X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fopenjp2%2Ft1.c;h=b32daeaee4de457ad7c8ff4b44631405d5453137;hb=79b199a8fee2a0d51d4389fcde3f5f4dd01971eb;hp=44a2f2437e61a7740945d3bd5c8866ae07ebd849;hpb=98b93103613c90753fb6c57696f9403f8ea0b1d6;p=openjpeg.git diff --git a/src/lib/openjp2/t1.c b/src/lib/openjp2/t1.c index 44a2f243..b32daeae 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 */ @@ -335,7 +348,7 @@ static INLINE void opj_t1_enc_sigpass_step(opj_t1_t *t1, 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; + v = (opj_int_abs(*datap) & one) ? 1 : 0; #ifdef DEBUG_ENC_SIG fprintf(stderr, " ctxt1=%d\n", ctxt1); #endif @@ -722,7 +735,7 @@ static INLINE void opj_t1_enc_refpass_step(opj_t1_t *t1, 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; + v = (opj_int_abs(*datap) & one) ? 1 : 0; #ifdef DEBUG_ENC_REF fprintf(stderr, " ctxt=%d\n", ctxt); #endif @@ -1052,6 +1065,7 @@ static void opj_t1_enc_clnpass_step( for (ci = runlen; ci < lim; ++ci) { OPJ_UINT32 vsc; opj_flag_t flags; + OPJ_UINT32 ctxt1; flags = *flagsp; @@ -1060,12 +1074,12 @@ static void opj_t1_enc_clnpass_step( } if (!(flags & ((T1_SIGMA_THIS | T1_PI_THIS) << (ci * 3U)))) { - OPJ_UINT32 ctxt1 = opj_t1_getctxno_zc(mqc, flags >> (ci * 3U)); + 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; + v = (opj_int_abs(*datap) & one) ? 1 : 0; opj_mqc_encode(mqc, v); if (v) { OPJ_UINT32 ctxt2, spb; @@ -1439,7 +1453,7 @@ static OPJ_BOOL opj_t1_allocate_buffers( if (!t1->encoder) { 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) { @@ -1563,6 +1577,7 @@ void opj_t1_destroy(opj_t1_t *p_t1) } typedef struct { + OPJ_BOOL whole_tile_decoding; OPJ_UINT32 resno; opj_tcd_cblk_dec_t* cblk; opj_tcd_band_t* band; @@ -1596,8 +1611,38 @@ 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; @@ -1613,7 +1658,21 @@ 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; @@ -1668,36 +1727,48 @@ static void opj_t1_clbl_decode_processor(void* user_data, opj_tls_t* tls) } } + /* 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 (j = 0; j < cblk_h; ++j) { - i = 0; - for (; i < (cblk_w & ~(OPJ_UINT32)3U); i += 4U) { - OPJ_INT32 tmp0 = datap[(j * cblk_w) + i + 0U]; - 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]; - datap[(j * cblk_w) + i + 0U] = tmp0 / 2; - datap[(j * cblk_w) + i + 1U] = tmp1 / 2; - datap[(j * cblk_w) + i + 2U] = tmp2 / 2; - datap[(j * cblk_w) + i + 3U] = tmp3 / 2; - } - for (; i < cblk_w; ++i) { - datap[(j * cblk_w) + i] /= 2; - } + for (i = 0; i < cblk_size; ++i) { + datap[i] /= 2; } } else { /* if (tccp->qmfbid == 0) */ - for (j = 0; j < cblk_h; ++j) { - for (i = 0; i < cblk_w; ++i) { - OPJ_FLOAT32 tmp = ((OPJ_FLOAT32)(*datap)) * band->stepsize; - memcpy(datap, &tmp, sizeof(tmp)); - datap++; + i = 0; +#ifdef __SSE2__ + { + const __m128 xmm_stepsize = _mm_set1_ps(band->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)) * band->stepsize; + memcpy(datap, &tmp, sizeof(tmp)); + datap++; + } } } else if (tccp->qmfbid == 1) { - OPJ_INT32* OPJ_RESTRICT tiledp = &tilec->data[(size_t)y * tile_w + - (size_t)x]; + 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) { @@ -1705,19 +1776,19 @@ 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 * (size_t)tile_w) + i + 0U] = tmp0 / 2; - ((OPJ_INT32*)tiledp)[(j * (size_t)tile_w) + i + 1U] = tmp1 / 2; - ((OPJ_INT32*)tiledp)[(j * (size_t)tile_w) + i + 2U] = tmp2 / 2; - ((OPJ_INT32*)tiledp)[(j * (size_t)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 * (size_t)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[(size_t)y * - tile_w + (size_t)x]; + 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) { @@ -1746,6 +1817,11 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, 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]; @@ -1763,6 +1839,17 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, (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; } @@ -1770,8 +1857,6 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, opj_tcd_cblk_dec_t* cblk = &precinct->cblks.dec[cblkno]; opj_t1_cblk_decode_processing_job_t* job; - assert(cblk->decoded_data == NULL); - if (!opj_tcd_is_subband_area_of_interest(tcd, tilec->compno, resno, @@ -1780,29 +1865,34 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, (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_w == 0 || cblk_h == 0) { + 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; } - /* Zero-init required */ - cblk->decoded_data = opj_calloc(1, cblk_w * cblk_h * sizeof(OPJ_INT32)); - if (cblk->decoded_data == NULL) { - if (p_manager_mutex) { - opj_mutex_lock(p_manager_mutex); - } - opj_event_msg(p_manager, EVT_ERROR, - "Cannot allocate cblk->decoded_data\n"); - if (p_manager_mutex) { - opj_mutex_unlock(p_manager_mutex); - } - *pret = OPJ_FALSE; - return; + 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, @@ -1811,6 +1901,7 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, *pret = OPJ_FALSE; return; } + job->whole_tile_decoding = tcd->whole_tile_decoding; job->resno = resno; job->cblk = cblk; job->band = band; @@ -1822,6 +1913,9 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, 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; } @@ -1830,6 +1924,9 @@ void opj_t1_decode_cblks(opj_tcd_t* tcd, } /* bandno */ } /* resno */ +#ifdef DEBUG_VERBOSE + printf("Leave opj_t1_decode_cblks(). Number decoded: %d\n", codeblocks_decoded); +#endif return; } @@ -1916,6 +2013,10 @@ static OPJ_BOOL opj_t1_decode_cblk(opj_t1_t *t1, } } 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 */ @@ -2053,7 +2154,7 @@ OPJ_BOOL opj_t1_encode_cblks(opj_t1_t *t1, OPJ_UINT32 cblk_w; OPJ_UINT32 cblk_h; OPJ_UINT32 i, j, tileLineAdvance; - size_t tileIndex = 0; + OPJ_SIZE_T tileIndex = 0; OPJ_INT32 x = cblk->x0 - band->x0; OPJ_INT32 y = cblk->y0 - band->y0; @@ -2077,13 +2178,22 @@ OPJ_BOOL opj_t1_encode_cblks(opj_t1_t *t1, cblk_h = t1->h; tileLineAdvance = tile_w - cblk_w; - tiledp = &tilec->data[(size_t)y * tile_w + (size_t)x]; + tiledp = &tilec->data[(OPJ_SIZE_T)y * tile_w + (OPJ_SIZE_T)x]; t1->data = tiledp; t1->data_stride = tile_w; 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; for (j = 0; j < cblk_h; ++j) { for (i = 0; i < cblk_w; ++i) { - tiledp[tileIndex] *= (1 << T1_NMSEDEC_FRACBITS); + tiledp_u[tileIndex] <<= T1_NMSEDEC_FRACBITS; tileIndex++; } tileIndex += tileLineAdvance;