mirrors
/
stb
mirror of https://github.com/nothings/stb
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

unroll channel loops in downsampler

This commit is contained in:
Sean Barrett 2014-09-13 06:23:07 -07:00
parent 3ee97c221f
commit c9600c012f
2 changed files with 116 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

114
stb_image_resize.h
View File

@ -1429,16 +1429,64 @@ static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, int n
int max_n = n1; int max_n = n1;
int coefficient_group = coefficient_width * x; int coefficient_group = coefficient_width * x;
for (k = n0; k <= max_n; k++) switch (channels) {
{ case 1:
int out_pixel_index = k * channels; for (k = n0; k <= max_n; k++)
float coefficient = horizontal_coefficients[coefficient_group + k - n0]; {
int c; int out_pixel_index = k * 1;
float coefficient = horizontal_coefficients[coefficient_group + k - n0];
STBIR__DEBUG_ASSERT(coefficient != 0);
output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
}
break;
STBIR__DEBUG_ASSERT(coefficient != 0); case 2:
for (k = n0; k <= max_n; k++)
{
int out_pixel_index = k * 2;
float coefficient = horizontal_coefficients[coefficient_group + k - n0];
STBIR__DEBUG_ASSERT(coefficient != 0);
output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
}
break;
for (c = 0; c < channels; c++) case 3:
output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient; for (k = n0; k <= max_n; k++)
{
int out_pixel_index = k * 3;
float coefficient = horizontal_coefficients[coefficient_group + k - n0];
STBIR__DEBUG_ASSERT(coefficient != 0);
output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
}
break;
case 4:
for (k = n0; k <= max_n; k++)
{
int out_pixel_index = k * 4;
float coefficient = horizontal_coefficients[coefficient_group + k - n0];
STBIR__DEBUG_ASSERT(coefficient != 0);
output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
}
break;
default:
for (k = n0; k <= max_n; k++)
{
int c;
int out_pixel_index = k * channels;
float coefficient = horizontal_coefficients[coefficient_group + k - n0];
STBIR__DEBUG_ASSERT(coefficient != 0);
for (c = 0; c < channels; c++)
output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
}
break;
} }
} }
} }
@ -1717,13 +1765,51 @@ static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n,
float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, kernel_pixel_width, ring_buffer_length); float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, kernel_pixel_width, ring_buffer_length);
for (x = 0; x < output_w; x++) switch (channels) {
{ case 1:
int in_pixel_index = x * channels; for (x = 0; x < output_w; x++)
{
int in_pixel_index = x * 1;
ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
}
break;
case 2:
for (x = 0; x < output_w; x++)
{
int in_pixel_index = x * 2;
ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
}
break;
case 3:
for (x = 0; x < output_w; x++)
{
int in_pixel_index = x * 3;
ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
}
break;
case 4:
for (x = 0; x < output_w; x++)
{
int in_pixel_index = x * 4;
ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
}
break;
default:
for (x = 0; x < output_w; x++)
{
int in_pixel_index = x * channels;
int c; int c;
for (c = 0; c < channels; c++) for (c = 0; c < channels; c++)
ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient; ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
}
break;
} }
} }
} }

16
tests/resample_test.cpp
View File

@ -141,6 +141,21 @@ static void resizer(int argc, char **argv)
exit(0); exit(0);
} }
static void performance(int argc, char **argv)
{
unsigned char* input_pixels;
unsigned char* output_pixels;
int w, h;
int n, i;
int out_w, out_h;
input_pixels = stbi_load(argv[1], &w, &h, &n, 0);
out_w = w/4;
out_h = h/4;
output_pixels = (unsigned char*) malloc(out_w*out_h*n);
for (i=0; i < 100; ++i)
stbir_resize_uint8_srgb(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, n, -1,0);
exit(0);
}
void test_suite(int argc, char **argv); void test_suite(int argc, char **argv);
@ -153,6 +168,7 @@ int main(int argc, char** argv)
int out_w, out_h, out_stride; int out_w, out_h, out_stride;
//resizer(argc, argv); //resizer(argc, argv);
performance(argc, argv);
#if 1 #if 1
test_suite(argc, argv); test_suite(argc, argv);