPJ_UINT32 const shift_flags = - (*flagsp >> (ci * 3U)); - - if ((shift_flags & (T1_SIGMA_THIS | T1_PI_THIS)) == T1_SIGMA_THIS) { - OPJ_UINT32 ctxt = opj_t1_getctxno_mag(shift_flags); - *nmsedec += opj_t1_getnmsedec_ref((OPJ_UINT32)opj_int_abs(*datap), - (OPJ_UINT32)bpno); - v = opj_int_abs(*datap) & one ? 1 : 0; -#ifdef DEBUG_ENC_REF - fprintf(stderr, " ctxt=%d\n", ctxt); -#endif - opj_mqc_setcurctx(mqc, ctxt); - if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ - opj_mqc_bypass_enc(mqc, v); - } else { - opj_mqc_encode(mqc, v); - } - *flagsp |= T1_MU_THIS << (ci * 3U); - } +#define opj_t1_enc_refpass_step_macro(mqc, curctx, a, c, ct, flags, flagsUpdated, datap, bpno, one, nmsedec, type, ci) \ +{\ + OPJ_UINT32 v; \ + if ((flags & ((T1_SIGMA_THIS | T1_PI_THIS) << ((ci) * 3U))) == (T1_SIGMA_THIS << ((ci) * 3U))) { \ + const OPJ_UINT32 shift_flags = (flags >> ((ci) * 3U)); \ + OPJ_UINT32 ctxt = opj_t1_getctxno_mag(shift_flags); \ + OPJ_UINT32 abs_data = opj_smr_abs(*datap); \ + *nmsedec += opj_t1_getnmsedec_ref(abs_data, \ + (OPJ_UINT32)bpno); \ + v = ((OPJ_INT32)abs_data & one) ? 1 : 0; \ +/* #ifdef DEBUG_ENC_REF */ \ +/* fprintf(stderr, " ctxt=%d\n", ctxt); */ \ +/* #endif */ \ + opj_t1_setcurctx(curctx, ctxt); \ + if (type == T1_TYPE_RAW) { /* BYPASS/LAZY MODE */ \ + opj_mqc_bypass_enc_macro(mqc, c, ct, v); \ + } else { \ + opj_mqc_encode_macro(mqc, curctx, a, c, ct, v); \ + } \ + flagsUpdated |= T1_MU_THIS << ((ci) * 3U); \ + } \ } @@ -791,100 +818,104 @@ static void opj_t1_enc_refpass( const OPJ_INT32 one = 1 << (bpno + T1_NMSEDEC_FRACBITS); opj_flag_t* f = &T1_FLAGS(0, 0); const OPJ_UINT32 extra = 2U; + opj_mqc_t* mqc = &(t1->mqc); + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); + const OPJ_INT32* datap = t1->data; *nmsedec = 0; #ifdef DEBUG_ENC_REF fprintf(stderr, "enc_refpass: bpno=%d\n", bpno); #endif - for (k = 0; k < (t1->h & ~3U); k += 4) { + for (k = 0; k < (t1->h & ~3U); k += 4, f += extra) { #ifdef DEBUG_ENC_REF fprintf(stderr, " k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < t1->w; ++i, f++, datap += 4) { + const OPJ_UINT32 flags = *f; + OPJ_UINT32 flagsUpdated = flags; #ifdef DEBUG_ENC_REF fprintf(stderr, " i=%d\n", i); #endif - if ((*f & (T1_SIGMA_4 | T1_SIGMA_7 | T1_SIGMA_10 | T1_SIGMA_13)) == 0) { + if ((flags & (T1_SIGMA_4 | T1_SIGMA_7 | T1_SIGMA_10 | T1_SIGMA_13)) == 0) { /* none significant */ - f++; continue; } - if ((*f & (T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3)) == + if ((flags & (T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3)) == (T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3)) { /* all processed by sigpass */ - f++; continue; } - opj_t1_enc_refpass_step( - t1, - f, - &t1->data[((k + 0) * t1->data_stride) + i], + opj_t1_enc_refpass_step_macro( + mqc, curctx, a, c, ct, + flags, flagsUpdated, + &datap[0], bpno, one, nmsedec, type, 0); - opj_t1_enc_refpass_step( - t1, - f, - &t1->data[((k + 1) * t1->data_stride) + i], + opj_t1_enc_refpass_step_macro( + mqc, curctx, a, c, ct, + flags, flagsUpdated, + &datap[1], bpno, one, nmsedec, type, 1); - opj_t1_enc_refpass_step( - t1, - f, - &t1->data[((k + 2) * t1->data_stride) + i], + opj_t1_enc_refpass_step_macro( + mqc, curctx, a, c, ct, + flags, flagsUpdated, + &datap[2], bpno, one, nmsedec, type, 2); - opj_t1_enc_refpass_step( - t1, - f, - &t1->data[((k + 3) * t1->data_stride) + i], + opj_t1_enc_refpass_step_macro( + mqc, curctx, a, c, ct, + flags, flagsUpdated, + &datap[3], bpno, one, nmsedec, type, 3); - ++f; + *f = flagsUpdated; } - f += extra; } if (k < t1->h) { OPJ_UINT32 j; + const OPJ_UINT32 remaining_lines = t1->h - k; #ifdef DEBUG_ENC_REF fprintf(stderr, " k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < t1->w; ++i, ++f) { #ifdef DEBUG_ENC_REF fprintf(stderr, " i=%d\n", i); #endif if ((*f & (T1_SIGMA_4 | T1_SIGMA_7 | T1_SIGMA_10 | T1_SIGMA_13)) == 0) { /* none significant */ - f++; + datap += remaining_lines; continue; } - for (j = k; j < t1->h; ++j) { - opj_t1_enc_refpass_step( - t1, - f, - &t1->data[(j * t1->data_stride) + i], + for (j = 0; j < remaining_lines; ++j, datap ++) { + opj_t1_enc_refpass_step_macro( + mqc, curctx, a, c, ct, + *f, *f, + &datap[0], bpno, one, nmsedec, type, - j - k); + j); } - ++f; } } + + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); } @@ -952,7 +983,7 @@ static void opj_t1_dec_refpass_raw( register opj_flag_t *flagsp = &t1->flags[flags_stride + 1]; \ const OPJ_UINT32 l_w = w; \ opj_mqc_t* mqc = &(t1->mqc); \ - DOWNLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ register OPJ_UINT32 v; \ one = 1 << bpno; \ poshalf = one >> 1; \ @@ -976,7 +1007,7 @@ static void opj_t1_dec_refpass_raw( } \ } \ } \ - UPLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ if( k < h ) { \ for (i = 0; i < l_w; ++i, ++data, ++flagsp) { \ for (j = 0; j < h - k; ++j) { \ @@ -1014,85 +1045,71 @@ static void opj_t1_dec_refpass_mqc( /** Encode clean-up pass step */ -static void opj_t1_enc_clnpass_step( - opj_t1_t *t1, - opj_flag_t *flagsp, - OPJ_INT32 *datap, - OPJ_INT32 bpno, - OPJ_INT32 one, - OPJ_INT32 *nmsedec, - OPJ_UINT32 agg, - OPJ_UINT32 runlen, - OPJ_UINT32 lim, - OPJ_UINT32 cblksty) -{ - OPJ_UINT32 v; - OPJ_UINT32 ci; - opj_mqc_t *mqc = &(t1->mqc); /* MQC component */ - - const OPJ_UINT32 check = (T1_SIGMA_4 | T1_SIGMA_7 | T1_SIGMA_10 | T1_SIGMA_13 | - T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3); - - if ((*flagsp & check) == check) { - if (runlen == 0) { - *flagsp &= ~(T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3); - } else if (runlen == 1) { - *flagsp &= ~(T1_PI_1 | T1_PI_2 | T1_PI_3); - } else if (runlen == 2) { - *flagsp &= ~(T1_PI_2 | T1_PI_3); - } else if (runlen == 3) { - *flagsp &= ~(T1_PI_3); - } - return; - } - - for (ci = runlen; ci < lim; ++ci) { - OPJ_UINT32 vsc; - opj_flag_t flags; - - flags = *flagsp; - - if ((agg != 0) && (ci == runlen)) { - goto LABEL_PARTIAL; - } - - if (!(flags & ((T1_SIGMA_THIS | T1_PI_THIS) << (ci * 3U)))) { - OPJ_UINT32 ctxt1 = opj_t1_getctxno_zc(mqc, flags >> (ci * 3U)); -#ifdef DEBUG_ENC_CLN - printf(" ctxt1=%d\n", ctxt1); -#endif - opj_mqc_setcurctx(mqc, ctxt1); - v = opj_int_abs(*datap) & one ? 1 : 0; - opj_mqc_encode(mqc, v); - if (v) { - OPJ_UINT32 ctxt2, spb; - OPJ_UINT32 lu; -LABEL_PARTIAL: - lu = opj_t1_getctxtno_sc_or_spb_index( - *flagsp, - flagsp[-1], flagsp[1], - ci); - *nmsedec += opj_t1_getnmsedec_sig((OPJ_UINT32)opj_int_abs(*datap), - (OPJ_UINT32)bpno); - ctxt2 = opj_t1_getctxno_sc(lu); -#ifdef DEBUG_ENC_CLN - printf(" ctxt2=%d\n", ctxt2); -#endif - opj_mqc_setcurctx(mqc, ctxt2); - - v = *datap < 0 ? 1U : 0U; - spb = opj_t1_getspb(lu); -#ifdef DEBUG_ENC_CLN - printf(" spb=%d\n", spb); -#endif - opj_mqc_encode(mqc, v ^ spb); - vsc = ((cblksty & J2K_CCP_CBLKSTY_VSC) && (ci == 0)) ? 1 : 0; - opj_t1_update_flags(flagsp, ci, v, t1->w + 2U, vsc); - } - } - *flagsp &= ~(T1_PI_THIS << (3U * ci)); - datap += t1->data_stride; - } +#define opj_t1_enc_clnpass_step_macro(mqc, curctx, a, c, ct, flagspIn, datapIn, bpno, one, nmsedec, agg, runlen, lim, cblksty) \ +{ \ + OPJ_UINT32 v; \ + OPJ_UINT32 ci; \ + opj_flag_t* const flagsp = (flagspIn); \ + const OPJ_INT32* l_datap = (datapIn); \ + const OPJ_UINT32 check = (T1_SIGMA_4 | T1_SIGMA_7 | T1_SIGMA_10 | T1_SIGMA_13 | \ + T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3); \ + \ + if ((*flagsp & check) == check) { \ + if (runlen == 0) { \ + *flagsp &= ~(T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3); \ + } else if (runlen == 1) { \ + *flagsp &= ~(T1_PI_1 | T1_PI_2 | T1_PI_3); \ + } else if (runlen == 2) { \ + *flagsp &= ~(T1_PI_2 | T1_PI_3); \ + } else if (runlen == 3) { \ + *flagsp &= ~(T1_PI_3); \ + } \ + } \ + else \ + for (ci = runlen; ci < lim; ++ci) { \ + OPJ_BOOL goto_PARTIAL = OPJ_FALSE; \ + if ((agg != 0) && (ci == runlen)) { \ + goto_PARTIAL = OPJ_TRUE; \ + } \ + else if (!(*flagsp & ((T1_SIGMA_THIS | T1_PI_THIS) << (ci * 3U)))) { \ + OPJ_UINT32 ctxt1 = opj_t1_getctxno_zc(mqc, *flagsp >> (ci * 3U)); \ +/* #ifdef DEBUG_ENC_CLN */ \ +/* printf(" ctxt1=%d\n", ctxt1); */ \ +/* #endif */ \ + opj_t1_setcurctx(curctx, ctxt1); \ + v = (opj_smr_abs(*l_datap) & (OPJ_UINT32)one) ? 1 : 0; \ + opj_mqc_encode_macro(mqc, curctx, a, c, ct, v); \ + if (v) { \ + goto_PARTIAL = OPJ_TRUE; \ + } \ + } \ + if( goto_PARTIAL ) { \ + OPJ_UINT32 vsc; \ + OPJ_UINT32 ctxt2, spb; \ + OPJ_UINT32 lu = opj_t1_getctxtno_sc_or_spb_index( \ + *flagsp, \ + flagsp[-1], flagsp[1], \ + ci); \ + *nmsedec += opj_t1_getnmsedec_sig(opj_smr_abs(*l_datap), \ + (OPJ_UINT32)bpno); \ + ctxt2 = opj_t1_getctxno_sc(lu); \ +/* #ifdef DEBUG_ENC_CLN */ \ +/* printf(" ctxt2=%d\n", ctxt2); */ \ +/* #endif */ \ + opj_t1_setcurctx(curctx, ctxt2); \ + \ + v = opj_smr_sign(*l_datap); \ + spb = opj_t1_getspb(lu); \ +/* #ifdef DEBUG_ENC_CLN */ \ +/* printf(" spb=%d\n", spb); */\ +/* #endif */ \ + opj_mqc_encode_macro(mqc, curctx, a, c, ct, v ^ spb); \ + vsc = ((cblksty & J2K_CCP_CBLKSTY_VSC) && (ci == 0)) ? 1 : 0; \ + opj_t1_update_flags(flagsp, ci, v, t1->w + 2U, vsc); \ + } \ + *flagsp &= ~(T1_PI_THIS << (3U * ci)); \ + l_datap ++; \ + } \ } #define opj_t1_dec_clnpass_step_macro(check_flags, partial, \ @@ -1148,47 +1165,50 @@ static void opj_t1_enc_clnpass( { OPJ_UINT32 i, k; const OPJ_INT32 one = 1 << (bpno + T1_NMSEDEC_FRACBITS); - OPJ_UINT32 agg, runlen; - - opj_mqc_t *mqc = &(t1->mqc); /* MQC component */ + opj_mqc_t* mqc = &(t1->mqc); + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); + const OPJ_INT32* datap = t1->data; + opj_flag_t *f = &T1_FLAGS(0, 0); + const OPJ_UINT32 extra = 2U; *nmsedec = 0; #ifdef DEBUG_ENC_CLN printf("enc_clnpass: bpno=%d\n", bpno); #endif - for (k = 0; k < (t1->h & ~3U); k += 4) { + for (k = 0; k < (t1->h & ~3U); k += 4, f += extra) { #ifdef DEBUG_ENC_CLN printf(" k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < t1->w; ++i, f++) { + OPJ_UINT32 agg, runlen; #ifdef DEBUG_ENC_CLN printf(" i=%d\n", i); #endif - agg = !(T1_FLAGS(i, k)); + agg = !*f; #ifdef DEBUG_ENC_CLN printf(" agg=%d\n", agg); #endif if (agg) { - for (runlen = 0; runlen < 4; ++runlen) { - if (opj_int_abs(t1->data[((k + runlen)*t1->data_stride) + i]) & one) { + for (runlen = 0; runlen < 4; ++runlen, ++datap) { + if (opj_smr_abs(*datap) & (OPJ_UINT32)one) { break; } } - opj_mqc_setcurctx(mqc, T1_CTXNO_AGG); - opj_mqc_encode(mqc, runlen != 4); + opj_t1_setcurctx(curctx, T1_CTXNO_AGG); + opj_mqc_encode_macro(mqc, curctx, a, c, ct, runlen != 4); if (runlen == 4) { continue; } - opj_mqc_setcurctx(mqc, T1_CTXNO_UNI); - opj_mqc_encode(mqc, runlen >> 1); - opj_mqc_encode(mqc, runlen & 1); + opj_t1_setcurctx(curctx, T1_CTXNO_UNI); + opj_mqc_encode_macro(mqc, curctx, a, c, ct, runlen >> 1); + opj_mqc_encode_macro(mqc, curctx, a, c, ct, runlen & 1); } else { runlen = 0; } - opj_t1_enc_clnpass_step( - t1, - &T1_FLAGS(i, k), - &t1->data[((k + runlen) * t1->data_stride) + i], + opj_t1_enc_clnpass_step_macro( + mqc, curctx, a, c, ct, + f, + datap, bpno, one, nmsedec, @@ -1196,23 +1216,24 @@ static void opj_t1_enc_clnpass( runlen, 4U, cblksty); + datap += 4 - runlen; } } if (k < t1->h) { - agg = 0; - runlen = 0; + const OPJ_UINT32 agg = 0; + const OPJ_UINT32 runlen = 0; #ifdef DEBUG_ENC_CLN printf(" k=%d\n", k); #endif - for (i = 0; i < t1->w; ++i) { + for (i = 0; i < t1->w; ++i, f++) { #ifdef DEBUG_ENC_CLN printf(" i=%d\n", i); printf(" agg=%d\n", agg); #endif - opj_t1_enc_clnpass_step( - t1, - &T1_FLAGS(i, k), - &t1->data[((k + runlen) * t1->data_stride) + i], + opj_t1_enc_clnpass_step_macro( + mqc, curctx, a, c, ct, + f, + datap, bpno, one, nmsedec, @@ -1220,8 +1241,11 @@ static void opj_t1_enc_clnpass( runlen, t1->h - k, cblksty); + datap += t1->h - k; } } + + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); } #define opj_t1_dec_clnpass_internal(t1, bpno, vsc, w, h, flags_stride) \ @@ -1233,7 +1257,7 @@ static void opj_t1_enc_clnpass( opj_mqc_t* mqc = &(t1->mqc); \ register OPJ_INT32 *data = t1->data; \ register opj_flag_t *flagsp = &t1->flags[flags_stride + 1]; \ - DOWNLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + DOWNLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ register OPJ_UINT32 v; \ one = 1 << bpno; \ half = one >> 1; \ @@ -1302,7 +1326,7 @@ static void opj_t1_enc_clnpass( *flagsp = flags & ~(T1_PI_0 | T1_PI_1 | T1_PI_2 | T1_PI_3); \ } \ } \ - UPLOAD_MQC_VARIABLES(mqc, curctx, c, a, ct); \ + UPLOAD_MQC_VARIABLES(mqc, curctx, a, c, ct); \ if( k < h ) { \ for (i = 0; i < l_w; ++i, ++flagsp, ++data) { \ for (j = 0; j < h - k; ++j) { \ @@ -1386,7 +1410,6 @@ static void opj_t1_dec_clnpass( } -/** mod fixed_quality */ static OPJ_FLOAT64 opj_t1_getwmsedec( OPJ_INT32 nmsedec, OPJ_UINT32 compno, @@ -1409,7 +1432,11 @@ static OPJ_FLOAT64 opj_t1_getwmsedec( if (qmfbid == 1) { w2 = opj_dwt_getnorm(level, orient); } else { /* if (qmfbid == 0) */ + const OPJ_INT32 log2_gain = (orient == 0) ? 0 : + (orient == 3) ? 2 : 1; w2 = opj_dwt_getnorm_real(level, orient); + /* Not sure this is right. But preserves past behaviour */ + stepsize /= (1 << log2_gain); } wmsedec = w1 * w2 * stepsize * (1 << bpno); @@ -1423,44 +1450,27 @@ static OPJ_BOOL opj_t1_allocate_buffers( OPJ_UINT32 w, OPJ_UINT32 h) { - size_t flagssize; + OPJ_UINT32 flagssize; OPJ_UINT32 flags_stride; - /* encoder uses tile buffer, so no need to allocate */ - if (!t1->encoder) { - size_t datasize; - -#if (SIZE_MAX / 0xFFFFFFFFU) < 0xFFFFFFFFU /* UINT32_MAX */ - /* Overflow check */ - if ((w > 0U) && ((size_t)h > (SIZE_MAX / (size_t)w))) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } -#endif - datasize = (size_t)w * h; + /* No risk of overflow. Prior checks ensure those assert are met */ + /* They are per the specification */ + assert(w <= 1024); + assert(h <= 1024); + assert(w * h <= 4096); - /* Overflow check */ - if (datasize > (SIZE_MAX / sizeof(OPJ_INT32))) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } + /* encoder uses tile buffer, so no need to allocate */ + { + OPJ_UINT32 datasize = w * h; - if (datasize > (size_t)t1->datasize) { + if (datasize > t1->datasize) { opj_aligned_free(t1->data); t1->data = (OPJ_INT32*) opj_aligned_malloc(datasize * sizeof(OPJ_INT32)); if (!t1->data) { /* FIXME event manager error callback */ return OPJ_FALSE; } -#if SIZE_MAX > 0xFFFFFFFFU /* UINT32_MAX */ - /* TODO remove this if t1->datasize type changes to size_t */ - /* Overflow check */ - if (datasize > (size_t)0xFFFFFFFFU /* UINT32_MAX */) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } -#endif - t1->datasize = (OPJ_UINT32)datasize; + t1->datasize = datasize; } /* memset first arg is declared to never be null by gcc */ if (t1->data != NULL) { @@ -1468,40 +1478,18 @@ static OPJ_BOOL opj_t1_allocate_buffers( } } - /* Overflow check */ - if (w > (0xFFFFFFFFU /* UINT32_MAX */ - 2U)) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } flags_stride = w + 2U; /* can't be 0U */ -#if (SIZE_MAX - 3U) < 0xFFFFFFFFU /* UINT32_MAX */ - /* Overflow check */ - if (h > (0xFFFFFFFFU /* UINT32_MAX */ - 3U)) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } -#endif flagssize = (h + 3U) / 4U + 2U; - /* Overflow check */ - if (flagssize > (SIZE_MAX / (size_t)flags_stride)) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } - flagssize *= (size_t)flags_stride; + flagssize *= flags_stride; { - /* BIG FAT XXX */ opj_flag_t* p; OPJ_UINT32 x; OPJ_UINT32 flags_height = (h + 3U) / 4U; - if (flagssize > (size_t)t1->flagssize) { - /* Overflow check */ - if (flagssize > (SIZE_MAX / sizeof(opj_flag_t))) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } + if (flagssize > t1->flagssize) { + opj_aligned_free(t1->flags); t1->flags = (opj_flag_t*) opj_aligned_malloc(flagssize * sizeof( opj_flag_t)); @@ -1509,16 +1497,8 @@ static OPJ_BOOL opj_t1_allocate_buffers( /* FIXME event manager error callback */ return OPJ_FALSE; } -#if SIZE_MAX > 0xFFFFFFFFU /* UINT32_MAX */ - /* TODO remove this if t1->flagssize type changes to size_t */ - /* Overflow check */ - if (flagssize > (size_t)0xFFFFFFFFU /* UINT32_MAX */) { - /* FIXME event manager error callback */ - return OPJ_FALSE; - } -#endif } - t1->flagssize = (OPJ_UINT32)flagssize; + t1->flagssize = flagssize; memset(t1->flags, 0, flagssize * sizeof(opj_flag_t)); @@ -1590,8 +1570,7 @@ void opj_t1_destroy(opj_t1_t *p_t1) return; } - /* encoder uses tile buffer, so no need to free */ - if (!p_t1->encoder && p_t1->data) { + if (p_t1->data) { opj_aligned_free(p_t1->data); p_t1->data = 00; } @@ -1601,15 +1580,19 @@ void opj_t1_destroy(opj_t1_t *p_t1) p_t1->flags = 00; } + opj_free(p_t1->cblkdatabuffer); + opj_free(p_t1); } typedef struct { + OPJ_BOOL whole_tile_decoding; OPJ_UINT32 resno; opj_tcd_cblk_dec_t* cblk; opj_tcd_band_t* band; opj_tcd_tilecomp_t* tilec; opj_tccp_t* tccp; + OPJ_BOOL mustuse_cblkdatabuffer; volatile OPJ_BOOL* pret; opj_event_mgr_t *p_manager; opj_mutex_t* p_manager_mutex; @@ -1637,12 +1620,44 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) OPJ_UINT32 tile_w; job = (opj_t1_cblk_decode_processing_job_t*) user_data; - resno = job->resno; + cblk = job->cblk; + + if (!job->whole_tile_decoding) { + cblk_w = (OPJ_UINT32)(cblk->x1 - cblk->x0); + cblk_h = (OPJ_UINT32)(cblk->y1 - cblk->y0); + + cblk->decoded_data = (OPJ_INT32*)opj_aligned_malloc(sizeof(OPJ_INT32) * + cblk_w * cblk_h); + if (cblk->decoded_data == NULL) { + if (job->p_manager_mutex) { + opj_mutex_lock(job->p_manager_mutex); + } + opj_event_msg(job->p_manager, EVT_ERROR, + "Cannot allocate cblk->decoded_data\n"); + if (job->p_manager_mutex) { + opj_mutex_unlock(job->p_manager_mutex); + } + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } + /* Zero-init required */ + memset(cblk->decoded_data, 0, sizeof(OPJ_INT32) * cblk_w * cblk_h); + } else if (cblk->decoded_data) { + /* Not sure if that code path can happen, but better be */ + /* safe than sorry */ + opj_aligned_free(cblk->decoded_data); + cblk->decoded_data = NULL; + } + + resno = job->resno; band = job->band; tilec = job->tilec; tccp = job->tccp; - tile_w = (OPJ_UINT32)(tilec->x1 - tilec->x0); + tile_w = (OPJ_UINT32)(tilec->resolutions[tilec->minimum_num_resolutions - 1].x1 + - + tilec->resolutions[tilec->minimum_num_resolutions - 1].x0); if (!*(job->pret)) { opj_free(job); @@ -1652,21 +1667,52 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) t1 = (opj_t1_t*) opj_tls_get(tls, OPJ_TLS_KEY_T1); if (t1 == NULL) { t1 = opj_t1_create(OPJ_FALSE); - opj_tls_set(tls, OPJ_TLS_KEY_T1, t1, opj_t1_destroy_wrapper); + if (t1 == NULL) { + opj_event_msg(job->p_manager, EVT_ERROR, + "Cannot allocate Tier 1 handle\n"); + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } + if (!opj_tls_set(tls, OPJ_TLS_KEY_T1, t1, opj_t1_destroy_wrapper)) { + opj_event_msg(job->p_manager, EVT_ERROR, + "Unable to set t1 handle as TLS\n"); + opj_t1_destroy(t1); + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } } + t1->mustuse_cblkdatabuffer = job->mustuse_cblkdatabuffer; - if (OPJ_FALSE == opj_t1_decode_cblk( - t1, - cblk, - band->bandno, - (OPJ_UINT32)tccp->roishift, - tccp->cblksty, - job->p_manager, - job->p_manager_mutex, - job->check_pterm)) { - *(job->pret) = OPJ_FALSE; - opj_free(job); - return; + if ((tccp->cblksty & J2K_CCP_CBLKSTY_HT) != 0) { + if (OPJ_FALSE == opj_t1_ht_decode_cblk( + t1, + cblk, + band->bandno, + (OPJ_UINT32)tccp->roishift, + tccp->cblksty, + job->p_manager, + job->p_manager_mutex, + job->check_pterm)) { + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } + } else { + if (OPJ_FALSE == opj_t1_decode_cblk( + t1, + cblk, + band->bandno, + (OPJ_UINT32)tccp->roishift, + tccp->cblksty, + job->p_manager, + job->p_manager_mutex, + job->check_pterm)) { + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } } x = cblk->x0 - band->x0; @@ -1680,7 +1726,7 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) y += pres->y1 - pres->y0; } - datap = t1->data; + datap = cblk->decoded_data ? cblk->decoded_data : t1->data; cblk_w = t1->w; cblk_h = t1->h; @@ -1705,9 +1751,50 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) } } } - if (tccp->qmfbid == 1) { - OPJ_INT32* OPJ_RESTRICT tiledp = &tilec->data[(OPJ_UINT32)y * tile_w + - (OPJ_UINT32)x]; + + /* Both can be non NULL if for example decoding a full tile and then */ + /* partially a tile. In which case partial decoding should be the */ + /* priority */ + assert((cblk->decoded_data != NULL) || (tilec->data != NULL)); + + if (cblk->decoded_data) { + OPJ_UINT32 cblk_size = cblk_w * cblk_h; + if (tccp->qmfbid == 1) { + for (i = 0; i < cblk_size; ++i) { + datap[i] /= 2; + } + } else { /* if (tccp->qmfbid == 0) */ + const float stepsize = 0.5f * band->stepsize; + i = 0; +#ifdef __SSE2__ + { + const __m128 xmm_stepsize = _mm_set1_ps(stepsize); + for (; i < (cblk_size & ~15U); i += 16) { + __m128 xmm0_data = _mm_cvtepi32_ps(_mm_load_si128((__m128i * const)( + datap + 0))); + __m128 xmm1_data = _mm_cvtepi32_ps(_mm_load_si128((__m128i * const)( + datap + 4))); + __m128 xmm2_data = _mm_cvtepi32_ps(_mm_load_si128((__m128i * const)( + datap + 8))); + __m128 xmm3_data = _mm_cvtepi32_ps(_mm_load_si128((__m128i * const)( + datap + 12))); + _mm_store_ps((float*)(datap + 0), _mm_mul_ps(xmm0_data, xmm_stepsize)); + _mm_store_ps((float*)(datap + 4), _mm_mul_ps(xmm1_data, xmm_stepsize)); + _mm_store_ps((float*)(datap + 8), _mm_mul_ps(xmm2_data, xmm_stepsize)); + _mm_store_ps((float*)(datap + 12), _mm_mul_ps(xmm3_data, xmm_stepsize)); + datap += 16; + } + } +#endif + for (; i < cblk_size; ++i) { + OPJ_FLOAT32 tmp = ((OPJ_FLOAT32)(*datap)) * stepsize; + memcpy(datap, &tmp, sizeof(tmp)); + datap++; + } + } + } else if (tccp->qmfbid == 1) { + OPJ_INT32* OPJ_RESTRICT tiledp = &tilec->data[(OPJ_SIZE_T)y * tile_w + + (OPJ_SIZE_T)x]; for (j = 0; j < cblk_h; ++j) { i = 0; for (; i < (cblk_w & ~(OPJ_UINT32)3U); i += 4U) { @@ -1715,23 +1802,24 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) OPJ_INT32 tmp1 = datap[(j * cblk_w) + i + 1U]; OPJ_INT32 tmp2 = datap[(j * cblk_w) + i + 2U]; OPJ_INT32 tmp3 = datap[(j * cblk_w) + i + 3U]; - ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 0U] = tmp0 / 2; - ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 1U] = tmp1 / 2; - ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 2U] = tmp2 / 2; - ((OPJ_INT32*)tiledp)[(j * tile_w) + i + 3U] = tmp3 / 2; + ((OPJ_INT32*)tiledp)[(j * (OPJ_SIZE_T)tile_w) + i + 0U] = tmp0 / 2; + ((OPJ_INT32*)tiledp)[(j * (OPJ_SIZE_T)tile_w) + i + 1U] = tmp1 / 2; + ((OPJ_INT32*)tiledp)[(j * (OPJ_SIZE_T)tile_w) + i + 2U] = tmp2 / 2; + ((OPJ_INT32*)tiledp)[(j * (OPJ_SIZE_T)tile_w) + i + 3U] = tmp3 / 2; } for (; 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 * (OPJ_SIZE_T)tile_w) + i] = tmp / 2; } } } else { /* if (tccp->qmfbid == 0) */ - OPJ_FLOAT32* OPJ_RESTRICT tiledp = (OPJ_FLOAT32*) &tilec->data[(OPJ_UINT32)y * - tile_w + (OPJ_UINT32)x]; + const float stepsize = 0.5f * band->stepsize; + OPJ_FLOAT32* OPJ_RESTRICT tiledp = (OPJ_FLOAT32*) &tilec->data[(OPJ_SIZE_T)y * + tile_w + (OPJ_SIZE_T)x]; for (j = 0; j < cblk_h; ++j) { OPJ_FLOAT32* OPJ_RESTRICT tiledp2 = tiledp; for (i = 0; i < cblk_w; ++i) { - OPJ_FLOAT32 tmp = (OPJ_FLOAT32) * datap * band->stepsize; + OPJ_FLOAT32 tmp = (OPJ_FLOAT32) * datap * stepsize; *tiledp2 = tmp; datap++; tiledp2++; @@ -1744,7 +1832,7 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) } -void opj_t1_decode_cblks(opj_thread_pool_t* tp, +void opj_t1_decode_cblks(opj_tcd_t* tcd, volatile OPJ_BOOL* pret, opj_tcd_tilecomp_t* tilec, opj_tccp_t* tccp, @@ -1753,8 +1841,14 @@ void opj_t1_decode_cblks(opj_thread_pool_t* tp, OPJ_BOOL check_pterm ) { + opj_thread_pool_t* tp = tcd->thread_pool; OPJ_UINT32 resno, bandno, precno, cblkno; +#ifdef DEBUG_VERBOSE + OPJ_UINT32 codeblocks_decoded = 0; + printf("Enter opj_t1_decode_cblks()\n"); +#endif + for (resno = 0; resno < tilec->minimum_num_resolutions; ++resno) { opj_tcd_resolution_t* res = &tilec->resolutions[resno]; @@ -1764,16 +1858,77 @@ void opj_t1_decode_cblks(opj_thread_pool_t* tp, for (precno = 0; precno < res->pw * res->ph; ++precno) { opj_tcd_precinct_t* precinct = &band->precincts[precno]; + if (!opj_tcd_is_subband_area_of_interest(tcd, + tilec->compno, + resno, + band->bandno, + (OPJ_UINT32)precinct->x0, + (OPJ_UINT32)precinct->y0, + (OPJ_UINT32)precinct->x1, + (OPJ_UINT32)precinct->y1)) { + for (cblkno = 0; cblkno < precinct->cw * precinct->ch; ++cblkno) { + opj_tcd_cblk_dec_t* cblk = &precinct->cblks.dec[cblkno]; + if (cblk->decoded_data) { +#ifdef DEBUG_VERBOSE + printf("Discarding codeblock %d,%d at resno=%d, bandno=%d\n", + cblk->x0, cblk->y0, resno, bandno); +#endif + opj_aligned_free(cblk->decoded_data); + cblk->decoded_data = NULL; + } + } + continue; + } + for (cblkno = 0; cblkno < precinct->cw * precinct->ch; ++cblkno) { opj_tcd_cblk_dec_t* cblk = &precinct->cblks.dec[cblkno]; opj_t1_cblk_decode_processing_job_t* job; + if (!opj_tcd_is_subband_area_of_interest(tcd, + tilec->compno, + resno, + band->bandno, + (OPJ_UINT32)cblk->x0, + (OPJ_UINT32)cblk->y0, + (OPJ_UINT32)cblk->x1, + (OPJ_UINT32)cblk->y1)) { + if (cblk->decoded_data) { +#ifdef DEBUG_VERBOSE + printf("Discarding codeblock %d,%d at resno=%d, bandno=%d\n", + cblk->x0, cblk->y0, resno, bandno); +#endif + opj_aligned_free(cblk->decoded_data); + cblk->decoded_data = NULL; + } + continue; + } + + if (!tcd->whole_tile_decoding) { + OPJ_UINT32 cblk_w = (OPJ_UINT32)(cblk->x1 - cblk->x0); + OPJ_UINT32 cblk_h = (OPJ_UINT32)(cblk->y1 - cblk->y0); + if (cblk->decoded_data != NULL) { +#ifdef DEBUG_VERBOSE + printf("Reusing codeblock %d,%d at resno=%d, bandno=%d\n", + cblk->x0, cblk->y0, resno, bandno); +#endif + continue; + } + if (cblk_w == 0 || cblk_h == 0) { + continue; + } +#ifdef DEBUG_VERBOSE + printf("Decoding codeblock %d,%d at resno=%d, bandno=%d\n", + cblk->x0, cblk->y0, resno, bandno); +#endif + } + job = (opj_t1_cblk_decode_processing_job_t*) opj_calloc(1, sizeof(opj_t1_cblk_decode_processing_job_t)); if (!job) { *pret = OPJ_FALSE; return; } + job->whole_tile_decoding = tcd->whole_tile_decoding; job->resno = resno; job->cblk = cblk; job->band = band; @@ -1783,7 +1938,11 @@ void opj_t1_decode_cblks(opj_thread_pool_t* tp, job->p_manager_mutex = p_manager_mutex; job->p_manager = p_manager; job->check_pterm = check_pterm; + job->mustuse_cblkdatabuffer = opj_thread_pool_get_thread_count(tp) > 1; opj_thread_pool_submit_job(tp, opj_t1_clbl_decode_processor, job); +#ifdef DEBUG_VERBOSE + codeblocks_decoded ++; +#endif if (!(*pret)) { return; } @@ -1792,6 +1951,9 @@ void opj_t1_decode_cblks(opj_thread_pool_t* tp, } /* bandno */ } /* resno */ +#ifdef DEBUG_VERBOSE + printf("Leave opj_t1_decode_cblks(). Number decoded: %d\n", codeblocks_decoded); +#endif return; } @@ -1810,7 +1972,10 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, OPJ_INT32 bpno_plus_one; OPJ_UINT32 passtype; OPJ_UINT32 segno, passno; + OPJ_BYTE* cblkdata = NULL; + OPJ_UINT32 cblkdataindex = 0; OPJ_BYTE type = T1_TYPE_MQ; /* BYPASS mode */ + OPJ_INT32* original_t1_data = NULL; mqc->lut_ctxno_zc_orient = lut_ctxno_zc + (orient << 9); @@ -1822,6 +1987,18 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, } bpno_plus_one = (OPJ_INT32)(roishift + cblk->numbps); + if (bpno_plus_one >= 31) { + if (p_manager_mutex) { + opj_mutex_lock(p_manager_mutex); + } + opj_event_msg(p_manager, EVT_WARNING, + "opj_t1_decode_cblk(): unsupported bpno_plus_one = %d >= 31\n", + bpno_plus_one); + if (p_manager_mutex) { + opj_mutex_unlock(p_manager_mutex); + } + return OPJ_FALSE; + } passtype = 2; opj_mqc_resetstates(mqc); @@ -1829,23 +2006,68 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, opj_mqc_setstate(mqc, T1_CTXNO_AGG, 0, 3); opj_mqc_setstate(mqc, T1_CTXNO_ZC, 0, 4); + /* Even if we have a single chunk, in multi-threaded decoding */ + /* the insertion of our synthetic marker might potentially override */ + /* valid codestream of other codeblocks decoded in parallel. */ + if (cblk->numchunks > 1 || t1->mustuse_cblkdatabuffer) { + OPJ_UINT32 i; + OPJ_UINT32 cblk_len; + + /* Compute whole codeblock length from chunk lengths */ + cblk_len = 0; + for (i = 0; i < cblk->numchunks; i++) { + cblk_len += cblk->chunks[i].len; + } + + /* Allocate temporary memory if needed */ + if (cblk_len + OPJ_COMMON_CBLK_DATA_EXTRA > t1->cblkdatabuffersize) { + cblkdata = (OPJ_BYTE*)opj_realloc(t1->cblkdatabuffer, + cblk_len + OPJ_COMMON_CBLK_DATA_EXTRA); + if (cblkdata == NULL) { + return OPJ_FALSE; + } + t1->cblkdatabuffer = cblkdata; + memset(t1->cblkdatabuffer + cblk_len, 0, OPJ_COMMON_CBLK_DATA_EXTRA); + t1->cblkdatabuffersize = cblk_len + OPJ_COMMON_CBLK_DATA_EXTRA; + } + + /* Concatenate all chunks */ + cblkdata = t1->cblkdatabuffer; + cblk_len = 0; + for (i = 0; i < cblk->numchunks; i++) { + memcpy(cblkdata + cblk_len, cblk->chunks[i].data, cblk->chunks[i].len); + cblk_len += cblk->chunks[i].len; + } + } else if (cblk->numchunks == 1) { + cblkdata = cblk->chunks[0].data; + } else { + /* Not sure if that can happen in practice, but avoid Coverity to */ + /* think we will dereference a null cblkdta pointer */ + return OPJ_TRUE; + } + + /* For subtile decoding, directly decode in the decoded_data buffer of */ + /* the code-block. Hack t1->data to point to it, and restore it later */ + if (cblk->decoded_data) { + original_t1_data = t1->data; + t1->data = cblk->decoded_data; + } + for (segno = 0; segno < cblk->real_num_segs; ++segno) { opj_tcd_seg_t *seg = &cblk->segs[segno]; /* BYPASS mode */ 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; - } + if (type == T1_TYPE_RAW) { - opj_mqc_raw_init_dec(mqc, (*seg->data) + seg->dataindex, seg->len, + opj_mqc_raw_init_dec(mqc, cblkdata + cblkdataindex, seg->len, OPJ_COMMON_CBLK_DATA_EXTRA); } else { - opj_mqc_init_dec(mqc, (*seg->data) + seg->dataindex, seg->len, + opj_mqc_init_dec(mqc, cblkdata + cblkdataindex, seg->len, OPJ_COMMON_CBLK_DATA_EXTRA); } + cblkdataindex += seg->len; for (passno = 0; (passno < seg->real_num_passes) && (bpno_plus_one >= 1); ++passno) { @@ -1910,118 +2132,241 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, } } + /* Restore original t1->data is needed */ + if (cblk->decoded_data) { + t1->data = original_t1_data; + } + return OPJ_TRUE; } +typedef struct { + OPJ_UINT32 compno; + OPJ_UINT32 resno; + opj_tcd_cblk_enc_t* cblk; + opj_tcd_tile_t *tile; + opj_tcd_band_t* band; + opj_tcd_tilecomp_t* tilec; + opj_tccp_t* tccp; + const OPJ_FLOAT64 * mct_norms; + OPJ_UINT32 mct_numcomps; + volatile OPJ_BOOL* pret; + opj_mutex_t* mutex; +} opj_t1_cblk_encode_processing_job_t; +/** Procedure to deal with a asynchronous code-block encoding job. + * + * @param user_data Pointer to a opj_t1_cblk_encode_processing_job_t* structure + * @param tls TLS handle. + */ +static void opj_t1_cblk_encode_processor(void* user_data, opj_tls_t* tls) +{ + opj_t1_cblk_encode_processing_job_t* job = + (opj_t1_cblk_encode_processing_job_t*)user_data; + opj_tcd_cblk_enc_t* cblk = job->cblk; + const opj_tcd_band_t* band = job->band; + const opj_tcd_tilecomp_t* tilec = job->tilec; + const opj_tccp_t* tccp = job->tccp; + const OPJ_UINT32 resno = job->resno; + opj_t1_t* t1; + const OPJ_UINT32 tile_w = (OPJ_UINT32)(tilec->x1 - tilec->x0); + + OPJ_INT32* OPJ_RESTRICT tiledp; + OPJ_UINT32 cblk_w; + OPJ_UINT32 cblk_h; + OPJ_UINT32 i, j; + + OPJ_INT32 x = cblk->x0 - band->x0; + OPJ_INT32 y = cblk->y0 - band->y0; + + if (!*(job->pret)) { + opj_free(job); + return; + } + + t1 = (opj_t1_t*) opj_tls_get(tls, OPJ_TLS_KEY_T1); + if (t1 == NULL) { + t1 = opj_t1_create(OPJ_TRUE); /* OPJ_TRUE == T1 for encoding */ + opj_tls_set(tls, OPJ_TLS_KEY_T1, t1, opj_t1_destroy_wrapper); + } + + if (band->bandno & 1) { + opj_tcd_resolution_t *pres = &tilec->resolutions[resno - 1]; + x += pres->x1 - pres->x0; + } + if (band->bandno & 2) { + opj_tcd_resolution_t *pres = &tilec->resolutions[resno - 1]; + y += pres->y1 - pres->y0; + } -OPJ_BOOL opj_t1_encode_cblks(opj_t1_t *t1, + if (!opj_t1_allocate_buffers( + t1, + (OPJ_UINT32)(cblk->x1 - cblk->x0), + (OPJ_UINT32)(cblk->y1 - cblk->y0))) { + *(job->pret) = OPJ_FALSE; + opj_free(job); + return; + } + + cblk_w = t1->w; + cblk_h = t1->h; + + tiledp = &tilec->data[(OPJ_SIZE_T)y * tile_w + (OPJ_SIZE_T)x]; + + if (tccp->qmfbid == 1) { + /* Do multiplication on unsigned type, even if the + * underlying type is signed, to avoid potential + * int overflow on large value (the output will be + * incorrect in such situation, but whatever...) + * This assumes complement-to-2 signed integer + * representation + * Fixes https://github.com/uclouvain/openjpeg/issues/1053 + */ + OPJ_UINT32* OPJ_RESTRICT tiledp_u = (OPJ_UINT32*) tiledp; + OPJ_UINT32* OPJ_RESTRICT t1data = (OPJ_UINT32*) t1->data; + /* Change from "natural" order to "zigzag" order of T1 passes */ + for (j = 0; j < (cblk_h & ~3U); j += 4) { + for (i = 0; i < cblk_w; ++i) { + t1data[0] = tiledp_u[(j + 0) * tile_w + i] << T1_NMSEDEC_FRACBITS; + t1data[1] = tiledp_u[(j + 1) * tile_w + i] << T1_NMSEDEC_FRACBITS; + t1data[2] = tiledp_u[(j + 2) * tile_w + i] << T1_NMSEDEC_FRACBITS; + t1data[3] = tiledp_u[(j + 3) * tile_w + i] << T1_NMSEDEC_FRACBITS; + t1data += 4; + } + } + if (j < cblk_h) { + for (i = 0; i < cblk_w; ++i) { + OPJ_UINT32 k; + for (k = j; k < cblk_h; k++) { + t1data[0] = tiledp_u[k * tile_w + i] << T1_NMSEDEC_FRACBITS; + t1data ++; + } + } + } + } else { /* if (tccp->qmfbid == 0) */ + OPJ_FLOAT32* OPJ_RESTRICT tiledp_f = (OPJ_FLOAT32*) tiledp; + OPJ_INT32* OPJ_RESTRICT t1data = t1->data; + /* Change from "natural" order to "zigzag" order of T1 passes */ + for (j = 0; j < (cblk_h & ~3U); j += 4) { + for (i = 0; i < cblk_w; ++i) { + t1data[0] = (OPJ_INT32)opj_lrintf((tiledp_f[(j + 0) * tile_w + i] / + band->stepsize) * (1 << T1_NMSEDEC_FRACBITS)); + t1data[1] = (OPJ_INT32)opj_lrintf((tiledp_f[(j + 1) * tile_w + i] / + band->stepsize) * (1 << T1_NMSEDEC_FRACBITS)); + t1data[2] = (OPJ_INT32)opj_lrintf((tiledp_f[(j + 2) * tile_w + i] / + band->stepsize) * (1 << T1_NMSEDEC_FRACBITS)); + t1data[3] = (OPJ_INT32)opj_lrintf((tiledp_f[(j + 3) * tile_w + i] / + band->stepsize) * (1 << T1_NMSEDEC_FRACBITS)); + t1data += 4; + } + } + if (j < cblk_h) { + for (i = 0; i < cblk_w; ++i) { + OPJ_UINT32 k; + for (k = j; k < cblk_h; k++) { + t1data[0] = (OPJ_INT32)opj_lrintf((tiledp_f[k * tile_w + i] / band->stepsize) + * (1 << T1_NMSEDEC_FRACBITS)); + t1data ++; + } + } + } + } + + { + OPJ_FLOAT64 cumwmsedec = + opj_t1_encode_cblk( + t1, + cblk, + band->bandno, + job->compno, + tilec->numresolutions - 1 - resno, + tccp->qmfbid, + band->stepsize, + tccp->cblksty, + job->tile->numcomps, + job->mct_norms, + job->mct_numcomps); + if (job->mutex) { + opj_mutex_lock(job->mutex); + } + job->tile->distotile += cumwmsedec; + if (job->mutex) { + opj_mutex_unlock(job->mutex); + } + } + + opj_free(job); +} + + +OPJ_BOOL opj_t1_encode_cblks(opj_tcd_t* tcd, opj_tcd_tile_t *tile, opj_tcp_t *tcp, const OPJ_FLOAT64 * mct_norms, OPJ_UINT32 mct_numcomps ) { + volatile OPJ_BOOL ret = OPJ_TRUE; + opj_thread_pool_t* tp = tcd->thread_pool; OPJ_UINT32 compno, resno, bandno, precno, cblkno; + opj_mutex_t* mutex = opj_mutex_create(); - tile->distotile = 0; /* fixed_quality */ + tile->distotile = 0; for (compno = 0; compno < tile->numcomps; ++compno) { opj_tcd_tilecomp_t* tilec = &tile->comps[compno]; opj_tccp_t* tccp = &tcp->tccps[compno]; - OPJ_UINT32 tile_w = (OPJ_UINT32)(tilec->x1 - tilec->x0); for (resno = 0; resno < tilec->numresolutions; ++resno) { opj_tcd_resolution_t *res = &tilec->resolutions[resno]; for (bandno = 0; bandno < res->numbands; ++bandno) { opj_tcd_band_t* OPJ_RESTRICT band = &res->bands[bandno]; - OPJ_INT32 bandconst; /* Skip empty bands */ if (opj_tcd_is_band_empty(band)) { continue; } - - bandconst = 8192 * 8192 / ((OPJ_INT32) floor(band->stepsize * 8192)); for (precno = 0; precno < res->pw * res->ph; ++precno) { opj_tcd_precinct_t *prc = &band->precincts[precno]; for (cblkno = 0; cblkno < prc->cw * prc->ch; ++cblkno) { opj_tcd_cblk_enc_t* cblk = &prc->cblks.enc[cblkno]; - OPJ_INT32* OPJ_RESTRICT tiledp; - OPJ_UINT32 cblk_w; - OPJ_UINT32 cblk_h; - OPJ_UINT32 i, j, tileIndex = 0, tileLineAdvance; - - OPJ_INT32 x = cblk->x0 - band->x0; - OPJ_INT32 y = cblk->y0 - band->y0; - if (band->bandno & 1) { - opj_tcd_resolution_t *pres = &tilec->resolutions[resno - 1]; - x += pres->x1 - pres->x0; - } - if (band->bandno & 2) { - opj_tcd_resolution_t *pres = &tilec->resolutions[resno - 1]; - y += pres->y1 - pres->y0; - } - if (!opj_t1_allocate_buffers( - t1, - (OPJ_UINT32)(cblk->x1 - cblk->x0), - (OPJ_UINT32)(cblk->y1 - cblk->y0))) { - return OPJ_FALSE; + opj_t1_cblk_encode_processing_job_t* job = + (opj_t1_cblk_encode_processing_job_t*) opj_calloc(1, + sizeof(opj_t1_cblk_encode_processing_job_t)); + if (!job) { + ret = OPJ_FALSE; + goto end; } - - cblk_w = t1->w; - cblk_h = t1->h; - tileLineAdvance = tile_w - cblk_w; - - tiledp = &tilec->data[(OPJ_UINT32)y * tile_w + (OPJ_UINT32)x]; - t1->data = tiledp; - t1->data_stride = tile_w; - if (tccp->qmfbid == 1) { - for (j = 0; j < cblk_h; ++j) { - for (i = 0; i < cblk_w; ++i) { - tiledp[tileIndex] *= (1 << T1_NMSEDEC_FRACBITS); - tileIndex++; - } - tileIndex += tileLineAdvance; - } - } else { /* if (tccp->qmfbid == 0) */ - for (j = 0; j < cblk_h; ++j) { - for (i = 0; i < cblk_w; ++i) { - OPJ_INT32 tmp = tiledp[tileIndex]; - tiledp[tileIndex] = - opj_int_fix_mul_t1( - tmp, - bandconst); - tileIndex++; - } - tileIndex += tileLineAdvance; - } - } - - opj_t1_encode_cblk( - t1, - cblk, - band->bandno, - compno, - tilec->numresolutions - 1 - resno, - tccp->qmfbid, - band->stepsize, - tccp->cblksty, - tile->numcomps, - tile, - mct_norms, - mct_numcomps); + job->compno = compno; + job->tile = tile; + job->resno = resno; + job->cblk = cblk; + job->band = band; + job->tilec = tilec; + job->tccp = tccp; + job->mct_norms = mct_norms; + job->mct_numcomps = mct_numcomps; + job->pret = &ret; + job->mutex = mutex; + opj_thread_pool_submit_job(tp, opj_t1_cblk_encode_processor, job); } /* cblkno */ } /* precno */ } /* bandno */ } /* resno */ } /* compno */ - return OPJ_TRUE; + +end: + opj_thread_pool_wait_completion(tcd->thread_pool, 0); + if (mutex) { + opj_mutex_destroy(mutex); + } + + return ret; } /* Returns whether the pass (bpno, passtype) is terminated */ @@ -2055,19 +2400,17 @@ static int opj_t1_enc_is_term_pass(opj_tcd_cblk_enc_t* cblk, } -/** mod fixed_quality */ -static void opj_t1_encode_cblk(opj_t1_t *t1, - opj_tcd_cblk_enc_t* cblk, - OPJ_UINT32 orient, - OPJ_UINT32 compno, - OPJ_UINT32 level, - OPJ_UINT32 qmfbid, - OPJ_FLOAT64 stepsize, - OPJ_UINT32 cblksty, - OPJ_UINT32 numcomps, - opj_tcd_tile_t * tile, - const OPJ_FLOAT64 * mct_norms, - OPJ_UINT32 mct_numcomps) +static OPJ_FLOAT64 opj_t1_encode_cblk(opj_t1_t *t1, + opj_tcd_cblk_enc_t* cblk, + OPJ_UINT32 orient, + OPJ_UINT32 compno, + OPJ_UINT32 level, + OPJ_UINT32 qmfbid, + OPJ_FLOAT64 stepsize, + OPJ_UINT32 cblksty, + OPJ_UINT32 numcomps, + const OPJ_FLOAT64 * mct_norms, + OPJ_UINT32 mct_numcomps) { OPJ_FLOAT64 cumwmsedec = 0.0; @@ -2081,6 +2424,7 @@ static void opj_t1_encode_cblk(opj_t1_t *t1, OPJ_UINT32 i, j; OPJ_BYTE type = T1_TYPE_MQ; OPJ_FLOAT64 tempwmsedec; + OPJ_INT32* datap; #ifdef EXTRA_DEBUG printf("encode_cblk(x=%d,y=%d,x1=%d,y1=%d,orient=%d,compno=%d,level=%d\n", @@ -2090,15 +2434,35 @@ static void opj_t1_encode_cblk(opj_t1_t *t1, mqc->lut_ctxno_zc_orient = lut_ctxno_zc + (orient << 9); max = 0; - for (i = 0; i < t1->w; ++i) { - for (j = 0; j < t1->h; ++j) { - OPJ_INT32 tmp = abs(t1->data[i + j * t1->data_stride]); - max = opj_int_max(max, tmp); + datap = t1->data; + for (j = 0; j < t1->h; ++j) { + const OPJ_UINT32 w = t1->w; + for (i = 0; i < w; ++i, ++datap) { + OPJ_INT32 tmp = *datap; + if (tmp < 0) { + OPJ_UINT32 tmp_unsigned; + if (tmp == INT_MIN) { + /* To avoid undefined behaviour when negating INT_MIN */ + /* but if we go here, it means we have supplied an input */ + /* with more bit depth than we we can really support. */ + /* Cf https://github.com/uclouvain/openjpeg/issues/1432 */ + tmp = INT_MIN + 1; + } + max = opj_int_max(max, -tmp); + tmp_unsigned = opj_to_smr(tmp); + memcpy(datap, &tmp_unsigned, sizeof(OPJ_INT32)); + } else { + max = opj_int_max(max, tmp); + } } } cblk->numbps = max ? (OPJ_UINT32)((opj_int_floorlog2(max) + 1) - T1_NMSEDEC_FRACBITS) : 0; + if (cblk->numbps == 0) { + cblk->totalpasses = 0; + return cumwmsedec; + } bpno = (OPJ_INT32)(cblk->numbps - 1); passtype = 2; @@ -2139,11 +2503,9 @@ static void opj_t1_encode_cblk(opj_t1_t *t1, break; } - /* fixed_quality */ tempwmsedec = opj_t1_getwmsedec(nmsedec, compno, level, orient, bpno, qmfbid, stepsize, numcomps, mct_norms, mct_numcomps) ; cumwmsedec += tempwmsedec; - tile->distotile += tempwmsedec; pass->distortiondec = cumwmsedec; if (opj_t1_enc_is_term_pass(cblk, cblksty, bpno, passtype)) { @@ -2225,4 +2587,6 @@ static void opj_t1_encode_cblk(opj_t1_t *t1, } } #endif + + return cumwmsedec; }