avcodec/jpeg2000: Remove CBLK limit

This also reduces the amount of memory needed
Fixes Ticket4672

The new code seems slightly faster as well, probably due to better cache usage

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

4 files changed, 92 insertions, 99 deletions

diff --git a/libavcodec/j2kenc.c b/libavcodec/j2kenc.c
index 293515c1ee..dcdf4112c0 100644
--- a/libavcodec/j2kenc.c
+++ b/libavcodec/j2kenc.c
@@ -493,18 +493,18 @@ static void encode_sigpass(Jpeg2000T1Context *t1, int width, int height, int ban
     for (y0 = 0; y0 < height; y0 += 4)
         for (x = 0; x < width; x++)
             for (y = y0; y < height && y < y0+4; y++){
-                if (!(t1->flags[y+1][x+1] & JPEG2000_T1_SIG) && (t1->flags[y+1][x+1] & JPEG2000_T1_SIG_NB)){
-                    int ctxno = ff_jpeg2000_getsigctxno(t1->flags[y+1][x+1], bandno),
-                        bit = t1->data[y][x] & mask ? 1 : 0;
+                if (!(t1->flags[(y+1) * t1->stride + x+1] & JPEG2000_T1_SIG) && (t1->flags[(y+1) * t1->stride + x+1] & JPEG2000_T1_SIG_NB)){
+                    int ctxno = ff_jpeg2000_getsigctxno(t1->flags[(y+1) * t1->stride + x+1], bandno),
+                        bit = t1->data[(y) * t1->stride + x] & mask ? 1 : 0;
                     ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, bit);
                     if (bit){
                         int xorbit;
-                        int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y+1][x+1], &xorbit);
-                        ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[y+1][x+1] >> 15) ^ xorbit);
-                        *nmsedec += getnmsedec_sig(t1->data[y][x], bpno + NMSEDEC_FRACBITS);
-                        ff_jpeg2000_set_significance(t1, x, y, t1->flags[y+1][x+1] >> 15);
+                        int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[(y+1) * t1->stride + x+1], &xorbit);
+                        ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[(y+1) * t1->stride + x+1] >> 15) ^ xorbit);
+                        *nmsedec += getnmsedec_sig(t1->data[(y) * t1->stride + x], bpno + NMSEDEC_FRACBITS);
+                        ff_jpeg2000_set_significance(t1, x, y, t1->flags[(y+1) * t1->stride + x+1] >> 15);
                     }
-                    t1->flags[y+1][x+1] |= JPEG2000_T1_VIS;
+                    t1->flags[(y+1) * t1->stride + x+1] |= JPEG2000_T1_VIS;
                 }
             }
 }
@@ -515,11 +515,11 @@ static void encode_refpass(Jpeg2000T1Context *t1, int width, int height, int *nm
     for (y0 = 0; y0 < height; y0 += 4)
         for (x = 0; x < width; x++)
             for (y = y0; y < height && y < y0+4; y++)
-                if ((t1->flags[y+1][x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG){
-                    int ctxno = ff_jpeg2000_getrefctxno(t1->flags[y+1][x+1]);
-                    *nmsedec += getnmsedec_ref(t1->data[y][x], bpno + NMSEDEC_FRACBITS);
-                    ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[y][x] & mask ? 1:0);
-                    t1->flags[y+1][x+1] |= JPEG2000_T1_REF;
+                if ((t1->flags[(y+1) * t1->stride + x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG){
+                    int ctxno = ff_jpeg2000_getrefctxno(t1->flags[(y+1) * t1->stride + x+1]);
+                    *nmsedec += getnmsedec_ref(t1->data[(y) * t1->stride + x], bpno + NMSEDEC_FRACBITS);
+                    ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[(y) * t1->stride + x] & mask ? 1:0);
+                    t1->flags[(y+1) * t1->stride + x+1] |= JPEG2000_T1_REF;
                 }
 }
 
@@ -529,15 +529,15 @@ static void encode_clnpass(Jpeg2000T1Context *t1, int width, int height, int ban
     for (y0 = 0; y0 < height; y0 += 4)
         for (x = 0; x < width; x++){
             if (y0 + 3 < height && !(
-            (t1->flags[y0+1][x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-            (t1->flags[y0+2][x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-            (t1->flags[y0+3][x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-            (t1->flags[y0+4][x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG))))
+            (t1->flags[(y0+1) * t1->stride + x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+            (t1->flags[(y0+2) * t1->stride + x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+            (t1->flags[(y0+3) * t1->stride + x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+            (t1->flags[(y0+4) * t1->stride + x+1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG))))
             {
                 // aggregation mode
                 int rlen;
                 for (rlen = 0; rlen < 4; rlen++)
-                    if (t1->data[y0+rlen][x] & mask)
+                    if (t1->data[(y0+rlen) * t1->stride + x] & mask)
                         break;
                 ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + MQC_CX_RL, rlen != 4);
                 if (rlen == 4)
@@ -545,34 +545,34 @@ static void encode_clnpass(Jpeg2000T1Context *t1, int width, int height, int ban
                 ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI, rlen >> 1);
                 ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI, rlen & 1);
                 for (y = y0 + rlen; y < y0 + 4; y++){
-                    if (!(t1->flags[y+1][x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))){
-                        int ctxno = ff_jpeg2000_getsigctxno(t1->flags[y+1][x+1], bandno);
+                    if (!(t1->flags[(y+1) * t1->stride + x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))){
+                        int ctxno = ff_jpeg2000_getsigctxno(t1->flags[(y+1) * t1->stride + x+1], bandno);
                         if (y > y0 + rlen)
-                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[y][x] & mask ? 1:0);
-                        if (t1->data[y][x] & mask){ // newly significant
+                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[(y) * t1->stride + x] & mask ? 1:0);
+                        if (t1->data[(y) * t1->stride + x] & mask){ // newly significant
                             int xorbit;
-                            int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y+1][x+1], &xorbit);
-                            *nmsedec += getnmsedec_sig(t1->data[y][x], bpno + NMSEDEC_FRACBITS);
-                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[y+1][x+1] >> 15) ^ xorbit);
-                            ff_jpeg2000_set_significance(t1, x, y, t1->flags[y+1][x+1] >> 15);
+                            int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[(y+1) * t1->stride + x+1], &xorbit);
+                            *nmsedec += getnmsedec_sig(t1->data[(y) * t1->stride + x], bpno + NMSEDEC_FRACBITS);
+                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[(y+1) * t1->stride + x+1] >> 15) ^ xorbit);
+                            ff_jpeg2000_set_significance(t1, x, y, t1->flags[(y+1) * t1->stride + x+1] >> 15);
                         }
                     }
-                    t1->flags[y+1][x+1] &= ~JPEG2000_T1_VIS;
+                    t1->flags[(y+1) * t1->stride + x+1] &= ~JPEG2000_T1_VIS;
                 }
             } else{
                 for (y = y0; y < y0 + 4 && y < height; y++){
-                    if (!(t1->flags[y+1][x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))){
-                        int ctxno = ff_jpeg2000_getsigctxno(t1->flags[y+1][x+1], bandno);
-                        ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[y][x] & mask ? 1:0);
-                        if (t1->data[y][x] & mask){ // newly significant
+                    if (!(t1->flags[(y+1) * t1->stride + x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))){
+                        int ctxno = ff_jpeg2000_getsigctxno(t1->flags[(y+1) * t1->stride + x+1], bandno);
+                        ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, t1->data[(y) * t1->stride + x] & mask ? 1:0);
+                        if (t1->data[(y) * t1->stride + x] & mask){ // newly significant
                             int xorbit;
-                            int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y+1][x+1], &xorbit);
-                            *nmsedec += getnmsedec_sig(t1->data[y][x], bpno + NMSEDEC_FRACBITS);
-                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[y+1][x+1] >> 15) ^ xorbit);
-                            ff_jpeg2000_set_significance(t1, x, y, t1->flags[y+1][x+1] >> 15);
+                            int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[(y+1) * t1->stride + x+1], &xorbit);
+                            *nmsedec += getnmsedec_sig(t1->data[(y) * t1->stride + x], bpno + NMSEDEC_FRACBITS);
+                            ff_mqc_encode(&t1->mqc, t1->mqc.cx_states + ctxno, (t1->flags[(y+1) * t1->stride + x+1] >> 15) ^ xorbit);
+                            ff_jpeg2000_set_significance(t1, x, y, t1->flags[(y+1) * t1->stride + x+1] >> 15);
                         }
                     }
-                    t1->flags[y+1][x+1] &= ~JPEG2000_T1_VIS;
+                    t1->flags[(y+1) * t1->stride + x+1] &= ~JPEG2000_T1_VIS;
                 }
             }
         }
@@ -584,16 +584,15 @@ static void encode_cblk(Jpeg2000EncoderContext *s, Jpeg2000T1Context *t1, Jpeg20
     int pass_t = 2, passno, x, y, max=0, nmsedec, bpno;
     int64_t wmsedec = 0;
 
-    for (y = 0; y < height+2; y++)
-        memset(t1->flags[y], 0, (width+2)*sizeof(int));
+    memset(t1->flags, 0, t1->stride * (height + 2) * sizeof(*t1->flags));
 
     for (y = 0; y < height; y++){
         for (x = 0; x < width; x++){
-            if (t1->data[y][x] < 0){
-                t1->flags[y+1][x+1] |= JPEG2000_T1_SGN;
-                t1->data[y][x] = -t1->data[y][x];
+            if (t1->data[(y) * t1->stride + x] < 0){
+                t1->flags[(y+1) * t1->stride + x+1] |= JPEG2000_T1_SGN;
+                t1->data[(y) * t1->stride + x] = -t1->data[(y) * t1->stride + x];
             }
-            max = FFMAX(max, t1->data[y][x]);
+            max = FFMAX(max, t1->data[(y) * t1->stride + x]);
         }
     }
 
@@ -827,6 +826,8 @@ static int encode_tile(Jpeg2000EncoderContext *s, Jpeg2000Tile *tile, int tileno
     for (compno = 0; compno < s->ncomponents; compno++){
         Jpeg2000Component *comp = s->tile[tileno].comp + compno;
 
+        t1.stride = (1<<codsty->log2_cblk_width) + 2;
+
         av_log(s->avctx, AV_LOG_DEBUG,"dwt\n");
         if ((ret = ff_dwt_encode(&comp->dwt, comp->i_data)) < 0)
             return ret;
@@ -862,14 +863,14 @@ static int encode_tile(Jpeg2000EncoderContext *s, Jpeg2000Tile *tile, int tileno
                         int y, x;
                         if (codsty->transform == FF_DWT53){
                             for (y = yy0; y < yy1; y++){
-                                int *ptr = t1.data[y-yy0];
+                                int *ptr = t1.data + (y-yy0)*t1.stride;
                                 for (x = xx0; x < xx1; x++){
                                     *ptr++ = comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * y + x] << NMSEDEC_FRACBITS;
                                 }
                             }
                         } else{
                             for (y = yy0; y < yy1; y++){
-                                int *ptr = t1.data[y-yy0];
+                                int *ptr = t1.data + (y-yy0)*t1.stride;
                                 for (x = xx0; x < xx1; x++){
                                     *ptr = (comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * y + x]);
                                     *ptr = (int64_t)*ptr * (int64_t)(16384 * 65536 / band->i_stepsize) >> 15 - NMSEDEC_FRACBITS;
diff --git a/libavcodec/jpeg2000.c b/libavcodec/jpeg2000.c
index ebf8320f81..7d2b18f70a 100644
--- a/libavcodec/jpeg2000.c
+++ b/libavcodec/jpeg2000.c
@@ -171,22 +171,22 @@ void ff_jpeg2000_set_significance(Jpeg2000T1Context *t1, int x, int y,
 {
     x++;
     y++;
-    t1->flags[y][x] |= JPEG2000_T1_SIG;
+    t1->flags[(y) * t1->stride + x] |= JPEG2000_T1_SIG;
     if (negative) {
-        t1->flags[y][x + 1] |= JPEG2000_T1_SIG_W | JPEG2000_T1_SGN_W;
-        t1->flags[y][x - 1] |= JPEG2000_T1_SIG_E | JPEG2000_T1_SGN_E;
-        t1->flags[y + 1][x] |= JPEG2000_T1_SIG_N | JPEG2000_T1_SGN_N;
-        t1->flags[y - 1][x] |= JPEG2000_T1_SIG_S | JPEG2000_T1_SGN_S;
+        t1->flags[(y) * t1->stride + x + 1] |= JPEG2000_T1_SIG_W | JPEG2000_T1_SGN_W;
+        t1->flags[(y) * t1->stride + x - 1] |= JPEG2000_T1_SIG_E | JPEG2000_T1_SGN_E;
+        t1->flags[(y + 1) * t1->stride + x] |= JPEG2000_T1_SIG_N | JPEG2000_T1_SGN_N;
+        t1->flags[(y - 1) * t1->stride + x] |= JPEG2000_T1_SIG_S | JPEG2000_T1_SGN_S;
     } else {
-        t1->flags[y][x + 1] |= JPEG2000_T1_SIG_W;
-        t1->flags[y][x - 1] |= JPEG2000_T1_SIG_E;
-        t1->flags[y + 1][x] |= JPEG2000_T1_SIG_N;
-        t1->flags[y - 1][x] |= JPEG2000_T1_SIG_S;
+        t1->flags[(y) * t1->stride + x + 1] |= JPEG2000_T1_SIG_W;
+        t1->flags[(y) * t1->stride + x - 1] |= JPEG2000_T1_SIG_E;
+        t1->flags[(y + 1) * t1->stride + x] |= JPEG2000_T1_SIG_N;
+        t1->flags[(y - 1) * t1->stride + x] |= JPEG2000_T1_SIG_S;
     }
-    t1->flags[y + 1][x + 1] |= JPEG2000_T1_SIG_NW;
-    t1->flags[y + 1][x - 1] |= JPEG2000_T1_SIG_NE;
-    t1->flags[y - 1][x + 1] |= JPEG2000_T1_SIG_SW;
-    t1->flags[y - 1][x - 1] |= JPEG2000_T1_SIG_SE;
+    t1->flags[(y + 1) * t1->stride + x + 1] |= JPEG2000_T1_SIG_NW;
+    t1->flags[(y + 1) * t1->stride + x - 1] |= JPEG2000_T1_SIG_NE;
+    t1->flags[(y - 1) * t1->stride + x + 1] |= JPEG2000_T1_SIG_SW;
+    t1->flags[(y - 1) * t1->stride + x - 1] |= JPEG2000_T1_SIG_SE;
 }
 
 static const uint8_t lut_gain[2][4] = { { 0, 0, 0, 0 }, { 0, 1, 1, 2 } };
diff --git a/libavcodec/jpeg2000.h b/libavcodec/jpeg2000.h
index 24395a24eb..0acad79eff 100644
--- a/libavcodec/jpeg2000.h
+++ b/libavcodec/jpeg2000.h
@@ -67,10 +67,6 @@ enum Jpeg2000Quantsty { // quantization style
     JPEG2000_QSTY_SE    // scalar expounded
 };
 
-#define JPEG2000_MAX_CBLKW 128
-#define JPEG2000_MAX_CBLKH 128
-
-
 #define JPEG2000_MAX_DECLEVELS 32
 #define JPEG2000_MAX_RESLEVELS (JPEG2000_MAX_DECLEVELS + 1)
 
@@ -123,9 +119,10 @@ enum Jpeg2000Quantsty { // quantization style
 #define JPEG2000_PGOD_CPRL      0x04  // Component-position-resolution level-layer progression
 
 typedef struct Jpeg2000T1Context {
-    int data[JPEG2000_MAX_CBLKW][JPEG2000_MAX_CBLKH];
-    uint16_t flags[JPEG2000_MAX_CBLKW + 2][JPEG2000_MAX_CBLKH + 2];
+    int data[6144];
+    uint16_t flags[6156];
     MqcState mqc;
+    int stride;
 } Jpeg2000T1Context;
 
 typedef struct Jpeg2000TgtNode {
diff --git a/libavcodec/jpeg2000dec.c b/libavcodec/jpeg2000dec.c
index 3d518e3c4c..1090f37b68 100644
--- a/libavcodec/jpeg2000dec.c
+++ b/libavcodec/jpeg2000dec.c
@@ -439,11 +439,6 @@ static int get_cox(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c)
         return AVERROR_INVALIDDATA;
     }
 
-    if (c->log2_cblk_width > 7 || c->log2_cblk_height > 7) {
-        avpriv_request_sample(s->avctx, "cblk size > 128");
-        return AVERROR_PATCHWELCOME;
-    }
-
     c->cblk_style = bytestream2_get_byteu(&s->g);
     if (c->cblk_style != 0) { // cblk style
         av_log(s->avctx, AV_LOG_WARNING, "extra cblk styles %X\n", c->cblk_style);
@@ -1233,20 +1228,20 @@ static void decode_sigpass(Jpeg2000T1Context *t1, int width, int height,
                 int flags_mask = -1;
                 if (vert_causal_ctx_csty_symbol && y == y0 + 3)
                     flags_mask &= ~(JPEG2000_T1_SIG_S | JPEG2000_T1_SIG_SW | JPEG2000_T1_SIG_SE | JPEG2000_T1_SGN_S);
-                if ((t1->flags[y+1][x+1] & JPEG2000_T1_SIG_NB & flags_mask)
-                && !(t1->flags[y+1][x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
-                    if (ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ff_jpeg2000_getsigctxno(t1->flags[y+1][x+1] & flags_mask, bandno))) {
-                        int xorbit, ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y+1][x+1] & flags_mask, &xorbit);
+                if ((t1->flags[(y+1) * t1->stride + x+1] & JPEG2000_T1_SIG_NB & flags_mask)
+                && !(t1->flags[(y+1) * t1->stride + x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
+                    if (ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ff_jpeg2000_getsigctxno(t1->flags[(y+1) * t1->stride + x+1] & flags_mask, bandno))) {
+                        int xorbit, ctxno = ff_jpeg2000_getsgnctxno(t1->flags[(y+1) * t1->stride + x+1] & flags_mask, &xorbit);
                         if (t1->mqc.raw)
-                             t1->data[y][x] = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ctxno) ? -mask : mask;
+                             t1->data[(y) * t1->stride + x] = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ctxno) ? -mask : mask;
                         else
-                             t1->data[y][x] = (ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ctxno) ^ xorbit) ?
+                             t1->data[(y) * t1->stride + x] = (ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ctxno) ^ xorbit) ?
                                                -mask : mask;
 
                         ff_jpeg2000_set_significance(t1, x, y,
-                                                     t1->data[y][x] < 0);
+                                                     t1->data[(y) * t1->stride + x] < 0);
                     }
-                    t1->flags[y + 1][x + 1] |= JPEG2000_T1_VIS;
+                    t1->flags[(y + 1) * t1->stride + x + 1] |= JPEG2000_T1_VIS;
                 }
             }
 }
@@ -1263,15 +1258,15 @@ static void decode_refpass(Jpeg2000T1Context *t1, int width, int height,
     for (y0 = 0; y0 < height; y0 += 4)
         for (x = 0; x < width; x++)
             for (y = y0; y < height && y < y0 + 4; y++)
-                if ((t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG) {
+                if ((t1->flags[(y + 1) * t1->stride + x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG) {
                     int flags_mask = (vert_causal_ctx_csty_symbol && y == y0 + 3) ?
                         ~(JPEG2000_T1_SIG_S | JPEG2000_T1_SIG_SW | JPEG2000_T1_SIG_SE | JPEG2000_T1_SGN_S) : -1;
-                    int ctxno = ff_jpeg2000_getrefctxno(t1->flags[y + 1][x + 1] & flags_mask);
+                    int ctxno = ff_jpeg2000_getrefctxno(t1->flags[(y + 1) * t1->stride + x + 1] & flags_mask);
                     int r     = ff_mqc_decode(&t1->mqc,
                                               t1->mqc.cx_states + ctxno)
                                 ? phalf : nhalf;
-                    t1->data[y][x]          += t1->data[y][x] < 0 ? -r : r;
-                    t1->flags[y + 1][x + 1] |= JPEG2000_T1_REF;
+                    t1->data[(y) * t1->stride + x]          += t1->data[(y) * t1->stride + x] < 0 ? -r : r;
+                    t1->flags[(y + 1) * t1->stride + x + 1] |= JPEG2000_T1_REF;
                 }
 }
 
@@ -1287,10 +1282,10 @@ static void decode_clnpass(Jpeg2000DecoderContext *s, Jpeg2000T1Context *t1,
             if (vert_causal_ctx_csty_symbol)
                 flags_mask &= ~(JPEG2000_T1_SIG_S | JPEG2000_T1_SIG_SW | JPEG2000_T1_SIG_SE | JPEG2000_T1_SGN_S);
             if (y0 + 3 < height &&
-                !((t1->flags[y0 + 1][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-                  (t1->flags[y0 + 2][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-                  (t1->flags[y0 + 3][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
-                  (t1->flags[y0 + 4][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG) & flags_mask))) {
+                !((t1->flags[(y0 + 1) * t1->stride + x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+                  (t1->flags[(y0 + 2) * t1->stride + x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+                  (t1->flags[(y0 + 3) * t1->stride + x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
+                  (t1->flags[(y0 + 4) * t1->stride + x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG) & flags_mask))) {
                 if (!ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_RL))
                     continue;
                 runlen = ff_mqc_decode(&t1->mqc,
@@ -1309,23 +1304,23 @@ static void decode_clnpass(Jpeg2000DecoderContext *s, Jpeg2000T1Context *t1,
                 if (vert_causal_ctx_csty_symbol && y == y0 + 3)
                     flags_mask &= ~(JPEG2000_T1_SIG_S | JPEG2000_T1_SIG_SW | JPEG2000_T1_SIG_SE | JPEG2000_T1_SGN_S);
                 if (!dec) {
-                    if (!(t1->flags[y+1][x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
-                        dec = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ff_jpeg2000_getsigctxno(t1->flags[y+1][x+1] & flags_mask,
+                    if (!(t1->flags[(y+1) * t1->stride + x+1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
+                        dec = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + ff_jpeg2000_getsigctxno(t1->flags[(y+1) * t1->stride + x+1] & flags_mask,
                                                                                              bandno));
                     }
                 }
                 if (dec) {
                     int xorbit;
-                    int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y + 1][x + 1] & flags_mask,
+                    int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[(y + 1) * t1->stride + x + 1] & flags_mask,
                                                         &xorbit);
-                    t1->data[y][x] = (ff_mqc_decode(&t1->mqc,
+                    t1->data[(y) * t1->stride + x] = (ff_mqc_decode(&t1->mqc,
                                                     t1->mqc.cx_states + ctxno) ^
                                       xorbit)
                                      ? -mask : mask;
-                    ff_jpeg2000_set_significance(t1, x, y, t1->data[y][x] < 0);
+                    ff_jpeg2000_set_significance(t1, x, y, t1->data[(y) * t1->stride + x] < 0);
                 }
                 dec = 0;
-                t1->flags[y + 1][x + 1] &= ~JPEG2000_T1_VIS;
+                t1->flags[(y + 1) * t1->stride + x + 1] &= ~JPEG2000_T1_VIS;
             }
         }
     }
@@ -1351,18 +1346,16 @@ static int decode_cblk(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *codsty,
     int term_cnt = 0;
     int coder_type;
 
-    av_assert0(width  <= JPEG2000_MAX_CBLKW);
-    av_assert0(height <= JPEG2000_MAX_CBLKH);
+    av_assert0(width <= 1024U && height <= 1024U);
+    av_assert0(width*height <= 4096);
 
-    for (y = 0; y < height; y++)
-        memset(t1->data[y], 0, width * sizeof(**t1->data));
+    memset(t1->data, 0, t1->stride * height * sizeof(*t1->data));
 
     /* If code-block contains no compressed data: nothing to do. */
     if (!cblk->length)
         return 0;
 
-    for (y = 0; y < height + 2; y++)
-        memset(t1->flags[y], 0, (width + 2) * sizeof(**t1->flags));
+    memset(t1->flags, 0, t1->stride * (height + 2) * sizeof(*t1->flags));
 
     cblk->data[cblk->length] = 0xff;
     cblk->data[cblk->length+1] = 0xff;
@@ -1426,7 +1419,7 @@ static void dequantization_float(int x, int y, Jpeg2000Cblk *cblk,
     int w = cblk->coord[0][1] - cblk->coord[0][0];
     for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j) {
         float *datap = &comp->f_data[(comp->coord[0][1] - comp->coord[0][0]) * (y + j) + x];
-        int *src = t1->data[j];
+        int *src = t1->data + j*t1->stride;
         for (i = 0; i < w; ++i)
             datap[i] = src[i] * band->f_stepsize;
     }
@@ -1441,7 +1434,7 @@ static void dequantization_int(int x, int y, Jpeg2000Cblk *cblk,
     int w = cblk->coord[0][1] - cblk->coord[0][0];
     for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j) {
         int32_t *datap = &comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * (y + j) + x];
-        int *src = t1->data[j];
+        int *src = t1->data + j*t1->stride;
         if (band->i_stepsize == 16384) {
             for (i = 0; i < w; ++i)
                 datap[i] = src[i] / 2;
@@ -1461,7 +1454,7 @@ static void dequantization_int_97(int x, int y, Jpeg2000Cblk *cblk,
     int w = cblk->coord[0][1] - cblk->coord[0][0];
     for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j) {
         int32_t *datap = &comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * (y + j) + x];
-        int *src = t1->data[j];
+        int *src = t1->data + j*t1->stride;
         for (i = 0; i < w; ++i)
             datap[i] = (src[i] * (int64_t)band->i_stepsize + (1<<14)) >> 15;
     }
@@ -1512,6 +1505,8 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
         Jpeg2000Component *comp     = tile->comp + compno;
         Jpeg2000CodingStyle *codsty = tile->codsty + compno;
 
+        t1.stride = (1<<codsty->log2_cblk_width) + 2;
+
         /* Loop on resolution levels */
         for (reslevelno = 0; reslevelno < codsty->nreslevels2decode; reslevelno++) {
             Jpeg2000ResLevel *rlevel = comp->reslevel + reslevelno;