aboutsummaryrefslogtreecommitdiff
path: root/source/convert_argb.cc
diff options
context:
space:
mode:
Diffstat (limited to 'source/convert_argb.cc')
-rw-r--r--source/convert_argb.cc18
1 files changed, 18 insertions, 0 deletions
diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index 5c946c3a..ad4470c7 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -4670,6 +4670,7 @@ int Android420ToARGBMatrix(const uint8_t* src_y,
// General case fallback creates NV12
align_buffer_64(plane_uv, halfwidth * 2 * halfheight);
+ if (!plane_uv) return 1;
dst_uv = plane_uv;
for (y = 0; y < halfheight; ++y) {
WeavePixels(src_u, src_v, src_pixel_stride_uv, dst_uv, halfwidth);
@@ -5982,6 +5983,7 @@ int I420ToRGB565Dither(const uint8_t* src_y,
{
// Allocate a row of argb.
align_buffer_64(row_argb, width * 4);
+ if (!row_argb) return 1;
for (y = 0; y < height; ++y) {
I422ToARGBRow(src_y, src_u, src_v, row_argb, &kYuvI601Constants, width);
ARGBToRGB565DitherRow(row_argb, dst_rgb565,
@@ -6240,6 +6242,7 @@ static int I420ToARGBMatrixBilinear(const uint8_t* src_y,
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4);
+ if (!row) return 1;
uint8_t* temp_u_1 = row;
uint8_t* temp_u_2 = row + row_size;
uint8_t* temp_v_1 = row + row_size * 2;
@@ -6376,6 +6379,7 @@ static int I422ToARGBMatrixLinear(const uint8_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2);
+ if (!row) return 1;
uint8_t* temp_u = row;
uint8_t* temp_v = row + row_size;
@@ -6507,6 +6511,7 @@ static int I420ToRGB24MatrixBilinear(const uint8_t* src_y,
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4);
+ if (!row) return 1;
uint8_t* temp_u_1 = row;
uint8_t* temp_u_2 = row + row_size;
uint8_t* temp_v_1 = row + row_size * 2;
@@ -6613,6 +6618,7 @@ static int I010ToAR30MatrixBilinear(const uint16_t* src_y,
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u_1 = (uint16_t*)(row);
uint16_t* temp_u_2 = (uint16_t*)(row) + row_size;
uint16_t* temp_v_1 = (uint16_t*)(row) + row_size * 2;
@@ -6712,6 +6718,7 @@ static int I210ToAR30MatrixLinear(const uint16_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u = (uint16_t*)(row);
uint16_t* temp_v = (uint16_t*)(row) + row_size;
@@ -6800,6 +6807,7 @@ static int I010ToARGBMatrixBilinear(const uint16_t* src_y,
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u_1 = (uint16_t*)(row);
uint16_t* temp_u_2 = (uint16_t*)(row) + row_size;
uint16_t* temp_v_1 = (uint16_t*)(row) + row_size * 2;
@@ -6898,6 +6906,7 @@ static int I210ToARGBMatrixLinear(const uint16_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u = (uint16_t*)(row);
uint16_t* temp_v = (uint16_t*)(row) + row_size;
@@ -7074,6 +7083,7 @@ static int I420AlphaToARGBMatrixBilinear(
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4);
+ if (!row) return 1;
uint8_t* temp_u_1 = row;
uint8_t* temp_u_2 = row + row_size;
uint8_t* temp_v_1 = row + row_size * 2;
@@ -7274,6 +7284,7 @@ static int I422AlphaToARGBMatrixLinear(const uint8_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2);
+ if (!row) return 1;
uint8_t* temp_u = row;
uint8_t* temp_v = row + row_size;
@@ -7413,6 +7424,7 @@ static int I010AlphaToARGBMatrixBilinear(
// alloc 4 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 4 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u_1 = (uint16_t*)(row);
uint16_t* temp_u_2 = (uint16_t*)(row) + row_size;
uint16_t* temp_v_1 = (uint16_t*)(row) + row_size * 2;
@@ -7574,6 +7586,7 @@ static int I210AlphaToARGBMatrixLinear(const uint16_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_u = (uint16_t*)(row);
uint16_t* temp_v = (uint16_t*)(row) + row_size;
@@ -7659,6 +7672,7 @@ static int P010ToARGBMatrixBilinear(const uint16_t* src_y,
// alloc 2 lines temp
const int row_size = (2 * width + 31) & ~31;
align_buffer_64(row, row_size * 2 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_uv_1 = (uint16_t*)(row);
uint16_t* temp_uv_2 = (uint16_t*)(row) + row_size;
@@ -7749,6 +7763,7 @@ static int P210ToARGBMatrixLinear(const uint16_t* src_y,
const int row_size = (2 * width + 31) & ~31;
align_buffer_64(row, row_size * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_uv = (uint16_t*)(row);
for (y = 0; y < height; ++y) {
@@ -7827,6 +7842,7 @@ static int P010ToAR30MatrixBilinear(const uint16_t* src_y,
// alloc 2 lines temp
const int row_size = (2 * width + 31) & ~31;
align_buffer_64(row, row_size * 2 * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_uv_1 = (uint16_t*)(row);
uint16_t* temp_uv_2 = (uint16_t*)(row) + row_size;
@@ -7917,6 +7933,7 @@ static int P210ToAR30MatrixLinear(const uint16_t* src_y,
const int row_size = (2 * width + 31) & ~31;
align_buffer_64(row, row_size * sizeof(uint16_t));
+ if (!row) return 1;
uint16_t* temp_uv = (uint16_t*)(row);
for (y = 0; y < height; ++y) {
@@ -8017,6 +8034,7 @@ static int I422ToRGB24MatrixLinear(const uint8_t* src_y,
// alloc 2 lines temp
const int row_size = (width + 31) & ~31;
align_buffer_64(row, row_size * 2);
+ if (!row) return 1;
uint8_t* temp_u = row;
uint8_t* temp_v = row + row_size;