Optimize reading/write into sparse array

This commit is contained in:
Even Rouault 2017-09-01 16:30:50 +02:00
parent 0ae3cba340
commit b2cc8f7f81
3 changed files with 111 additions and 45 deletions

View File

@ -91,7 +91,7 @@ void opj_sparse_array_int32_free(opj_sparse_array_int32_t* sa)
}
}
OPJ_BOOL opj_sparse_array_is_region_valid(opj_sparse_array_int32_t* sa,
OPJ_BOOL opj_sparse_array_is_region_valid(const opj_sparse_array_int32_t* sa,
OPJ_UINT32 x0,
OPJ_UINT32 y0,
OPJ_UINT32 x1,
@ -102,7 +102,7 @@ OPJ_BOOL opj_sparse_array_is_region_valid(opj_sparse_array_int32_t* sa,
}
static OPJ_BOOL opj_sparse_array_int32_read_or_write(
opj_sparse_array_int32_t* sa,
const opj_sparse_array_int32_t* sa,
OPJ_UINT32 x0,
OPJ_UINT32 y0,
OPJ_UINT32 x1,
@ -115,6 +115,8 @@ static OPJ_BOOL opj_sparse_array_int32_read_or_write(
{
OPJ_UINT32 y, block_y;
OPJ_UINT32 y_incr = 0;
const OPJ_UINT32 block_width = sa->block_width;
if (!opj_sparse_array_is_region_valid(sa, x0, y0, x1, y1)) {
return forgiving;
}
@ -128,43 +130,64 @@ static OPJ_BOOL opj_sparse_array_int32_read_or_write(
sa->block_height;
block_y_offset = sa->block_height - y_incr;
y_incr = opj_uint_min(y_incr, y1 - y);
block_x = x0 / sa->block_width;
block_x = x0 / block_width;
for (x = x0; x < x1; block_x ++, x += x_incr) {
OPJ_UINT32 j;
OPJ_UINT32 block_x_offset;
OPJ_INT32* src_block;
x_incr = (x == x0) ? sa->block_width - (x0 % sa->block_width) : sa->block_width;
block_x_offset = sa->block_width - x_incr;
x_incr = (x == x0) ? block_width - (x0 % block_width) : block_width;
block_x_offset = block_width - x_incr;
x_incr = opj_uint_min(x_incr, x1 - x);
src_block = sa->data_blocks[block_y * sa->block_count_hor + block_x];
if (is_read_op) {
if (src_block == NULL) {
for (j = 0; j < y_incr; j++) {
if (buf_col_stride == 1) {
memset(buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0) * buf_col_stride,
0,
sizeof(OPJ_INT32) * x_incr);
} else {
if (buf_col_stride == 1) {
OPJ_INT32* dest_ptr = buf + (y - y0) * (size_t)buf_line_stride +
(x - x0) * buf_col_stride;
for (j = 0; j < y_incr; j++) {
memset(dest_ptr, 0, sizeof(OPJ_INT32) * x_incr);
dest_ptr += buf_line_stride;
}
} else {
OPJ_INT32* dest_ptr = buf + (y - y0) * (size_t)buf_line_stride +
(x - x0) * buf_col_stride;
for (j = 0; j < y_incr; j++) {
OPJ_UINT32 k;
for (k = 0; k < x_incr; k++) {
*(buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0 + k) * buf_col_stride)
= 0;
dest_ptr[k * buf_col_stride] = 0;
}
dest_ptr += buf_line_stride;
}
}
} else {
for (j = 0; j < y_incr; j++) {
if (buf_col_stride == 1) {
memcpy(buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0) * buf_col_stride,
src_block + (block_y_offset + j) * (size_t)sa->block_width + block_x_offset,
sizeof(OPJ_INT32) * x_incr);
const OPJ_INT32* OPJ_RESTRICT src_ptr = src_block + block_y_offset *
(size_t)block_width + block_x_offset;
if (buf_col_stride == 1) {
OPJ_INT32* OPJ_RESTRICT dest_ptr = buf + (y - y0) * (size_t)buf_line_stride +
(x - x0) * buf_col_stride;
for (j = 0; j < y_incr; j++) {
memcpy(dest_ptr, src_ptr, sizeof(OPJ_INT32) * x_incr);
dest_ptr += buf_line_stride;
src_ptr += block_width;
}
} else {
OPJ_INT32* OPJ_RESTRICT dest_ptr = buf + (y - y0) * (size_t)buf_line_stride +
(x - x0) * buf_col_stride;
if (x_incr == 1) {
for (j = 0; j < y_incr; j++) {
*dest_ptr = *src_ptr;
dest_ptr += buf_line_stride;
src_ptr += block_width;
}
} else {
OPJ_UINT32 k;
for (k = 0; k < x_incr; k++) {
*(buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0 + k) * buf_col_stride)
=
*(src_block + (block_y_offset + j) * (size_t)sa->block_width + block_x_offset +
k);
/* General case */
for (j = 0; j < y_incr; j++) {
OPJ_UINT32 k;
for (k = 0; k < x_incr; k++) {
dest_ptr[k * buf_col_stride] = src_ptr[k];
}
dest_ptr += buf_line_stride;
src_ptr += block_width;
}
}
}
@ -179,18 +202,36 @@ static OPJ_BOOL opj_sparse_array_int32_read_or_write(
sa->data_blocks[block_y * sa->block_count_hor + block_x] = src_block;
}
for (j = 0; j < y_incr; j++) {
if (buf_col_stride == 1) {
memcpy(src_block + (block_y_offset + j) * (size_t)sa->block_width +
block_x_offset,
buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0) * buf_col_stride,
sizeof(OPJ_INT32) * x_incr);
if (buf_col_stride == 1) {
OPJ_INT32* OPJ_RESTRICT dest_ptr = src_block + block_y_offset *
(size_t)block_width + block_x_offset;
const OPJ_INT32* OPJ_RESTRICT src_ptr = buf + (y - y0) *
(size_t)buf_line_stride + (x - x0) * buf_col_stride;
for (j = 0; j < y_incr; j++) {
memcpy(dest_ptr, src_ptr, sizeof(OPJ_INT32) * x_incr);
dest_ptr += block_width;
src_ptr += buf_line_stride;
}
} else {
OPJ_INT32* OPJ_RESTRICT dest_ptr = src_block + block_y_offset *
(size_t)block_width + block_x_offset;
const OPJ_INT32* OPJ_RESTRICT src_ptr = buf + (y - y0) *
(size_t)buf_line_stride + (x - x0) * buf_col_stride;
if (x_incr == 1) {
for (j = 0; j < y_incr; j++) {
*dest_ptr = *src_ptr;
src_ptr += buf_line_stride;
dest_ptr += block_width;
}
} else {
OPJ_UINT32 k;
for (k = 0; k < x_incr; k++) {
*(src_block + (block_y_offset + j) * (size_t)sa->block_width + block_x_offset +
k) =
*(buf + (y - y0 + j) * (size_t)buf_line_stride + (x - x0 + k) * buf_col_stride);
/* General case */
for (j = 0; j < y_incr; j++) {
OPJ_UINT32 k;
for (k = 0; k < x_incr; k++) {
dest_ptr[k] = src_ptr[k * buf_col_stride];
}
src_ptr += buf_line_stride;
dest_ptr += block_width;
}
}
}
@ -201,7 +242,7 @@ static OPJ_BOOL opj_sparse_array_int32_read_or_write(
return OPJ_TRUE;
}
OPJ_BOOL opj_sparse_array_int32_read(opj_sparse_array_int32_t* sa,
OPJ_BOOL opj_sparse_array_int32_read(const opj_sparse_array_int32_t* sa,
OPJ_UINT32 x0,
OPJ_UINT32 y0,
OPJ_UINT32 x1,
@ -211,12 +252,13 @@ OPJ_BOOL opj_sparse_array_int32_read(opj_sparse_array_int32_t* sa,
OPJ_UINT32 dest_line_stride,
OPJ_BOOL forgiving)
{
return opj_sparse_array_int32_read_or_write(sa, x0, y0, x1, y1,
dest,
dest_col_stride,
dest_line_stride,
forgiving,
OPJ_TRUE);
return opj_sparse_array_int32_read_or_write(
(opj_sparse_array_int32_t*)sa, x0, y0, x1, y1,
dest,
dest_col_stride,
dest_line_stride,
forgiving,
OPJ_TRUE);
}
OPJ_BOOL opj_sparse_array_int32_write(opj_sparse_array_int32_t* sa,

View File

@ -77,7 +77,7 @@ void opj_sparse_array_int32_free(opj_sparse_array_int32_t* sa);
* @param y1 bottom y coordinate (not included) of the region. Must be greater than y0.
* @return OPJ_TRUE or OPJ_FALSE.
*/
OPJ_BOOL opj_sparse_array_is_region_valid(opj_sparse_array_int32_t* sa,
OPJ_BOOL opj_sparse_array_is_region_valid(const opj_sparse_array_int32_t* sa,
OPJ_UINT32 x0,
OPJ_UINT32 y0,
OPJ_UINT32 x1,
@ -99,7 +99,7 @@ OPJ_BOOL opj_sparse_array_is_region_valid(opj_sparse_array_int32_t* sa,
* @param forgiving if set to TRUE and the region is invalid, OPJ_TRUE will still be returned.
* @return OPJ_TRUE in case of success.
*/
OPJ_BOOL opj_sparse_array_int32_read(opj_sparse_array_int32_t* sa,
OPJ_BOOL opj_sparse_array_int32_read(const opj_sparse_array_int32_t* sa,
OPJ_UINT32 x0,
OPJ_UINT32 y0,
OPJ_UINT32 x1,

View File

@ -92,6 +92,7 @@ int main()
ret = opj_sparse_array_int32_write(sa, 4, 5, 4 + 1, 5 + 1, buffer, 1, 1,
OPJ_FALSE);
assert(ret);
buffer[0] = 2;
ret = opj_sparse_array_int32_write(sa, 4, 5, 4 + 1, 5 + 1, buffer, 1, 1,
OPJ_FALSE);
@ -105,6 +106,29 @@ int main()
assert(buffer[0] == 2);
assert(buffer[1] == 0xFF);
buffer[0] = 0xFF;
buffer[1] = 0xFF;
buffer[2] = 0xFF;
ret = opj_sparse_array_int32_read(sa, 4, 5, 4 + 1, 5 + 2, buffer, 0, 1,
OPJ_FALSE);
assert(ret);
assert(buffer[0] == 2);
assert(buffer[1] == 0);
assert(buffer[2] == 0xFF);
buffer[0] = 3;
ret = opj_sparse_array_int32_write(sa, 4, 5, 4 + 1, 5 + 1, buffer, 0, 1,
OPJ_FALSE);
assert(ret);
buffer[0] = 0;
buffer[1] = 0xFF;
ret = opj_sparse_array_int32_read(sa, 4, 5, 4 + 1, 5 + 1, buffer, 1, 1,
OPJ_FALSE);
assert(ret);
assert(buffer[0] == 3);
assert(buffer[1] == 0xFF);
w = 15 + 1;
h = 17 + 1;
memset(buffer, 0xFF, sizeof(buffer));
@ -114,7 +138,7 @@ int main()
for (j = 0; j < h; j++) {
for (i = 0; i < w; i++) {
if (i == 4 - 2 && j == 5 - 1) {
assert(buffer[ j * w + i ] == 2);
assert(buffer[ j * w + i ] == 3);
} else {
assert(buffer[ j * w + i ] == 0);
}