summaryrefslogtreecommitdiff
path: root/third_party/libopenjpeg20/dwt.c
diff options
context:
space:
mode:
authorNicolas Pena <npm@chromium.org>2017-10-23 10:30:46 -0400
committerChromium commit bot <commit-bot@chromium.org>2017-10-23 22:40:23 +0000
commit826480cf599f61fe0366ab2bd5803dd53c9d0562 (patch)
treed42bca4255a2ecf484b8498b08c6132dbcffe86a /third_party/libopenjpeg20/dwt.c
parentebdba614b9683ddd1d50e8960639bc54c9d4bb7a (diff)
downloadpdfium-826480cf599f61fe0366ab2bd5803dd53c9d0562.tar.xz
Upgrade LibopenJPEG to 2.3chromium/3249
Bug: Change-Id: I4c968a4e5f41037d80e5dc64a1297cd2cbda31b1 Reviewed-on: https://pdfium-review.googlesource.com/16350 Reviewed-by: dsinclair <dsinclair@chromium.org> Commit-Queue: Nicolás Peña Moreno <npm@chromium.org>
Diffstat (limited to 'third_party/libopenjpeg20/dwt.c')
-rw-r--r--third_party/libopenjpeg20/dwt.c1438
1 files changed, 1236 insertions, 202 deletions
diff --git a/third_party/libopenjpeg20/dwt.c b/third_party/libopenjpeg20/dwt.c
index 8bf23b6d60..1455ee84a1 100644
--- a/third_party/libopenjpeg20/dwt.c
+++ b/third_party/libopenjpeg20/dwt.c
@@ -80,9 +80,9 @@
typedef struct dwt_local {
OPJ_INT32* mem;
OPJ_SIZE_T mem_count;
- OPJ_INT32 dn;
- OPJ_INT32 sn;
- OPJ_INT32 cas;
+ OPJ_INT32 dn; /* number of elements in high pass band */
+ OPJ_INT32 sn; /* number of elements in low pass band */
+ OPJ_INT32 cas; /* 0 = start on even coord, 1 = start on odd coord */
} opj_dwt_t;
typedef union {
@@ -91,9 +91,13 @@ typedef union {
typedef struct v4dwt_local {
opj_v4_t* wavelet ;
- OPJ_INT32 dn ;
- OPJ_INT32 sn ;
- OPJ_INT32 cas ;
+ OPJ_INT32 dn ; /* number of elements in high pass band */
+ OPJ_INT32 sn ; /* number of elements in low pass band */
+ OPJ_INT32 cas ; /* 0 = start on even coord, 1 = start on odd coord */
+ OPJ_UINT32 win_l_x0; /* start coord in low pass band */
+ OPJ_UINT32 win_l_x1; /* end coord in low pass band */
+ OPJ_UINT32 win_h_x0; /* start coord in high pass band */
+ OPJ_UINT32 win_h_x1; /* end coord in high pass band */
} opj_v4dwt_t ;
static const OPJ_FLOAT32 opj_dwt_alpha = 1.586134342f; /* 12994 */
@@ -129,13 +133,13 @@ Forward 5-3 wavelet transform in 1-D
*/
static void opj_dwt_encode_1(OPJ_INT32 *a, OPJ_SIZE_T a_count, OPJ_INT32 dn,
OPJ_INT32 sn, OPJ_INT32 cas);
-
/**
Forward 9-7 wavelet transform in 1-D
*/
static void opj_dwt_encode_1_real(OPJ_INT32 *a, OPJ_SIZE_T a_count,
OPJ_INT32 dn, OPJ_INT32 sn, OPJ_INT32 cas);
+
/**
Explicit calculation of the Quantization Stepsizes
*/
@@ -147,6 +151,10 @@ Inverse wavelet transform in 2-D.
static OPJ_BOOL opj_dwt_decode_tile(opj_thread_pool_t* tp,
const opj_tcd_tilecomp_t* tilec, OPJ_UINT32 i);
+static OPJ_BOOL opj_dwt_decode_partial_tile(
+ opj_tcd_tilecomp_t* tilec,
+ OPJ_UINT32 numres);
+
static OPJ_BOOL opj_dwt_encode_procedure(const opj_tcd_tilecomp_t * tilec,
void(*p_function)(OPJ_INT32 *, OPJ_SIZE_T, OPJ_INT32, OPJ_INT32, OPJ_INT32));
@@ -158,25 +166,38 @@ static OPJ_UINT32 opj_dwt_max_resolution(opj_tcd_resolution_t* OPJ_RESTRICT r,
/* </summary> */
static void opj_v4dwt_decode(opj_v4dwt_t* OPJ_RESTRICT dwt);
-static void opj_v4dwt_interleave_h(opj_v4dwt_t* OPJ_RESTRICT w,
- OPJ_FLOAT32* OPJ_RESTRICT a, OPJ_INT32 x, OPJ_INT32 size);
+static void opj_v4dwt_interleave_h(opj_v4dwt_t* OPJ_RESTRICT dwt,
+ OPJ_FLOAT32* OPJ_RESTRICT a,
+ OPJ_UINT32 width,
+ OPJ_UINT32 remaining_height);
-static void opj_v4dwt_interleave_v(opj_v4dwt_t* OPJ_RESTRICT v,
- OPJ_FLOAT32* OPJ_RESTRICT a, OPJ_INT32 x, OPJ_INT32 nb_elts_read);
+static void opj_v4dwt_interleave_v(opj_v4dwt_t* OPJ_RESTRICT dwt,
+ OPJ_FLOAT32* OPJ_RESTRICT a,
+ OPJ_UINT32 width,
+ OPJ_UINT32 nb_elts_read);
#ifdef __SSE__
-static void opj_v4dwt_decode_step1_sse(opj_v4_t* w, OPJ_INT32 count,
+static void opj_v4dwt_decode_step1_sse(opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
const __m128 c);
-static void opj_v4dwt_decode_step2_sse(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
- OPJ_INT32 m, __m128 c);
+static void opj_v4dwt_decode_step2_sse(opj_v4_t* l, opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
+ OPJ_UINT32 m, __m128 c);
#else
-static void opj_v4dwt_decode_step1(opj_v4_t* w, OPJ_INT32 count,
+static void opj_v4dwt_decode_step1(opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
const OPJ_FLOAT32 c);
-static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
- OPJ_INT32 m, OPJ_FLOAT32 c);
+static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
+ OPJ_UINT32 m,
+ OPJ_FLOAT32 c);
#endif
@@ -202,6 +223,8 @@ static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
/* <summary> */
/* This table contains the norms of the 5-3 wavelets for different bands. */
/* </summary> */
+/* FIXME! the array should really be extended up to 33 resolution levels */
+/* See https://github.com/uclouvain/openjpeg/issues/493 */
static const OPJ_FLOAT64 opj_dwt_norms[4][10] = {
{1.000, 1.500, 2.750, 5.375, 10.68, 21.34, 42.67, 85.33, 170.7, 341.3},
{1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
@@ -212,6 +235,8 @@ static const OPJ_FLOAT64 opj_dwt_norms[4][10] = {
/* <summary> */
/* This table contains the norms of the 9-7 wavelets for different bands. */
/* </summary> */
+/* FIXME! the array should really be extended up to 33 resolution levels */
+/* See https://github.com/uclouvain/openjpeg/issues/493 */
static const OPJ_FLOAT64 opj_dwt_norms_real[4][10] = {
{1.000, 1.965, 4.177, 8.403, 16.90, 33.84, 67.69, 135.3, 270.6, 540.9},
{2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
@@ -265,7 +290,7 @@ static void opj_dwt_deinterleave_v(OPJ_INT32 *a, OPJ_INT32 *b, OPJ_INT32 dn,
l_src += 2;
} /* b[i*x]=a[2*i+cas]; */
- l_dest = b + sn * x;
+ l_dest = b + (OPJ_SIZE_T)sn * (OPJ_SIZE_T)x;
l_src = a + 1 - cas;
i = dn;
@@ -311,7 +336,7 @@ static void opj_dwt_interleave_v(const opj_dwt_t* v, OPJ_INT32 *a, OPJ_INT32 x)
bi += 2;
ai += x;
}
- ai = a + (v->sn * x);
+ ai = a + (v->sn * (OPJ_SIZE_T)x);
bi = v->mem + 1 - v->cas;
i = v->dn ;
while (i--) {
@@ -602,7 +627,7 @@ static
void opj_idwt53_v_final_memcpy(OPJ_INT32* tiledp_col,
const OPJ_INT32* tmp,
OPJ_INT32 len,
- OPJ_INT32 stride)
+ OPJ_SIZE_T stride)
{
OPJ_INT32 i;
for (i = 0; i < len; ++i) {
@@ -611,9 +636,9 @@ void opj_idwt53_v_final_memcpy(OPJ_INT32* tiledp_col,
PARALLEL_COLS_53 * sizeof(OPJ_INT32))
would do but would be a tiny bit slower.
We can take here advantage of our knowledge of alignment */
- STOREU(&tiledp_col[i * stride + 0],
+ STOREU(&tiledp_col[(OPJ_SIZE_T)i * stride + 0],
LOAD(&tmp[PARALLEL_COLS_53 * i + 0]));
- STOREU(&tiledp_col[i * stride + VREG_INT_COUNT],
+ STOREU(&tiledp_col[(OPJ_SIZE_T)i * stride + VREG_INT_COUNT],
LOAD(&tmp[PARALLEL_COLS_53 * i + VREG_INT_COUNT]));
}
}
@@ -625,12 +650,13 @@ static void opj_idwt53_v_cas0_mcols_SSE2_OR_AVX2(
const OPJ_INT32 sn,
const OPJ_INT32 len,
OPJ_INT32* tiledp_col,
- const OPJ_INT32 stride)
+ const OPJ_SIZE_T stride)
{
const OPJ_INT32* in_even = &tiledp_col[0];
- const OPJ_INT32* in_odd = &tiledp_col[sn * stride];
+ const OPJ_INT32* in_odd = &tiledp_col[(OPJ_SIZE_T)sn * stride];
- OPJ_INT32 i, j;
+ OPJ_INT32 i;
+ OPJ_SIZE_T j;
VREG d1c_0, d1n_0, s1n_0, s0c_0, s0n_0;
VREG d1c_1, d1n_1, s1n_1, s0c_1, s0n_1;
const VREG two = LOAD_CST(2);
@@ -647,7 +673,7 @@ static void opj_idwt53_v_cas0_mcols_SSE2_OR_AVX2(
/* Note: loads of input even/odd values must be done in a unaligned */
/* fashion. But stores in tmp can be done with aligned store, since */
/* the temporary buffer is properly aligned */
- assert((size_t)tmp % (sizeof(OPJ_INT32) * VREG_INT_COUNT) == 0);
+ assert((OPJ_SIZE_T)tmp % (sizeof(OPJ_INT32) * VREG_INT_COUNT) == 0);
s1n_0 = LOADU(in_even + 0);
s1n_1 = LOADU(in_even + VREG_INT_COUNT);
@@ -689,7 +715,7 @@ static void opj_idwt53_v_cas0_mcols_SSE2_OR_AVX2(
if (len & 1) {
VREG tmp_len_minus_1;
- s1n_0 = LOADU(in_even + ((len - 1) / 2) * stride);
+ s1n_0 = LOADU(in_even + (OPJ_SIZE_T)((len - 1) / 2) * stride);
/* tmp_len_minus_1 = s1n - ((d1n + 1) >> 1); */
tmp_len_minus_1 = SUB(s1n_0, SAR(ADD3(d1n_0, d1n_0, two), 2));
STORE(tmp + PARALLEL_COLS_53 * (len - 1), tmp_len_minus_1);
@@ -697,7 +723,7 @@ static void opj_idwt53_v_cas0_mcols_SSE2_OR_AVX2(
STORE(tmp + PARALLEL_COLS_53 * (len - 2),
ADD(d1n_0, SAR(ADD(s0n_0, tmp_len_minus_1), 1)));
- s1n_1 = LOADU(in_even + ((len - 1) / 2) * stride + VREG_INT_COUNT);
+ s1n_1 = LOADU(in_even + (OPJ_SIZE_T)((len - 1) / 2) * stride + VREG_INT_COUNT);
/* tmp_len_minus_1 = s1n - ((d1n + 1) >> 1); */
tmp_len_minus_1 = SUB(s1n_1, SAR(ADD3(d1n_1, d1n_1, two), 2));
STORE(tmp + PARALLEL_COLS_53 * (len - 1) + VREG_INT_COUNT,
@@ -725,15 +751,16 @@ static void opj_idwt53_v_cas1_mcols_SSE2_OR_AVX2(
const OPJ_INT32 sn,
const OPJ_INT32 len,
OPJ_INT32* tiledp_col,
- const OPJ_INT32 stride)
+ const OPJ_SIZE_T stride)
{
- OPJ_INT32 i, j;
+ OPJ_INT32 i;
+ OPJ_SIZE_T j;
VREG s1_0, s2_0, dc_0, dn_0;
VREG s1_1, s2_1, dc_1, dn_1;
const VREG two = LOAD_CST(2);
- const OPJ_INT32* in_even = &tiledp_col[sn * stride];
+ const OPJ_INT32* in_even = &tiledp_col[(OPJ_SIZE_T)sn * stride];
const OPJ_INT32* in_odd = &tiledp_col[0];
assert(len > 2);
@@ -748,7 +775,7 @@ static void opj_idwt53_v_cas1_mcols_SSE2_OR_AVX2(
/* Note: loads of input even/odd values must be done in a unaligned */
/* fashion. But stores in tmp can be done with aligned store, since */
/* the temporary buffer is properly aligned */
- assert((size_t)tmp % (sizeof(OPJ_INT32) * VREG_INT_COUNT) == 0);
+ assert((OPJ_SIZE_T)tmp % (sizeof(OPJ_INT32) * VREG_INT_COUNT) == 0);
s1_0 = LOADU(in_even + stride);
/* in_odd[0] - ((in_even[0] + s1 + 2) >> 2); */
@@ -793,9 +820,9 @@ static void opj_idwt53_v_cas1_mcols_SSE2_OR_AVX2(
if (!(len & 1)) {
/*dn = in_odd[(len / 2 - 1) * stride] - ((s1 + 1) >> 1); */
- dn_0 = SUB(LOADU(in_odd + (len / 2 - 1) * stride),
+ dn_0 = SUB(LOADU(in_odd + (OPJ_SIZE_T)(len / 2 - 1) * stride),
SAR(ADD3(s1_0, s1_0, two), 2));
- dn_1 = SUB(LOADU(in_odd + (len / 2 - 1) * stride + VREG_INT_COUNT),
+ dn_1 = SUB(LOADU(in_odd + (OPJ_SIZE_T)(len / 2 - 1) * stride + VREG_INT_COUNT),
SAR(ADD3(s1_1, s1_1, two), 2));
/* tmp[len - 2] = s1 + ((dn + dc) >> 1); */
@@ -835,7 +862,7 @@ static void opj_idwt3_v_cas0(OPJ_INT32* tmp,
const OPJ_INT32 sn,
const OPJ_INT32 len,
OPJ_INT32* tiledp_col,
- const OPJ_INT32 stride)
+ const OPJ_SIZE_T stride)
{
OPJ_INT32 i, j;
OPJ_INT32 d1c, d1n, s1n, s0c, s0n;
@@ -846,15 +873,15 @@ static void opj_idwt3_v_cas0(OPJ_INT32* tmp,
/* accesses and explicit interleaving. */
s1n = tiledp_col[0];
- d1n = tiledp_col[sn * stride];
+ d1n = tiledp_col[(OPJ_SIZE_T)sn * stride];
s0n = s1n - ((d1n + 1) >> 1);
for (i = 0, j = 0; i < (len - 3); i += 2, j++) {
d1c = d1n;
s0c = s0n;
- s1n = tiledp_col[(j + 1) * stride];
- d1n = tiledp_col[(sn + j + 1) * stride];
+ s1n = tiledp_col[(OPJ_SIZE_T)(j + 1) * stride];
+ d1n = tiledp_col[(OPJ_SIZE_T)(sn + j + 1) * stride];
s0n = s1n - ((d1c + d1n + 2) >> 2);
@@ -866,7 +893,7 @@ static void opj_idwt3_v_cas0(OPJ_INT32* tmp,
if (len & 1) {
tmp[len - 1] =
- tiledp_col[((len - 1) / 2) * stride] -
+ tiledp_col[(OPJ_SIZE_T)((len - 1) / 2) * stride] -
((d1n + 1) >> 1);
tmp[len - 2] = d1n + ((s0n + tmp[len - 1]) >> 1);
} else {
@@ -874,7 +901,7 @@ static void opj_idwt3_v_cas0(OPJ_INT32* tmp,
}
for (i = 0; i < len; ++i) {
- tiledp_col[i * stride] = tmp[i];
+ tiledp_col[(OPJ_SIZE_T)i * stride] = tmp[i];
}
}
@@ -885,11 +912,11 @@ static void opj_idwt3_v_cas1(OPJ_INT32* tmp,
const OPJ_INT32 sn,
const OPJ_INT32 len,
OPJ_INT32* tiledp_col,
- const OPJ_INT32 stride)
+ const OPJ_SIZE_T stride)
{
OPJ_INT32 i, j;
OPJ_INT32 s1, s2, dc, dn;
- const OPJ_INT32* in_even = &tiledp_col[sn * stride];
+ const OPJ_INT32* in_even = &tiledp_col[(OPJ_SIZE_T)sn * stride];
const OPJ_INT32* in_odd = &tiledp_col[0];
assert(len > 2);
@@ -902,9 +929,9 @@ static void opj_idwt3_v_cas1(OPJ_INT32* tmp,
tmp[0] = in_even[0] + dc;
for (i = 1, j = 1; i < (len - 2 - !(len & 1)); i += 2, j++) {
- s2 = in_even[(j + 1) * stride];
+ s2 = in_even[(OPJ_SIZE_T)(j + 1) * stride];
- dn = in_odd[j * stride] - ((s1 + s2 + 2) >> 2);
+ dn = in_odd[(OPJ_SIZE_T)j * stride] - ((s1 + s2 + 2) >> 2);
tmp[i ] = dc;
tmp[i + 1] = s1 + ((dn + dc) >> 1);
@@ -913,7 +940,7 @@ static void opj_idwt3_v_cas1(OPJ_INT32* tmp,
}
tmp[i] = dc;
if (!(len & 1)) {
- dn = in_odd[(len / 2 - 1) * stride] - ((s1 + 1) >> 1);
+ dn = in_odd[(OPJ_SIZE_T)(len / 2 - 1) * stride] - ((s1 + 1) >> 1);
tmp[len - 2] = s1 + ((dn + dc) >> 1);
tmp[len - 1] = dn;
} else {
@@ -921,7 +948,7 @@ static void opj_idwt3_v_cas1(OPJ_INT32* tmp,
}
for (i = 0; i < len; ++i) {
- tiledp_col[i * stride] = tmp[i];
+ tiledp_col[(OPJ_SIZE_T)i * stride] = tmp[i];
}
}
#endif /* !defined(STANDARD_SLOW_VERSION) */
@@ -932,7 +959,7 @@ static void opj_idwt3_v_cas1(OPJ_INT32* tmp,
/* Performs interleave, inverse wavelet transform and copy back to buffer */
static void opj_idwt53_v(const opj_dwt_t *dwt,
OPJ_INT32* tiledp_col,
- OPJ_INT32 stride,
+ OPJ_SIZE_T stride,
OPJ_INT32 nb_cols)
{
#ifdef STANDARD_SLOW_VERSION
@@ -980,14 +1007,14 @@ static void opj_idwt53_v(const opj_dwt_t *dwt,
OPJ_INT32* out = dwt->mem;
for (c = 0; c < nb_cols; c++, tiledp_col++) {
OPJ_INT32 i;
- const OPJ_INT32* in_even = &tiledp_col[sn * stride];
+ const OPJ_INT32* in_even = &tiledp_col[(OPJ_SIZE_T)sn * stride];
const OPJ_INT32* in_odd = &tiledp_col[0];
out[1] = in_odd[0] - ((in_even[0] + 1) >> 1);
out[0] = in_even[0] + out[1];
for (i = 0; i < len; ++i) {
- tiledp_col[i * stride] = out[i];
+ tiledp_col[(OPJ_SIZE_T)i * stride] = out[i];
}
}
@@ -1097,8 +1124,8 @@ static INLINE OPJ_BOOL opj_dwt_encode_procedure(const opj_tcd_tilecomp_t * tilec
OPJ_INT32 rw; /* width of the resolution level computed */
OPJ_INT32 rh; /* height of the resolution level computed */
- size_t l_data_count;
- size_t l_data_size;
+ OPJ_SIZE_T l_data_count;
+ OPJ_SIZE_T l_data_size;
opj_tcd_resolution_t * l_cur_res = 0;
opj_tcd_resolution_t * l_last_res = 0;
@@ -1184,10 +1211,14 @@ OPJ_BOOL opj_dwt_encode(opj_tcd_tilecomp_t * tilec)
/* <summary> */
/* Inverse 5-3 wavelet transform in 2-D. */
/* </summary> */
-OPJ_BOOL opj_dwt_decode(opj_thread_pool_t* tp, opj_tcd_tilecomp_t* tilec,
+OPJ_BOOL opj_dwt_decode(opj_tcd_t *p_tcd, opj_tcd_tilecomp_t* tilec,
OPJ_UINT32 numres)
{
- return opj_dwt_decode_tile(tp, tilec, numres);
+ if (p_tcd->whole_tile_decoding) {
+ return opj_dwt_decode_tile(p_tcd->thread_pool, tilec, numres);
+ } else {
+ return opj_dwt_decode_partial_tile(tilec, numres);
+ }
}
@@ -1210,6 +1241,14 @@ OPJ_UINT32 opj_dwt_getgain(OPJ_UINT32 orient)
/* </summary> */
OPJ_FLOAT64 opj_dwt_getnorm(OPJ_UINT32 level, OPJ_UINT32 orient)
{
+ /* FIXME ! This is just a band-aid to avoid a buffer overflow */
+ /* but the array should really be extended up to 33 resolution levels */
+ /* See https://github.com/uclouvain/openjpeg/issues/493 */
+ if (orient == 0 && level >= 10) {
+ level = 9;
+ } else if (orient > 0 && level >= 9) {
+ level = 8;
+ }
return opj_dwt_norms[orient][level];
}
@@ -1235,6 +1274,14 @@ OPJ_UINT32 opj_dwt_getgain_real(OPJ_UINT32 orient)
/* </summary> */
OPJ_FLOAT64 opj_dwt_getnorm_real(OPJ_UINT32 level, OPJ_UINT32 orient)
{
+ /* FIXME ! This is just a band-aid to avoid a buffer overflow */
+ /* but the array should really be extended up to 33 resolution levels */
+ /* See https://github.com/uclouvain/openjpeg/issues/493 */
+ if (orient == 0 && level >= 10) {
+ level = 9;
+ } else if (orient > 0 && level >= 9) {
+ level = 8;
+ }
return opj_dwt_norms_real[orient][level];
}
@@ -1324,11 +1371,11 @@ static void opj_dwt_decode_v_func(void* user_data, opj_tls_t* tls)
job = (opj_dwd_decode_v_job_t*)user_data;
for (j = job->min_j; j + PARALLEL_COLS_53 <= job->max_j;
j += PARALLEL_COLS_53) {
- opj_idwt53_v(&job->v, &job->tiledp[j], (OPJ_INT32)job->w,
+ opj_idwt53_v(&job->v, &job->tiledp[j], (OPJ_SIZE_T)job->w,
PARALLEL_COLS_53);
}
if (j < job->max_j)
- opj_idwt53_v(&job->v, &job->tiledp[j], (OPJ_INT32)job->w,
+ opj_idwt53_v(&job->v, &job->tiledp[j], (OPJ_SIZE_T)job->w,
(OPJ_INT32)(job->max_j - j));
opj_aligned_free(job->v.mem);
@@ -1352,8 +1399,10 @@ static OPJ_BOOL opj_dwt_decode_tile(opj_thread_pool_t* tp,
OPJ_UINT32 rh = (OPJ_UINT32)(tr->y1 -
tr->y0); /* height of the resolution level computed */
- OPJ_UINT32 w = (OPJ_UINT32)(tilec->x1 - tilec->x0);
- size_t h_mem_size;
+ OPJ_UINT32 w = (OPJ_UINT32)(tilec->resolutions[tilec->minimum_num_resolutions -
+ 1].x1 -
+ tilec->resolutions[tilec->minimum_num_resolutions - 1].x0);
+ OPJ_SIZE_T h_mem_size;
int num_threads;
if (numres == 1U) {
@@ -1395,7 +1444,7 @@ static OPJ_BOOL opj_dwt_decode_tile(opj_thread_pool_t* tp,
if (num_threads <= 1 || rh <= 1) {
for (j = 0; j < rh; ++j) {
- opj_idwt53_h(&h, &tiledp[j * w]);
+ opj_idwt53_h(&h, &tiledp[(OPJ_SIZE_T)j * w]);
}
} else {
OPJ_UINT32 num_jobs = (OPJ_UINT32)num_threads;
@@ -1447,10 +1496,10 @@ static OPJ_BOOL opj_dwt_decode_tile(opj_thread_pool_t* tp,
if (num_threads <= 1 || rw <= 1) {
for (j = 0; j + PARALLEL_COLS_53 <= rw;
j += PARALLEL_COLS_53) {
- opj_idwt53_v(&v, &tiledp[j], (OPJ_INT32)w, PARALLEL_COLS_53);
+ opj_idwt53_v(&v, &tiledp[j], (OPJ_SIZE_T)w, PARALLEL_COLS_53);
}
if (j < rw) {
- opj_idwt53_v(&v, &tiledp[j], (OPJ_INT32)w, (OPJ_INT32)(rw - j));
+ opj_idwt53_v(&v, &tiledp[j], (OPJ_SIZE_T)w, (OPJ_INT32)(rw - j));
}
} else {
OPJ_UINT32 num_jobs = (OPJ_UINT32)num_threads;
@@ -1500,136 +1549,849 @@ static OPJ_BOOL opj_dwt_decode_tile(opj_thread_pool_t* tp,
return OPJ_TRUE;
}
-static void opj_v4dwt_interleave_h(opj_v4dwt_t* OPJ_RESTRICT w,
- OPJ_FLOAT32* OPJ_RESTRICT a, OPJ_INT32 x, OPJ_INT32 size)
+static void opj_dwt_interleave_partial_h(OPJ_INT32 *dest,
+ OPJ_INT32 cas,
+ opj_sparse_array_int32_t* sa,
+ OPJ_UINT32 sa_line,
+ OPJ_UINT32 sn,
+ OPJ_UINT32 win_l_x0,
+ OPJ_UINT32 win_l_x1,
+ OPJ_UINT32 win_h_x0,
+ OPJ_UINT32 win_h_x1)
+{
+ OPJ_BOOL ret;
+ ret = opj_sparse_array_int32_read(sa,
+ win_l_x0, sa_line,
+ win_l_x1, sa_line + 1,
+ dest + cas + 2 * win_l_x0,
+ 2, 0, OPJ_TRUE);
+ assert(ret);
+ ret = opj_sparse_array_int32_read(sa,
+ sn + win_h_x0, sa_line,
+ sn + win_h_x1, sa_line + 1,
+ dest + 1 - cas + 2 * win_h_x0,
+ 2, 0, OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+}
+
+
+static void opj_dwt_interleave_partial_v(OPJ_INT32 *dest,
+ OPJ_INT32 cas,
+ opj_sparse_array_int32_t* sa,
+ OPJ_UINT32 sa_col,
+ OPJ_UINT32 nb_cols,
+ OPJ_UINT32 sn,
+ OPJ_UINT32 win_l_y0,
+ OPJ_UINT32 win_l_y1,
+ OPJ_UINT32 win_h_y0,
+ OPJ_UINT32 win_h_y1)
+{
+ OPJ_BOOL ret;
+ ret = opj_sparse_array_int32_read(sa,
+ sa_col, win_l_y0,
+ sa_col + nb_cols, win_l_y1,
+ dest + cas * 4 + 2 * 4 * win_l_y0,
+ 1, 2 * 4, OPJ_TRUE);
+ assert(ret);
+ ret = opj_sparse_array_int32_read(sa,
+ sa_col, sn + win_h_y0,
+ sa_col + nb_cols, sn + win_h_y1,
+ dest + (1 - cas) * 4 + 2 * 4 * win_h_y0,
+ 1, 2 * 4, OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+}
+
+static void opj_dwt_decode_partial_1(OPJ_INT32 *a, OPJ_SIZE_T a_count,
+ OPJ_INT32 dn, OPJ_INT32 sn,
+ OPJ_INT32 cas,
+ OPJ_INT32 win_l_x0,
+ OPJ_INT32 win_l_x1,
+ OPJ_INT32 win_h_x0,
+ OPJ_INT32 win_h_x1)
+{
+ OPJ_INT32 i;
+
+ if (!cas) {
+ if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
+
+ /* Naive version is :
+ for (i = win_l_x0; i < i_max; i++) {
+ OPJ_S(i) -= (OPJ_D_(i - 1) + OPJ_D_(i) + 2) >> 2;
+ }
+ for (i = win_h_x0; i < win_h_x1; i++) {
+ OPJ_D(i) += (OPJ_S_(i) + OPJ_S_(i + 1)) >> 1;
+ }
+ but the compiler doesn't manage to unroll it to avoid bound
+ checking in OPJ_S_ and OPJ_D_ macros
+ */
+
+ i = win_l_x0;
+ if (i < win_l_x1) {
+ OPJ_INT32 i_max;
+
+ /* Left-most case */
+ OPJ_S(i) -= (OPJ_D_(i - 1) + OPJ_D_(i) + 2) >> 2;
+ i ++;
+
+ i_max = win_l_x1;
+ if (i_max > dn) {
+ i_max = dn;
+ }
+ for (; i < i_max; i++) {
+ /* No bound checking */
+ OPJ_S(i) -= (OPJ_D(i - 1) + OPJ_D(i) + 2) >> 2;
+ }
+ for (; i < win_l_x1; i++) {
+ /* Right-most case */
+ OPJ_S(i) -= (OPJ_D_(i - 1) + OPJ_D_(i) + 2) >> 2;
+ }
+ }
+
+ i = win_h_x0;
+ if (i < win_h_x1) {
+ OPJ_INT32 i_max = win_h_x1;
+ if (i_max >= sn) {
+ i_max = sn - 1;
+ }
+ for (; i < i_max; i++) {
+ /* No bound checking */
+ OPJ_D(i) += (OPJ_S(i) + OPJ_S(i + 1)) >> 1;
+ }
+ for (; i < win_h_x1; i++) {
+ /* Right-most case */
+ OPJ_D(i) += (OPJ_S_(i) + OPJ_S_(i + 1)) >> 1;
+ }
+ }
+ }
+ } else {
+ if (!sn && dn == 1) { /* NEW : CASE ONE ELEMENT */
+ OPJ_S(0) /= 2;
+ } else {
+ for (i = win_l_x0; i < win_l_x1; i++) {
+ OPJ_D(i) -= (OPJ_SS_(i) + OPJ_SS_(i + 1) + 2) >> 2;
+ }
+ for (i = win_h_x0; i < win_h_x1; i++) {
+ OPJ_S(i) += (OPJ_DD_(i) + OPJ_DD_(i - 1)) >> 1;
+ }
+ }
+ }
+}
+
+#define OPJ_S_off(i,off) a[(OPJ_UINT32)(i)*2*4+off]
+#define OPJ_D_off(i,off) a[(1+(OPJ_UINT32)(i)*2)*4+off]
+#define OPJ_S__off(i,off) ((i)<0?OPJ_S_off(0,off):((i)>=sn?OPJ_S_off(sn-1,off):OPJ_S_off(i,off)))
+#define OPJ_D__off(i,off) ((i)<0?OPJ_D_off(0,off):((i)>=dn?OPJ_D_off(dn-1,off):OPJ_D_off(i,off)))
+#define OPJ_SS__off(i,off) ((i)<0?OPJ_S_off(0,off):((i)>=dn?OPJ_S_off(dn-1,off):OPJ_S_off(i,off)))
+#define OPJ_DD__off(i,off) ((i)<0?OPJ_D_off(0,off):((i)>=sn?OPJ_D_off(sn-1,off):OPJ_D_off(i,off)))
+
+static void opj_dwt_decode_partial_1_parallel(OPJ_INT32 *a,
+ OPJ_UINT32 nb_cols,
+ OPJ_INT32 dn, OPJ_INT32 sn,
+ OPJ_INT32 cas,
+ OPJ_INT32 win_l_x0,
+ OPJ_INT32 win_l_x1,
+ OPJ_INT32 win_h_x0,
+ OPJ_INT32 win_h_x1)
{
- OPJ_FLOAT32* OPJ_RESTRICT bi = (OPJ_FLOAT32*)(w->wavelet + w->cas);
- OPJ_INT32 count = w->sn;
- OPJ_INT32 i, k;
+ OPJ_INT32 i;
+ OPJ_UINT32 off;
+
+ (void)nb_cols;
+
+ if (!cas) {
+ if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
+
+ /* Naive version is :
+ for (i = win_l_x0; i < i_max; i++) {
+ OPJ_S(i) -= (OPJ_D_(i - 1) + OPJ_D_(i) + 2) >> 2;
+ }
+ for (i = win_h_x0; i < win_h_x1; i++) {
+ OPJ_D(i) += (OPJ_S_(i) + OPJ_S_(i + 1)) >> 1;
+ }
+ but the compiler doesn't manage to unroll it to avoid bound
+ checking in OPJ_S_ and OPJ_D_ macros
+ */
+
+ i = win_l_x0;
+ if (i < win_l_x1) {
+ OPJ_INT32 i_max;
+
+ /* Left-most case */
+ for (off = 0; off < 4; off++) {
+ OPJ_S_off(i, off) -= (OPJ_D__off(i - 1, off) + OPJ_D__off(i, off) + 2) >> 2;
+ }
+ i ++;
+
+ i_max = win_l_x1;
+ if (i_max > dn) {
+ i_max = dn;
+ }
+
+#ifdef __SSE2__
+ if (i + 1 < i_max) {
+ const __m128i two = _mm_set1_epi32(2);
+ __m128i Dm1 = _mm_load_si128((__m128i * const)(a + 4 + (i - 1) * 8));
+ for (; i + 1 < i_max; i += 2) {
+ /* No bound checking */
+ __m128i S = _mm_load_si128((__m128i * const)(a + i * 8));
+ __m128i D = _mm_load_si128((__m128i * const)(a + 4 + i * 8));
+ __m128i S1 = _mm_load_si128((__m128i * const)(a + (i + 1) * 8));
+ __m128i D1 = _mm_load_si128((__m128i * const)(a + 4 + (i + 1) * 8));
+ S = _mm_sub_epi32(S,
+ _mm_srai_epi32(_mm_add_epi32(_mm_add_epi32(Dm1, D), two), 2));
+ S1 = _mm_sub_epi32(S1,
+ _mm_srai_epi32(_mm_add_epi32(_mm_add_epi32(D, D1), two), 2));
+ _mm_store_si128((__m128i*)(a + i * 8), S);
+ _mm_store_si128((__m128i*)(a + (i + 1) * 8), S1);
+ Dm1 = D1;
+ }
+ }
+#endif
+
+ for (; i < i_max; i++) {
+ /* No bound checking */
+ for (off = 0; off < 4; off++) {
+ OPJ_S_off(i, off) -= (OPJ_D_off(i - 1, off) + OPJ_D_off(i, off) + 2) >> 2;
+ }
+ }
+ for (; i < win_l_x1; i++) {
+ /* Right-most case */
+ for (off = 0; off < 4; off++) {
+ OPJ_S_off(i, off) -= (OPJ_D__off(i - 1, off) + OPJ_D__off(i, off) + 2) >> 2;
+ }
+ }
+ }
+
+ i = win_h_x0;
+ if (i < win_h_x1) {
+ OPJ_INT32 i_max = win_h_x1;
+ if (i_max >= sn) {
+ i_max = sn - 1;
+ }
+
+#ifdef __SSE2__
+ if (i + 1 < i_max) {
+ __m128i S = _mm_load_si128((__m128i * const)(a + i * 8));
+ for (; i + 1 < i_max; i += 2) {
+ /* No bound checking */
+ __m128i D = _mm_load_si128((__m128i * const)(a + 4 + i * 8));
+ __m128i S1 = _mm_load_si128((__m128i * const)(a + (i + 1) * 8));
+ __m128i D1 = _mm_load_si128((__m128i * const)(a + 4 + (i + 1) * 8));
+ __m128i S2 = _mm_load_si128((__m128i * const)(a + (i + 2) * 8));
+ D = _mm_add_epi32(D, _mm_srai_epi32(_mm_add_epi32(S, S1), 1));
+ D1 = _mm_add_epi32(D1, _mm_srai_epi32(_mm_add_epi32(S1, S2), 1));
+ _mm_store_si128((__m128i*)(a + 4 + i * 8), D);
+ _mm_store_si128((__m128i*)(a + 4 + (i + 1) * 8), D1);
+ S = S2;
+ }
+ }
+#endif
+
+ for (; i < i_max; i++) {
+ /* No bound checking */
+ for (off = 0; off < 4; off++) {
+ OPJ_D_off(i, off) += (OPJ_S_off(i, off) + OPJ_S_off(i + 1, off)) >> 1;
+ }
+ }
+ for (; i < win_h_x1; i++) {
+ /* Right-most case */
+ for (off = 0; off < 4; off++) {
+ OPJ_D_off(i, off) += (OPJ_S__off(i, off) + OPJ_S__off(i + 1, off)) >> 1;
+ }
+ }
+ }
+ }
+ } else {
+ if (!sn && dn == 1) { /* NEW : CASE ONE ELEMENT */
+ for (off = 0; off < 4; off++) {
+ OPJ_S_off(0, off) /= 2;
+ }
+ } else {
+ for (i = win_l_x0; i < win_l_x1; i++) {
+ for (off = 0; off < 4; off++) {
+ OPJ_D_off(i, off) -= (OPJ_SS__off(i, off) + OPJ_SS__off(i + 1, off) + 2) >> 2;
+ }
+ }
+ for (i = win_h_x0; i < win_h_x1; i++) {
+ for (off = 0; off < 4; off++) {
+ OPJ_S_off(i, off) += (OPJ_DD__off(i, off) + OPJ_DD__off(i - 1, off)) >> 1;
+ }
+ }
+ }
+ }
+}
+
+static void opj_dwt_get_band_coordinates(opj_tcd_tilecomp_t* tilec,
+ OPJ_UINT32 resno,
+ OPJ_UINT32 bandno,
+ OPJ_UINT32 tcx0,
+ OPJ_UINT32 tcy0,
+ OPJ_UINT32 tcx1,
+ OPJ_UINT32 tcy1,
+ OPJ_UINT32* tbx0,
+ OPJ_UINT32* tby0,
+ OPJ_UINT32* tbx1,
+ OPJ_UINT32* tby1)
+{
+ /* Compute number of decomposition for this band. See table F-1 */
+ OPJ_UINT32 nb = (resno == 0) ?
+ tilec->numresolutions - 1 :
+ tilec->numresolutions - resno;
+ /* Map above tile-based coordinates to sub-band-based coordinates per */
+ /* equation B-15 of the standard */
+ OPJ_UINT32 x0b = bandno & 1;
+ OPJ_UINT32 y0b = bandno >> 1;
+ if (tbx0) {
+ *tbx0 = (nb == 0) ? tcx0 :
+ (tcx0 <= (1U << (nb - 1)) * x0b) ? 0 :
+ opj_uint_ceildivpow2(tcx0 - (1U << (nb - 1)) * x0b, nb);
+ }
+ if (tby0) {
+ *tby0 = (nb == 0) ? tcy0 :
+ (tcy0 <= (1U << (nb - 1)) * y0b) ? 0 :
+ opj_uint_ceildivpow2(tcy0 - (1U << (nb - 1)) * y0b, nb);
+ }
+ if (tbx1) {
+ *tbx1 = (nb == 0) ? tcx1 :
+ (tcx1 <= (1U << (nb - 1)) * x0b) ? 0 :
+ opj_uint_ceildivpow2(tcx1 - (1U << (nb - 1)) * x0b, nb);
+ }
+ if (tby1) {
+ *tby1 = (nb == 0) ? tcy1 :
+ (tcy1 <= (1U << (nb - 1)) * y0b) ? 0 :
+ opj_uint_ceildivpow2(tcy1 - (1U << (nb - 1)) * y0b, nb);
+ }
+}
+
+static void opj_dwt_segment_grow(OPJ_UINT32 filter_width,
+ OPJ_UINT32 max_size,
+ OPJ_UINT32* start,
+ OPJ_UINT32* end)
+{
+ *start = opj_uint_subs(*start, filter_width);
+ *end = opj_uint_adds(*end, filter_width);
+ *end = opj_uint_min(*end, max_size);
+}
+
+
+static opj_sparse_array_int32_t* opj_dwt_init_sparse_array(
+ opj_tcd_tilecomp_t* tilec,
+ OPJ_UINT32 numres)
+{
+ opj_tcd_resolution_t* tr_max = &(tilec->resolutions[numres - 1]);
+ OPJ_UINT32 w = (OPJ_UINT32)(tr_max->x1 - tr_max->x0);
+ OPJ_UINT32 h = (OPJ_UINT32)(tr_max->y1 - tr_max->y0);
+ OPJ_UINT32 resno, bandno, precno, cblkno;
+ opj_sparse_array_int32_t* sa = opj_sparse_array_int32_create(
+ w, h, opj_uint_min(w, 64), opj_uint_min(h, 64));
+ if (sa == NULL) {
+ return NULL;
+ }
+
+ for (resno = 0; resno < numres; ++resno) {
+ opj_tcd_resolution_t* res = &tilec->resolutions[resno];
+
+ for (bandno = 0; bandno < res->numbands; ++bandno) {
+ opj_tcd_band_t* band = &res->bands[bandno];
+
+ for (precno = 0; precno < res->pw * res->ph; ++precno) {
+ opj_tcd_precinct_t* precinct = &band->precincts[precno];
+ for (cblkno = 0; cblkno < precinct->cw * precinct->ch; ++cblkno) {
+ opj_tcd_cblk_dec_t* cblk = &precinct->cblks.dec[cblkno];
+ if (cblk->decoded_data != NULL) {
+ OPJ_UINT32 x = (OPJ_UINT32)(cblk->x0 - band->x0);
+ OPJ_UINT32 y = (OPJ_UINT32)(cblk->y0 - band->y0);
+ OPJ_UINT32 cblk_w = (OPJ_UINT32)(cblk->x1 - cblk->x0);
+ OPJ_UINT32 cblk_h = (OPJ_UINT32)(cblk->y1 - cblk->y0);
+
+ if (band->bandno & 1) {
+ opj_tcd_resolution_t* pres = &tilec->resolutions[resno - 1];
+ x += (OPJ_UINT32)(pres->x1 - pres->x0);
+ }
+ if (band->bandno & 2) {
+ opj_tcd_resolution_t* pres = &tilec->resolutions[resno - 1];
+ y += (OPJ_UINT32)(pres->y1 - pres->y0);
+ }
+
+ if (!opj_sparse_array_int32_write(sa, x, y,
+ x + cblk_w, y + cblk_h,
+ cblk->decoded_data,
+ 1, cblk_w, OPJ_TRUE)) {
+ opj_sparse_array_int32_free(sa);
+ return NULL;
+ }
+ }
+ }
+ }
+ }
+ }
+
+ return sa;
+}
+
+
+static OPJ_BOOL opj_dwt_decode_partial_tile(
+ opj_tcd_tilecomp_t* tilec,
+ OPJ_UINT32 numres)
+{
+ opj_sparse_array_int32_t* sa;
+ opj_dwt_t h;
+ opj_dwt_t v;
+ OPJ_UINT32 resno;
+ /* This value matches the maximum left/right extension given in tables */
+ /* F.2 and F.3 of the standard. */
+ const OPJ_UINT32 filter_width = 2U;
+
+ opj_tcd_resolution_t* tr = tilec->resolutions;
+ opj_tcd_resolution_t* tr_max = &(tilec->resolutions[numres - 1]);
+
+ OPJ_UINT32 rw = (OPJ_UINT32)(tr->x1 -
+ tr->x0); /* width of the resolution level computed */
+ OPJ_UINT32 rh = (OPJ_UINT32)(tr->y1 -
+ tr->y0); /* height of the resolution level computed */
+
+ OPJ_SIZE_T h_mem_size;
+
+ /* Compute the intersection of the area of interest, expressed in tile coordinates */
+ /* with the tile coordinates */
+ OPJ_UINT32 win_tcx0 = tilec->win_x0;
+ OPJ_UINT32 win_tcy0 = tilec->win_y0;
+ OPJ_UINT32 win_tcx1 = tilec->win_x1;
+ OPJ_UINT32 win_tcy1 = tilec->win_y1;
+
+ if (tr_max->x0 == tr_max->x1 || tr_max->y0 == tr_max->y1) {
+ return OPJ_TRUE;
+ }
+
+ sa = opj_dwt_init_sparse_array(tilec, numres);
+ if (sa == NULL) {
+ return OPJ_FALSE;
+ }
+
+ if (numres == 1U) {
+ OPJ_BOOL ret = opj_sparse_array_int32_read(sa,
+ tr_max->win_x0 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y0 - (OPJ_UINT32)tr_max->y0,
+ tr_max->win_x1 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y1 - (OPJ_UINT32)tr_max->y0,
+ tilec->data_win,
+ 1, tr_max->win_x1 - tr_max->win_x0,
+ OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+ opj_sparse_array_int32_free(sa);
+ return OPJ_TRUE;
+ }
+ h.mem_count = opj_dwt_max_resolution(tr, numres);
+ /* overflow check */
+ /* in vertical pass, we process 4 columns at a time */
+ if (h.mem_count > (SIZE_MAX / (4 * sizeof(OPJ_INT32)))) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ return OPJ_FALSE;
+ }
+
+ h_mem_size = h.mem_count * 4 * sizeof(OPJ_INT32);
+ h.mem = (OPJ_INT32*)opj_aligned_32_malloc(h_mem_size);
+ if (! h.mem) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ return OPJ_FALSE;
+ }
+
+ v.mem_count = h.mem_count;
+ v.mem = h.mem;
+
+ for (resno = 1; resno < numres; resno ++) {
+ OPJ_UINT32 i, j;
+ /* Window of interest subband-based coordinates */
+ OPJ_UINT32 win_ll_x0, win_ll_y0, win_ll_x1, win_ll_y1;
+ OPJ_UINT32 win_hl_x0, win_hl_x1;
+ OPJ_UINT32 win_lh_y0, win_lh_y1;
+ /* Window of interest tile-resolution-based coordinates */
+ OPJ_UINT32 win_tr_x0, win_tr_x1, win_tr_y0, win_tr_y1;
+ /* Tile-resolution subband-based coordinates */
+ OPJ_UINT32 tr_ll_x0, tr_ll_y0, tr_hl_x0, tr_lh_y0;
+
+ ++tr;
+
+ h.sn = (OPJ_INT32)rw;
+ v.sn = (OPJ_INT32)rh;
+
+ rw = (OPJ_UINT32)(tr->x1 - tr->x0);
+ rh = (OPJ_UINT32)(tr->y1 - tr->y0);
+
+ h.dn = (OPJ_INT32)(rw - (OPJ_UINT32)h.sn);
+ h.cas = tr->x0 % 2;
+
+ v.dn = (OPJ_INT32)(rh - (OPJ_UINT32)v.sn);
+ v.cas = tr->y0 % 2;
+
+ /* Get the subband coordinates for the window of interest */
+ /* LL band */
+ opj_dwt_get_band_coordinates(tilec, resno, 0,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ &win_ll_x0, &win_ll_y0,
+ &win_ll_x1, &win_ll_y1);
+
+ /* HL band */
+ opj_dwt_get_band_coordinates(tilec, resno, 1,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ &win_hl_x0, NULL, &win_hl_x1, NULL);
+
+ /* LH band */
+ opj_dwt_get_band_coordinates(tilec, resno, 2,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ NULL, &win_lh_y0, NULL, &win_lh_y1);
+
+ /* Beware: band index for non-LL0 resolution are 0=HL, 1=LH and 2=HH */
+ tr_ll_x0 = (OPJ_UINT32)tr->bands[1].x0;
+ tr_ll_y0 = (OPJ_UINT32)tr->bands[0].y0;
+ tr_hl_x0 = (OPJ_UINT32)tr->bands[0].x0;
+ tr_lh_y0 = (OPJ_UINT32)tr->bands[1].y0;
+
+ /* Substract the origin of the bands for this tile, to the subwindow */
+ /* of interest band coordinates, so as to get them relative to the */
+ /* tile */
+ win_ll_x0 = opj_uint_subs(win_ll_x0, tr_ll_x0);
+ win_ll_y0 = opj_uint_subs(win_ll_y0, tr_ll_y0);
+ win_ll_x1 = opj_uint_subs(win_ll_x1, tr_ll_x0);
+ win_ll_y1 = opj_uint_subs(win_ll_y1, tr_ll_y0);
+ win_hl_x0 = opj_uint_subs(win_hl_x0, tr_hl_x0);
+ win_hl_x1 = opj_uint_subs(win_hl_x1, tr_hl_x0);
+ win_lh_y0 = opj_uint_subs(win_lh_y0, tr_lh_y0);
+ win_lh_y1 = opj_uint_subs(win_lh_y1, tr_lh_y0);
+
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)h.sn, &win_ll_x0, &win_ll_x1);
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)h.dn, &win_hl_x0, &win_hl_x1);
+
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)v.sn, &win_ll_y0, &win_ll_y1);
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)v.dn, &win_lh_y0, &win_lh_y1);
+
+ /* Compute the tile-resolution-based coordinates for the window of interest */
+ if (h.cas == 0) {
+ win_tr_x0 = opj_uint_min(2 * win_ll_x0, 2 * win_hl_x0 + 1);
+ win_tr_x1 = opj_uint_min(opj_uint_max(2 * win_ll_x1, 2 * win_hl_x1 + 1), rw);
+ } else {
+ win_tr_x0 = opj_uint_min(2 * win_hl_x0, 2 * win_ll_x0 + 1);
+ win_tr_x1 = opj_uint_min(opj_uint_max(2 * win_hl_x1, 2 * win_ll_x1 + 1), rw);
+ }
+
+ if (v.cas == 0) {
+ win_tr_y0 = opj_uint_min(2 * win_ll_y0, 2 * win_lh_y0 + 1);
+ win_tr_y1 = opj_uint_min(opj_uint_max(2 * win_ll_y1, 2 * win_lh_y1 + 1), rh);
+ } else {
+ win_tr_y0 = opj_uint_min(2 * win_lh_y0, 2 * win_ll_y0 + 1);
+ win_tr_y1 = opj_uint_min(opj_uint_max(2 * win_lh_y1, 2 * win_ll_y1 + 1), rh);
+ }
+
+ for (j = 0; j < rh; ++j) {
+ if ((j >= win_ll_y0 && j < win_ll_y1) ||
+ (j >= win_lh_y0 + (OPJ_UINT32)v.sn && j < win_lh_y1 + (OPJ_UINT32)v.sn)) {
+
+ /* Avoids dwt.c:1584:44 (in opj_dwt_decode_partial_1): runtime error: */
+ /* signed integer overflow: -1094795586 + -1094795586 cannot be represented in type 'int' */
+ /* on opj_decompress -i ../../openjpeg/MAPA.jp2 -o out.tif -d 0,0,256,256 */
+ /* This is less extreme than memsetting the whole buffer to 0 */
+ /* although we could potentially do better with better handling of edge conditions */
+ if (win_tr_x1 >= 1 && win_tr_x1 < rw) {
+ h.mem[win_tr_x1 - 1] = 0;
+ }
+ if (win_tr_x1 < rw) {
+ h.mem[win_tr_x1] = 0;
+ }
+
+ opj_dwt_interleave_partial_h(h.mem,
+ h.cas,
+ sa,
+ j,
+ (OPJ_UINT32)h.sn,
+ win_ll_x0,
+ win_ll_x1,
+ win_hl_x0,
+ win_hl_x1);
+ opj_dwt_decode_partial_1(h.mem, h.mem_count, h.dn, h.sn, h.cas,
+ (OPJ_INT32)win_ll_x0,
+ (OPJ_INT32)win_ll_x1,
+ (OPJ_INT32)win_hl_x0,
+ (OPJ_INT32)win_hl_x1);
+ if (!opj_sparse_array_int32_write(sa,
+ win_tr_x0, j,
+ win_tr_x1, j + 1,
+ h.mem + win_tr_x0,
+ 1, 0, OPJ_TRUE)) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ opj_aligned_free(h.mem);
+ return OPJ_FALSE;
+ }
+ }
+ }
+
+ for (i = win_tr_x0; i < win_tr_x1;) {
+ OPJ_UINT32 nb_cols = opj_uint_min(4U, win_tr_x1 - i);
+ opj_dwt_interleave_partial_v(v.mem,
+ v.cas,
+ sa,
+ i,
+ nb_cols,
+ (OPJ_UINT32)v.sn,
+ win_ll_y0,
+ win_ll_y1,
+ win_lh_y0,
+ win_lh_y1);
+ opj_dwt_decode_partial_1_parallel(v.mem, nb_cols, v.dn, v.sn, v.cas,
+ (OPJ_INT32)win_ll_y0,
+ (OPJ_INT32)win_ll_y1,
+ (OPJ_INT32)win_lh_y0,
+ (OPJ_INT32)win_lh_y1);
+ if (!opj_sparse_array_int32_write(sa,
+ i, win_tr_y0,
+ i + nb_cols, win_tr_y1,
+ v.mem + 4 * win_tr_y0,
+ 1, 4, OPJ_TRUE)) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ opj_aligned_free(h.mem);
+ return OPJ_FALSE;
+ }
+
+ i += nb_cols;
+ }
+ }
+ opj_aligned_free(h.mem);
+
+ {
+ OPJ_BOOL ret = opj_sparse_array_int32_read(sa,
+ tr_max->win_x0 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y0 - (OPJ_UINT32)tr_max->y0,
+ tr_max->win_x1 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y1 - (OPJ_UINT32)tr_max->y0,
+ tilec->data_win,
+ 1, tr_max->win_x1 - tr_max->win_x0,
+ OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+ }
+ opj_sparse_array_int32_free(sa);
+ return OPJ_TRUE;
+}
+
+static void opj_v4dwt_interleave_h(opj_v4dwt_t* OPJ_RESTRICT dwt,
+ OPJ_FLOAT32* OPJ_RESTRICT a,
+ OPJ_UINT32 width,
+ OPJ_UINT32 remaining_height)
+{
+ OPJ_FLOAT32* OPJ_RESTRICT bi = (OPJ_FLOAT32*)(dwt->wavelet + dwt->cas);
+ OPJ_UINT32 i, k;
+ OPJ_UINT32 x0 = dwt->win_l_x0;
+ OPJ_UINT32 x1 = dwt->win_l_x1;
for (k = 0; k < 2; ++k) {
- if (count + 3 * x < size && ((size_t) a & 0x0f) == 0 &&
- ((size_t) bi & 0x0f) == 0 && (x & 0x0f) == 0) {
+ if (remaining_height >= 4 && ((OPJ_SIZE_T) a & 0x0f) == 0 &&
+ ((OPJ_SIZE_T) bi & 0x0f) == 0 && (width & 0x0f) == 0) {
/* Fast code path */
- for (i = 0; i < count; ++i) {
- OPJ_INT32 j = i;
+ for (i = x0; i < x1; ++i) {
+ OPJ_UINT32 j = i;
bi[i * 8 ] = a[j];
- j += x;
+ j += width;
bi[i * 8 + 1] = a[j];
- j += x;
+ j += width;
bi[i * 8 + 2] = a[j];
- j += x;
+ j += width;
bi[i * 8 + 3] = a[j];
}
} else {
/* Slow code path */
- for (i = 0; i < count; ++i) {
- OPJ_INT32 j = i;
+ for (i = x0; i < x1; ++i) {
+ OPJ_UINT32 j = i;
bi[i * 8 ] = a[j];
- j += x;
- if (j >= size) {
+ j += width;
+ if (remaining_height == 1) {
continue;
}
bi[i * 8 + 1] = a[j];
- j += x;
- if (j >= size) {
+ j += width;
+ if (remaining_height == 2) {
continue;
}
bi[i * 8 + 2] = a[j];
- j += x;
- if (j >= size) {
+ j += width;
+ if (remaining_height == 3) {
continue;
}
bi[i * 8 + 3] = a[j]; /* This one*/
}
}
- bi = (OPJ_FLOAT32*)(w->wavelet + 1 - w->cas);
- a += w->sn;
- size -= w->sn;
- count = w->dn;
+ bi = (OPJ_FLOAT32*)(dwt->wavelet + 1 - dwt->cas);
+ a += dwt->sn;
+ x0 = dwt->win_h_x0;
+ x1 = dwt->win_h_x1;
}
}
-static void opj_v4dwt_interleave_v(opj_v4dwt_t* OPJ_RESTRICT v,
- OPJ_FLOAT32* OPJ_RESTRICT a, OPJ_INT32 x, OPJ_INT32 nb_elts_read)
+static void opj_v4dwt_interleave_partial_h(opj_v4dwt_t* dwt,
+ opj_sparse_array_int32_t* sa,
+ OPJ_UINT32 sa_line,
+ OPJ_UINT32 remaining_height)
{
- opj_v4_t* OPJ_RESTRICT bi = v->wavelet + v->cas;
- OPJ_INT32 i;
+ OPJ_UINT32 i;
+ for (i = 0; i < remaining_height; i++) {
+ OPJ_BOOL ret;
+ ret = opj_sparse_array_int32_read(sa,
+ dwt->win_l_x0, sa_line + i,
+ dwt->win_l_x1, sa_line + i + 1,
+ /* Nasty cast from float* to int32* */
+ (OPJ_INT32*)(dwt->wavelet + dwt->cas + 2 * dwt->win_l_x0) + i,
+ 8, 0, OPJ_TRUE);
+ assert(ret);
+ ret = opj_sparse_array_int32_read(sa,
+ (OPJ_UINT32)dwt->sn + dwt->win_h_x0, sa_line + i,
+ (OPJ_UINT32)dwt->sn + dwt->win_h_x1, sa_line + i + 1,
+ /* Nasty cast from float* to int32* */
+ (OPJ_INT32*)(dwt->wavelet + 1 - dwt->cas + 2 * dwt->win_h_x0) + i,
+ 8, 0, OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+ }
+}
+
+static void opj_v4dwt_interleave_v(opj_v4dwt_t* OPJ_RESTRICT dwt,
+ OPJ_FLOAT32* OPJ_RESTRICT a,
+ OPJ_UINT32 width,
+ OPJ_UINT32 nb_elts_read)
+{
+ opj_v4_t* OPJ_RESTRICT bi = dwt->wavelet + dwt->cas;
+ OPJ_UINT32 i;
- for (i = 0; i < v->sn; ++i) {
- memcpy(&bi[i * 2], &a[i * x], (size_t)nb_elts_read * sizeof(OPJ_FLOAT32));
+ for (i = dwt->win_l_x0; i < dwt->win_l_x1; ++i) {
+ memcpy(&bi[i * 2], &a[i * (OPJ_SIZE_T)width],
+ (OPJ_SIZE_T)nb_elts_read * sizeof(OPJ_FLOAT32));
}
- a += v->sn * x;
- bi = v->wavelet + 1 - v->cas;
+ a += (OPJ_UINT32)dwt->sn * (OPJ_SIZE_T)width;
+ bi = dwt->wavelet + 1 - dwt->cas;
- for (i = 0; i < v->dn; ++i) {
- memcpy(&bi[i * 2], &a[i * x], (size_t)nb_elts_read * sizeof(OPJ_FLOAT32));
+ for (i = dwt->win_h_x0; i < dwt->win_h_x1; ++i) {
+ memcpy(&bi[i * 2], &a[i * (OPJ_SIZE_T)width],
+ (OPJ_SIZE_T)nb_elts_read * sizeof(OPJ_FLOAT32));
}
}
+static void opj_v4dwt_interleave_partial_v(opj_v4dwt_t* OPJ_RESTRICT dwt,
+ opj_sparse_array_int32_t* sa,
+ OPJ_UINT32 sa_col,
+ OPJ_UINT32 nb_elts_read)
+{
+ OPJ_BOOL ret;
+ ret = opj_sparse_array_int32_read(sa,
+ sa_col, dwt->win_l_x0,
+ sa_col + nb_elts_read, dwt->win_l_x1,
+ (OPJ_INT32*)(dwt->wavelet + dwt->cas + 2 * dwt->win_l_x0),
+ 1, 8, OPJ_TRUE);
+ assert(ret);
+ ret = opj_sparse_array_int32_read(sa,
+ sa_col, (OPJ_UINT32)dwt->sn + dwt->win_h_x0,
+ sa_col + nb_elts_read, (OPJ_UINT32)dwt->sn + dwt->win_h_x1,
+ (OPJ_INT32*)(dwt->wavelet + 1 - dwt->cas + 2 * dwt->win_h_x0),
+ 1, 8, OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+}
+
#ifdef __SSE__
-static void opj_v4dwt_decode_step1_sse(opj_v4_t* w, OPJ_INT32 count,
+static void opj_v4dwt_decode_step1_sse(opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
const __m128 c)
{
__m128* OPJ_RESTRICT vw = (__m128*) w;
- OPJ_INT32 i;
+ OPJ_UINT32 i;
/* 4x unrolled loop */
- for (i = 0; i < count >> 2; ++i) {
- *vw = _mm_mul_ps(*vw, c);
- vw += 2;
- *vw = _mm_mul_ps(*vw, c);
- vw += 2;
- *vw = _mm_mul_ps(*vw, c);
- vw += 2;
- *vw = _mm_mul_ps(*vw, c);
- vw += 2;
- }
- count &= 3;
- for (i = 0; i < count; ++i) {
- *vw = _mm_mul_ps(*vw, c);
- vw += 2;
+ vw += 2 * start;
+ for (i = start; i + 3 < end; i += 4, vw += 8) {
+ __m128 xmm0 = _mm_mul_ps(vw[0], c);
+ __m128 xmm2 = _mm_mul_ps(vw[2], c);
+ __m128 xmm4 = _mm_mul_ps(vw[4], c);
+ __m128 xmm6 = _mm_mul_ps(vw[6], c);
+ vw[0] = xmm0;
+ vw[2] = xmm2;
+ vw[4] = xmm4;
+ vw[6] = xmm6;
+ }
+ for (; i < end; ++i, vw += 2) {
+ vw[0] = _mm_mul_ps(vw[0], c);
}
}
-void opj_v4dwt_decode_step2_sse(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
- OPJ_INT32 m, __m128 c)
+static void opj_v4dwt_decode_step2_sse(opj_v4_t* l, opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
+ OPJ_UINT32 m,
+ __m128 c)
{
__m128* OPJ_RESTRICT vl = (__m128*) l;
__m128* OPJ_RESTRICT vw = (__m128*) w;
- OPJ_INT32 i;
+ OPJ_UINT32 i;
+ OPJ_UINT32 imax = opj_uint_min(end, m);
__m128 tmp1, tmp2, tmp3;
- tmp1 = vl[0];
- for (i = 0; i < m; ++i) {
+ if (start == 0) {
+ tmp1 = vl[0];
+ } else {
+ vw += start * 2;
+ tmp1 = vw[-3];
+ }
+
+ i = start;
+
+ /* 4x loop unrolling */
+ for (; i + 3 < imax; i += 4) {
+ __m128 tmp4, tmp5, tmp6, tmp7, tmp8, tmp9;
+ tmp2 = vw[-1];
+ tmp3 = vw[ 0];
+ tmp4 = vw[ 1];
+ tmp5 = vw[ 2];
+ tmp6 = vw[ 3];
+ tmp7 = vw[ 4];
+ tmp8 = vw[ 5];
+ tmp9 = vw[ 6];
+ vw[-1] = _mm_add_ps(tmp2, _mm_mul_ps(_mm_add_ps(tmp1, tmp3), c));
+ vw[ 1] = _mm_add_ps(tmp4, _mm_mul_ps(_mm_add_ps(tmp3, tmp5), c));
+ vw[ 3] = _mm_add_ps(tmp6, _mm_mul_ps(_mm_add_ps(tmp5, tmp7), c));
+ vw[ 5] = _mm_add_ps(tmp8, _mm_mul_ps(_mm_add_ps(tmp7, tmp9), c));
+ tmp1 = tmp9;
+ vw += 8;
+ }
+
+ for (; i < imax; ++i) {
tmp2 = vw[-1];
tmp3 = vw[ 0];
vw[-1] = _mm_add_ps(tmp2, _mm_mul_ps(_mm_add_ps(tmp1, tmp3), c));
tmp1 = tmp3;
vw += 2;
}
- vl = vw - 2;
- if (m >= k) {
- return;
- }
- c = _mm_add_ps(c, c);
- c = _mm_mul_ps(c, vl[0]);
- for (; m < k; ++m) {
- __m128 tmp = vw[-1];
- vw[-1] = _mm_add_ps(tmp, c);
- vw += 2;
+ if (m < end) {
+ assert(m + 1 == end);
+ c = _mm_add_ps(c, c);
+ c = _mm_mul_ps(c, vw[-2]);
+ vw[-1] = _mm_add_ps(vw[-1], c);
}
}
#else
-static void opj_v4dwt_decode_step1(opj_v4_t* w, OPJ_INT32 count,
+static void opj_v4dwt_decode_step1(opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
const OPJ_FLOAT32 c)
{
OPJ_FLOAT32* OPJ_RESTRICT fw = (OPJ_FLOAT32*) w;
- OPJ_INT32 i;
- for (i = 0; i < count; ++i) {
+ OPJ_UINT32 i;
+ for (i = start; i < end; ++i) {
OPJ_FLOAT32 tmp1 = fw[i * 8 ];
OPJ_FLOAT32 tmp2 = fw[i * 8 + 1];
OPJ_FLOAT32 tmp3 = fw[i * 8 + 2];
@@ -1641,13 +2403,21 @@ static void opj_v4dwt_decode_step1(opj_v4_t* w, OPJ_INT32 count,
}
}
-static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
- OPJ_INT32 m, OPJ_FLOAT32 c)
+static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w,
+ OPJ_UINT32 start,
+ OPJ_UINT32 end,
+ OPJ_UINT32 m,
+ OPJ_FLOAT32 c)
{
OPJ_FLOAT32* fl = (OPJ_FLOAT32*) l;
OPJ_FLOAT32* fw = (OPJ_FLOAT32*) w;
- OPJ_INT32 i;
- for (i = 0; i < m; ++i) {
+ OPJ_UINT32 i;
+ OPJ_UINT32 imax = opj_uint_min(end, m);
+ if (start > 0) {
+ fw += 8 * start;
+ fl = fw - 8;
+ }
+ for (i = start; i < imax; ++i) {
OPJ_FLOAT32 tmp1_1 = fl[0];
OPJ_FLOAT32 tmp1_2 = fl[1];
OPJ_FLOAT32 tmp1_3 = fl[2];
@@ -1667,27 +2437,13 @@ static void opj_v4dwt_decode_step2(opj_v4_t* l, opj_v4_t* w, OPJ_INT32 k,
fl = fw;
fw += 8;
}
- if (m < k) {
- OPJ_FLOAT32 c1;
- OPJ_FLOAT32 c2;
- OPJ_FLOAT32 c3;
- OPJ_FLOAT32 c4;
+ if (m < end) {
+ assert(m + 1 == end);
c += c;
- c1 = fl[0] * c;
- c2 = fl[1] * c;
- c3 = fl[2] * c;
- c4 = fl[3] * c;
- for (; m < k; ++m) {
- OPJ_FLOAT32 tmp1 = fw[-4];
- OPJ_FLOAT32 tmp2 = fw[-3];
- OPJ_FLOAT32 tmp3 = fw[-2];
- OPJ_FLOAT32 tmp4 = fw[-1];
- fw[-4] = tmp1 + c1;
- fw[-3] = tmp2 + c2;
- fw[-2] = tmp3 + c3;
- fw[-1] = tmp4 + c4;
- fw += 8;
- }
+ fw[-4] = fw[-4] + fl[0] * c;
+ fw[-3] = fw[-3] + fl[1] * c;
+ fw[-2] = fw[-2] + fl[2] * c;
+ fw[-1] = fw[-1] + fl[3] * c;
}
}
@@ -1713,27 +2469,47 @@ static void opj_v4dwt_decode(opj_v4dwt_t* OPJ_RESTRICT dwt)
b = 0;
}
#ifdef __SSE__
- opj_v4dwt_decode_step1_sse(dwt->wavelet + a, dwt->sn, _mm_set1_ps(opj_K));
- opj_v4dwt_decode_step1_sse(dwt->wavelet + b, dwt->dn, _mm_set1_ps(opj_c13318));
- opj_v4dwt_decode_step2_sse(dwt->wavelet + b, dwt->wavelet + a + 1, dwt->sn,
- opj_int_min(dwt->sn, dwt->dn - a), _mm_set1_ps(opj_dwt_delta));
- opj_v4dwt_decode_step2_sse(dwt->wavelet + a, dwt->wavelet + b + 1, dwt->dn,
- opj_int_min(dwt->dn, dwt->sn - b), _mm_set1_ps(opj_dwt_gamma));
- opj_v4dwt_decode_step2_sse(dwt->wavelet + b, dwt->wavelet + a + 1, dwt->sn,
- opj_int_min(dwt->sn, dwt->dn - a), _mm_set1_ps(opj_dwt_beta));
- opj_v4dwt_decode_step2_sse(dwt->wavelet + a, dwt->wavelet + b + 1, dwt->dn,
- opj_int_min(dwt->dn, dwt->sn - b), _mm_set1_ps(opj_dwt_alpha));
+ opj_v4dwt_decode_step1_sse(dwt->wavelet + a, dwt->win_l_x0, dwt->win_l_x1,
+ _mm_set1_ps(opj_K));
+ opj_v4dwt_decode_step1_sse(dwt->wavelet + b, dwt->win_h_x0, dwt->win_h_x1,
+ _mm_set1_ps(opj_c13318));
+ opj_v4dwt_decode_step2_sse(dwt->wavelet + b, dwt->wavelet + a + 1,
+ dwt->win_l_x0, dwt->win_l_x1,
+ (OPJ_UINT32)opj_int_min(dwt->sn, dwt->dn - a),
+ _mm_set1_ps(opj_dwt_delta));
+ opj_v4dwt_decode_step2_sse(dwt->wavelet + a, dwt->wavelet + b + 1,
+ dwt->win_h_x0, dwt->win_h_x1,
+ (OPJ_UINT32)opj_int_min(dwt->dn, dwt->sn - b),
+ _mm_set1_ps(opj_dwt_gamma));
+ opj_v4dwt_decode_step2_sse(dwt->wavelet + b, dwt->wavelet + a + 1,
+ dwt->win_l_x0, dwt->win_l_x1,
+ (OPJ_UINT32)opj_int_min(dwt->sn, dwt->dn - a),
+ _mm_set1_ps(opj_dwt_beta));
+ opj_v4dwt_decode_step2_sse(dwt->wavelet + a, dwt->wavelet + b + 1,
+ dwt->win_h_x0, dwt->win_h_x1,
+ (OPJ_UINT32)opj_int_min(dwt->dn, dwt->sn - b),
+ _mm_set1_ps(opj_dwt_alpha));
#else
- opj_v4dwt_decode_step1(dwt->wavelet + a, dwt->sn, opj_K);
- opj_v4dwt_decode_step1(dwt->wavelet + b, dwt->dn, opj_c13318);
- opj_v4dwt_decode_step2(dwt->wavelet + b, dwt->wavelet + a + 1, dwt->sn,
- opj_int_min(dwt->sn, dwt->dn - a), opj_dwt_delta);
- opj_v4dwt_decode_step2(dwt->wavelet + a, dwt->wavelet + b + 1, dwt->dn,
- opj_int_min(dwt->dn, dwt->sn - b), opj_dwt_gamma);
- opj_v4dwt_decode_step2(dwt->wavelet + b, dwt->wavelet + a + 1, dwt->sn,
- opj_int_min(dwt->sn, dwt->dn - a), opj_dwt_beta);
- opj_v4dwt_decode_step2(dwt->wavelet + a, dwt->wavelet + b + 1, dwt->dn,
- opj_int_min(dwt->dn, dwt->sn - b), opj_dwt_alpha);
+ opj_v4dwt_decode_step1(dwt->wavelet + a, dwt->win_l_x0, dwt->win_l_x1,
+ opj_K);
+ opj_v4dwt_decode_step1(dwt->wavelet + b, dwt->win_h_x0, dwt->win_h_x1,
+ opj_c13318);
+ opj_v4dwt_decode_step2(dwt->wavelet + b, dwt->wavelet + a + 1,
+ dwt->win_l_x0, dwt->win_l_x1,
+ (OPJ_UINT32)opj_int_min(dwt->sn, dwt->dn - a),
+ opj_dwt_delta);
+ opj_v4dwt_decode_step2(dwt->wavelet + a, dwt->wavelet + b + 1,
+ dwt->win_h_x0, dwt->win_h_x1,
+ (OPJ_UINT32)opj_int_min(dwt->dn, dwt->sn - b),
+ opj_dwt_gamma);
+ opj_v4dwt_decode_step2(dwt->wavelet + b, dwt->wavelet + a + 1,
+ dwt->win_l_x0, dwt->win_l_x1,
+ (OPJ_UINT32)opj_int_min(dwt->sn, dwt->dn - a),
+ opj_dwt_beta);
+ opj_v4dwt_decode_step2(dwt->wavelet + a, dwt->wavelet + b + 1,
+ dwt->win_h_x0, dwt->win_h_x1,
+ (OPJ_UINT32)opj_int_min(dwt->dn, dwt->sn - b),
+ opj_dwt_alpha);
#endif
}
@@ -1741,8 +2517,9 @@ static void opj_v4dwt_decode(opj_v4dwt_t* OPJ_RESTRICT dwt)
/* <summary> */
/* Inverse 9-7 wavelet transform in 2-D. */
/* </summary> */
-OPJ_BOOL opj_dwt_decode_real(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
- OPJ_UINT32 numres)
+static
+OPJ_BOOL opj_dwt_decode_tile_97(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
+ OPJ_UINT32 numres)
{
opj_v4dwt_t h;
opj_v4dwt_t v;
@@ -1754,9 +2531,11 @@ OPJ_BOOL opj_dwt_decode_real(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
OPJ_UINT32 rh = (OPJ_UINT32)(res->y1 -
res->y0); /* height of the resolution level computed */
- OPJ_UINT32 w = (OPJ_UINT32)(tilec->x1 - tilec->x0);
+ OPJ_UINT32 w = (OPJ_UINT32)(tilec->resolutions[tilec->minimum_num_resolutions -
+ 1].x1 -
+ tilec->resolutions[tilec->minimum_num_resolutions - 1].x0);
- size_t l_data_size;
+ OPJ_SIZE_T l_data_size;
l_data_size = opj_dwt_max_resolution(res, numres);
/* overflow check */
@@ -1779,9 +2558,7 @@ OPJ_BOOL opj_dwt_decode_real(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
while (--numres) {
OPJ_FLOAT32 * OPJ_RESTRICT aj = (OPJ_FLOAT32*) tilec->data;
- OPJ_UINT32 bufsize = (OPJ_UINT32)((tilec->x1 - tilec->x0) *
- (tilec->y1 - tilec->y0));
- OPJ_INT32 j;
+ OPJ_UINT32 j;
h.sn = (OPJ_INT32)rw;
v.sn = (OPJ_INT32)rh;
@@ -1796,53 +2573,59 @@ OPJ_BOOL opj_dwt_decode_real(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
h.dn = (OPJ_INT32)(rw - (OPJ_UINT32)h.sn);
h.cas = res->x0 % 2;
- for (j = (OPJ_INT32)rh; j > 3; j -= 4) {
- OPJ_INT32 k;
- opj_v4dwt_interleave_h(&h, aj, (OPJ_INT32)w, (OPJ_INT32)bufsize);
+ h.win_l_x0 = 0;
+ h.win_l_x1 = (OPJ_UINT32)h.sn;
+ h.win_h_x0 = 0;
+ h.win_h_x1 = (OPJ_UINT32)h.dn;
+ for (j = 0; j + 3 < rh; j += 4) {
+ OPJ_UINT32 k;
+ opj_v4dwt_interleave_h(&h, aj, w, rh - j);
opj_v4dwt_decode(&h);
- for (k = (OPJ_INT32)rw; --k >= 0;) {
- aj[k ] = h.wavelet[k].f[0];
- aj[k + (OPJ_INT32)w ] = h.wavelet[k].f[1];
- aj[k + (OPJ_INT32)w * 2] = h.wavelet[k].f[2];
- aj[k + (OPJ_INT32)w * 3] = h.wavelet[k].f[3];
+ for (k = 0; k < rw; k++) {
+ aj[k ] = h.wavelet[k].f[0];
+ aj[k + (OPJ_SIZE_T)w ] = h.wavelet[k].f[1];
+ aj[k + (OPJ_SIZE_T)w * 2] = h.wavelet[k].f[2];
+ aj[k + (OPJ_SIZE_T)w * 3] = h.wavelet[k].f[3];
}
aj += w * 4;
- bufsize -= w * 4;
}
- if (rh & 0x03) {
- OPJ_INT32 k;
- j = rh & 0x03;
- opj_v4dwt_interleave_h(&h, aj, (OPJ_INT32)w, (OPJ_INT32)bufsize);
+ if (j < rh) {
+ OPJ_UINT32 k;
+ opj_v4dwt_interleave_h(&h, aj, w, rh - j);
opj_v4dwt_decode(&h);
- for (k = (OPJ_INT32)rw; --k >= 0;) {
- switch (j) {
+ for (k = 0; k < rw; k++) {
+ switch (rh - j) {
case 3:
- aj[k + (OPJ_INT32)w * 2] = h.wavelet[k].f[2];
+ aj[k + (OPJ_SIZE_T)w * 2] = h.wavelet[k].f[2];
/* FALLTHRU */
case 2:
- aj[k + (OPJ_INT32)w ] = h.wavelet[k].f[1];
+ aj[k + (OPJ_SIZE_T)w ] = h.wavelet[k].f[1];
/* FALLTHRU */
case 1:
- aj[k ] = h.wavelet[k].f[0];
+ aj[k] = h.wavelet[k].f[0];
}
}
}
v.dn = (OPJ_INT32)(rh - (OPJ_UINT32)v.sn);
v.cas = res->y0 % 2;
+ v.win_l_x0 = 0;
+ v.win_l_x1 = (OPJ_UINT32)v.sn;
+ v.win_h_x0 = 0;
+ v.win_h_x1 = (OPJ_UINT32)v.dn;
aj = (OPJ_FLOAT32*) tilec->data;
- for (j = (OPJ_INT32)rw; j > 3; j -= 4) {
+ for (j = rw; j > 3; j -= 4) {
OPJ_UINT32 k;
- opj_v4dwt_interleave_v(&v, aj, (OPJ_INT32)w, 4);
+ opj_v4dwt_interleave_v(&v, aj, w, 4);
opj_v4dwt_decode(&v);
for (k = 0; k < rh; ++k) {
- memcpy(&aj[k * w], &v.wavelet[k], 4 * sizeof(OPJ_FLOAT32));
+ memcpy(&aj[k * (OPJ_SIZE_T)w], &v.wavelet[k], 4 * sizeof(OPJ_FLOAT32));
}
aj += 4;
}
@@ -1852,15 +2635,266 @@ OPJ_BOOL opj_dwt_decode_real(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
j = rw & 0x03;
- opj_v4dwt_interleave_v(&v, aj, (OPJ_INT32)w, j);
+ opj_v4dwt_interleave_v(&v, aj, w, j);
opj_v4dwt_decode(&v);
for (k = 0; k < rh; ++k) {
- memcpy(&aj[k * w], &v.wavelet[k], (size_t)j * sizeof(OPJ_FLOAT32));
+ memcpy(&aj[k * (OPJ_SIZE_T)w], &v.wavelet[k],
+ (OPJ_SIZE_T)j * sizeof(OPJ_FLOAT32));
+ }
+ }
+ }
+
+ opj_aligned_free(h.wavelet);
+ return OPJ_TRUE;
+}
+
+static
+OPJ_BOOL opj_dwt_decode_partial_97(opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
+ OPJ_UINT32 numres)
+{
+ opj_sparse_array_int32_t* sa;
+ opj_v4dwt_t h;
+ opj_v4dwt_t v;
+ OPJ_UINT32 resno;
+ /* This value matches the maximum left/right extension given in tables */
+ /* F.2 and F.3 of the standard. Note: in opj_tcd_is_subband_area_of_interest() */
+ /* we currently use 3. */
+ const OPJ_UINT32 filter_width = 4U;
+
+ opj_tcd_resolution_t* tr = tilec->resolutions;
+ opj_tcd_resolution_t* tr_max = &(tilec->resolutions[numres - 1]);
+
+ OPJ_UINT32 rw = (OPJ_UINT32)(tr->x1 -
+ tr->x0); /* width of the resolution level computed */
+ OPJ_UINT32 rh = (OPJ_UINT32)(tr->y1 -
+ tr->y0); /* height of the resolution level computed */
+
+ OPJ_SIZE_T l_data_size;
+
+ /* Compute the intersection of the area of interest, expressed in tile coordinates */
+ /* with the tile coordinates */
+ OPJ_UINT32 win_tcx0 = tilec->win_x0;
+ OPJ_UINT32 win_tcy0 = tilec->win_y0;
+ OPJ_UINT32 win_tcx1 = tilec->win_x1;
+ OPJ_UINT32 win_tcy1 = tilec->win_y1;
+
+ if (tr_max->x0 == tr_max->x1 || tr_max->y0 == tr_max->y1) {
+ return OPJ_TRUE;
+ }
+
+ sa = opj_dwt_init_sparse_array(tilec, numres);
+ if (sa == NULL) {
+ return OPJ_FALSE;
+ }
+
+ if (numres == 1U) {
+ OPJ_BOOL ret = opj_sparse_array_int32_read(sa,
+ tr_max->win_x0 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y0 - (OPJ_UINT32)tr_max->y0,
+ tr_max->win_x1 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y1 - (OPJ_UINT32)tr_max->y0,
+ tilec->data_win,
+ 1, tr_max->win_x1 - tr_max->win_x0,
+ OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+ opj_sparse_array_int32_free(sa);
+ return OPJ_TRUE;
+ }
+
+ l_data_size = opj_dwt_max_resolution(tr, numres);
+ /* overflow check */
+ if (l_data_size > (SIZE_MAX - 5U)) {
+ /* FIXME event manager error callback */
+ return OPJ_FALSE;
+ }
+ l_data_size += 5U;
+ /* overflow check */
+ if (l_data_size > (SIZE_MAX / sizeof(opj_v4_t))) {
+ /* FIXME event manager error callback */
+ return OPJ_FALSE;
+ }
+ h.wavelet = (opj_v4_t*) opj_aligned_malloc(l_data_size * sizeof(opj_v4_t));
+ if (!h.wavelet) {
+ /* FIXME event manager error callback */
+ return OPJ_FALSE;
+ }
+ v.wavelet = h.wavelet;
+
+ for (resno = 1; resno < numres; resno ++) {
+ OPJ_UINT32 j;
+ /* Window of interest subband-based coordinates */
+ OPJ_UINT32 win_ll_x0, win_ll_y0, win_ll_x1, win_ll_y1;
+ OPJ_UINT32 win_hl_x0, win_hl_x1;
+ OPJ_UINT32 win_lh_y0, win_lh_y1;
+ /* Window of interest tile-resolution-based coordinates */
+ OPJ_UINT32 win_tr_x0, win_tr_x1, win_tr_y0, win_tr_y1;
+ /* Tile-resolution subband-based coordinates */
+ OPJ_UINT32 tr_ll_x0, tr_ll_y0, tr_hl_x0, tr_lh_y0;
+
+ ++tr;
+
+ h.sn = (OPJ_INT32)rw;
+ v.sn = (OPJ_INT32)rh;
+
+ rw = (OPJ_UINT32)(tr->x1 - tr->x0);
+ rh = (OPJ_UINT32)(tr->y1 - tr->y0);
+
+ h.dn = (OPJ_INT32)(rw - (OPJ_UINT32)h.sn);
+ h.cas = tr->x0 % 2;
+
+ v.dn = (OPJ_INT32)(rh - (OPJ_UINT32)v.sn);
+ v.cas = tr->y0 % 2;
+
+ /* Get the subband coordinates for the window of interest */
+ /* LL band */
+ opj_dwt_get_band_coordinates(tilec, resno, 0,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ &win_ll_x0, &win_ll_y0,
+ &win_ll_x1, &win_ll_y1);
+
+ /* HL band */
+ opj_dwt_get_band_coordinates(tilec, resno, 1,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ &win_hl_x0, NULL, &win_hl_x1, NULL);
+
+ /* LH band */
+ opj_dwt_get_band_coordinates(tilec, resno, 2,
+ win_tcx0, win_tcy0, win_tcx1, win_tcy1,
+ NULL, &win_lh_y0, NULL, &win_lh_y1);
+
+ /* Beware: band index for non-LL0 resolution are 0=HL, 1=LH and 2=HH */
+ tr_ll_x0 = (OPJ_UINT32)tr->bands[1].x0;
+ tr_ll_y0 = (OPJ_UINT32)tr->bands[0].y0;
+ tr_hl_x0 = (OPJ_UINT32)tr->bands[0].x0;
+ tr_lh_y0 = (OPJ_UINT32)tr->bands[1].y0;
+
+ /* Substract the origin of the bands for this tile, to the subwindow */
+ /* of interest band coordinates, so as to get them relative to the */
+ /* tile */
+ win_ll_x0 = opj_uint_subs(win_ll_x0, tr_ll_x0);
+ win_ll_y0 = opj_uint_subs(win_ll_y0, tr_ll_y0);
+ win_ll_x1 = opj_uint_subs(win_ll_x1, tr_ll_x0);
+ win_ll_y1 = opj_uint_subs(win_ll_y1, tr_ll_y0);
+ win_hl_x0 = opj_uint_subs(win_hl_x0, tr_hl_x0);
+ win_hl_x1 = opj_uint_subs(win_hl_x1, tr_hl_x0);
+ win_lh_y0 = opj_uint_subs(win_lh_y0, tr_lh_y0);
+ win_lh_y1 = opj_uint_subs(win_lh_y1, tr_lh_y0);
+
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)h.sn, &win_ll_x0, &win_ll_x1);
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)h.dn, &win_hl_x0, &win_hl_x1);
+
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)v.sn, &win_ll_y0, &win_ll_y1);
+ opj_dwt_segment_grow(filter_width, (OPJ_UINT32)v.dn, &win_lh_y0, &win_lh_y1);
+
+ /* Compute the tile-resolution-based coordinates for the window of interest */
+ if (h.cas == 0) {
+ win_tr_x0 = opj_uint_min(2 * win_ll_x0, 2 * win_hl_x0 + 1);
+ win_tr_x1 = opj_uint_min(opj_uint_max(2 * win_ll_x1, 2 * win_hl_x1 + 1), rw);
+ } else {
+ win_tr_x0 = opj_uint_min(2 * win_hl_x0, 2 * win_ll_x0 + 1);
+ win_tr_x1 = opj_uint_min(opj_uint_max(2 * win_hl_x1, 2 * win_ll_x1 + 1), rw);
+ }
+
+ if (v.cas == 0) {
+ win_tr_y0 = opj_uint_min(2 * win_ll_y0, 2 * win_lh_y0 + 1);
+ win_tr_y1 = opj_uint_min(opj_uint_max(2 * win_ll_y1, 2 * win_lh_y1 + 1), rh);
+ } else {
+ win_tr_y0 = opj_uint_min(2 * win_lh_y0, 2 * win_ll_y0 + 1);
+ win_tr_y1 = opj_uint_min(opj_uint_max(2 * win_lh_y1, 2 * win_ll_y1 + 1), rh);
+ }
+
+ h.win_l_x0 = win_ll_x0;
+ h.win_l_x1 = win_ll_x1;
+ h.win_h_x0 = win_hl_x0;
+ h.win_h_x1 = win_hl_x1;
+ for (j = 0; j + 3 < rh; j += 4) {
+ if ((j + 3 >= win_ll_y0 && j < win_ll_y1) ||
+ (j + 3 >= win_lh_y0 + (OPJ_UINT32)v.sn &&
+ j < win_lh_y1 + (OPJ_UINT32)v.sn)) {
+ opj_v4dwt_interleave_partial_h(&h, sa, j, opj_uint_min(4U, rh - j));
+ opj_v4dwt_decode(&h);
+ if (!opj_sparse_array_int32_write(sa,
+ win_tr_x0, j,
+ win_tr_x1, j + 4,
+ (OPJ_INT32*)&h.wavelet[win_tr_x0].f[0],
+ 4, 1, OPJ_TRUE)) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ opj_aligned_free(h.wavelet);
+ return OPJ_FALSE;
+ }
+ }
+ }
+
+ if (j < rh &&
+ ((j + 3 >= win_ll_y0 && j < win_ll_y1) ||
+ (j + 3 >= win_lh_y0 + (OPJ_UINT32)v.sn &&
+ j < win_lh_y1 + (OPJ_UINT32)v.sn))) {
+ opj_v4dwt_interleave_partial_h(&h, sa, j, rh - j);
+ opj_v4dwt_decode(&h);
+ if (!opj_sparse_array_int32_write(sa,
+ win_tr_x0, j,
+ win_tr_x1, rh,
+ (OPJ_INT32*)&h.wavelet[win_tr_x0].f[0],
+ 4, 1, OPJ_TRUE)) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ opj_aligned_free(h.wavelet);
+ return OPJ_FALSE;
+ }
+ }
+
+ v.win_l_x0 = win_ll_y0;
+ v.win_l_x1 = win_ll_y1;
+ v.win_h_x0 = win_lh_y0;
+ v.win_h_x1 = win_lh_y1;
+ for (j = win_tr_x0; j < win_tr_x1; j += 4) {
+ OPJ_UINT32 nb_elts = opj_uint_min(4U, win_tr_x1 - j);
+
+ opj_v4dwt_interleave_partial_v(&v, sa, j, nb_elts);
+ opj_v4dwt_decode(&v);
+
+ if (!opj_sparse_array_int32_write(sa,
+ j, win_tr_y0,
+ j + nb_elts, win_tr_y1,
+ (OPJ_INT32*)&h.wavelet[win_tr_y0].f[0],
+ 1, 4, OPJ_TRUE)) {
+ /* FIXME event manager error callback */
+ opj_sparse_array_int32_free(sa);
+ opj_aligned_free(h.wavelet);
+ return OPJ_FALSE;
}
}
}
+ {
+ OPJ_BOOL ret = opj_sparse_array_int32_read(sa,
+ tr_max->win_x0 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y0 - (OPJ_UINT32)tr_max->y0,
+ tr_max->win_x1 - (OPJ_UINT32)tr_max->x0,
+ tr_max->win_y1 - (OPJ_UINT32)tr_max->y0,
+ tilec->data_win,
+ 1, tr_max->win_x1 - tr_max->win_x0,
+ OPJ_TRUE);
+ assert(ret);
+ OPJ_UNUSED(ret);
+ }
+ opj_sparse_array_int32_free(sa);
+
opj_aligned_free(h.wavelet);
return OPJ_TRUE;
}
+
+
+OPJ_BOOL opj_dwt_decode_real(opj_tcd_t *p_tcd,
+ opj_tcd_tilecomp_t* OPJ_RESTRICT tilec,
+ OPJ_UINT32 numres)
+{
+ if (p_tcd->whole_tile_decoding) {
+ return opj_dwt_decode_tile_97(tilec, numres);
+ } else {
+ return opj_dwt_decode_partial_97(tilec, numres);
+ }
+}
generated by cgit v1.2.3 (git 2.25.1) at 2024-12-19 07:05:29 +0000