#include <math.h>
#include <string.h>
#include <stdlib.h>
+#include "jpeglib.h"
-#define CACHE_LINE_FACTOR 1
-
-void myreadline(char *buf)
-{
- char *ptr;
-
- if (fgets(buf, 256, stdin) == NULL) {
- fprintf(stderr, "Premuture EOF\n");
- exit(1);
- }
-
- if (strlen(buf) == 0) {
- return;
- }
-
- ptr = &buf[strlen(buf) - 1];
- if (*ptr == '\n') {
- *ptr-- = 0;
- if (*ptr == '\r') {
- *ptr = 0;
- }
- }
-}
+#define CACHE_LINE_FACTOR 8
double sinc(double x)
{
unsigned startcoeff;
};
-void hscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nw)
+void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw, unsigned dstride)
{
struct pix_desc *pd = (struct pix_desc *)malloc(nw * sizeof(struct pix_desc));
int size_coeffs = 8;
int num_coeffs = 0;
int x, y, sx;
double sf = (double)w / (double)nw;
- double support = 3.0 * sf;
+ double support = (w > nw) ? (3.0 * sf) : (3.0 / sf);
/* calculate the filter */
for (x = 0; x < nw; ++x) {
pd[x].startcoeff = num_coeffs;
for (sx = start; sx <= end; ++sx) {
- double nd = sx/sf - x;
+ double nd = (w > nw) ? (sx/sf - x) : (sx - x*sf);
double f = lanczos_tap(nd);
if (num_coeffs == size_coeffs) {
size_coeffs <<= 1;
coeffs = (float *)realloc(coeffs, size_coeffs * sizeof(float));
}
-
+
coeffs[num_coeffs++] = f;
sum += f;
}
}
for (y = 0; y < h; ++y) {
- unsigned char *sptr = pix + y*w;
- float *dptr = npix + y*nw;
+ float *sptr = pix + y*w;
+ unsigned char *dptr = npix + y*dstride;
+ unsigned char ch = 0;
for (x = 0; x < nw; ++x) {
float acc = 0.0;
float *cf = &coeffs[pd[x].startcoeff];
for (sx = pd[x].start; sx <= pd[x].end; ++sx) {
acc += sptr[sx] * *cf++;
}
- *dptr++ = acc;
+
+ if (acc < 0.0)
+ ch = 0;
+ else if (acc > 255.0)
+ ch = 255;
+ else
+ ch = (unsigned char)acc;
+ *dptr++ = ch;
+ }
+ for ( ; x < dstride; ++x) {
+ *dptr++ = ch;
}
}
}
-void vscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nh)
+void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh, unsigned dstride)
{
struct pix_desc *pd = (struct pix_desc *)malloc(nh * sizeof(struct pix_desc));
int size_coeffs = 8;
int num_coeffs = 0;
int x, y, sy;
double sf = (double)h / (double)nh;
- double support = 3.0 * sf;
+ double support = (h > nh) ? (3.0 * sf) : (3.0 / sf);
/* calculate the filter */
for (y = 0; y < nh; ++y) {
pd[y].startcoeff = num_coeffs;
for (sy = start; sy <= end; ++sy) {
- double nd = sy/sf - y;
+ double nd = (h > nh) ? (sy/sf - y) : (sy - y*sf);
double f = lanczos_tap(nd);
if (num_coeffs == size_coeffs) {
size_coeffs <<= 1;
coeffs[pd[y].startcoeff + sy - start] /= sum;
}
}
+
#if CACHE_LINE_FACTOR > 1
for (x = 0; x < w; x += CACHE_LINE_FACTOR) {
- float *sptr = pix + x;
- unsigned char *dptr = npix + x;
+ unsigned char *sptr = pix + x;
+ float *dptr = npix + x;
for (y = 0; y < nh; ++y) {
int i;
float acc[CACHE_LINE_FACTOR];
acc[i] = 0.0;
float *cf = &coeffs[pd[y].startcoeff];
unsigned sy;
- unsigned char ch;
for (sy = pd[y].start; sy <= pd[y].end; ++sy) {
+ //asm volatile ("prefetcht0 %0" :: "m" (sptr[(sy+1) * w]));
for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
acc[i] += sptr[sy * w + i] * *cf;
}
}
for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
- if (acc[i] < 0.0)
- ch = 0;
- else if (acc[i] > 255.0)
- ch = 255;
- else
- ch = (unsigned char)acc[i];
- dptr[i] = ch;
+ dptr[i] = acc[i];
}
- dptr += w;
+ dptr += dstride;
}
}
-#endif
for (x = (x/CACHE_LINE_FACTOR)*CACHE_LINE_FACTOR; x < w; ++x) {
- float *sptr = pix + x;
- unsigned char *dptr = npix + x;
+#else
+ for (x = 0; x < w; ++x) {
+#endif
+ unsigned char *sptr = pix + x;
+ float *dptr = npix + x;
for (y = 0; y < nh; ++y) {
float acc = 0.0;
float *cf = &coeffs[pd[y].startcoeff];
unsigned sy;
- unsigned char ch;
for (sy = pd[y].start; sy <= pd[y].end; ++sy) {
acc += sptr[sy * w] * *cf++;
}
- if (acc < 0.0)
- ch = 0;
- else if (acc > 255.0)
- ch = 255;
- else
- ch = (unsigned char)acc;
-
- *dptr = ch;
- dptr += w;
+ *dptr = acc;
+ dptr += dstride;
}
}
}
-int main(void)
+int main(int argc, char **argv)
{
- unsigned w, h;
- unsigned nw = 600, nh = 400;
- char buf[256];
- unsigned char *pix;
- float *npix;
- unsigned char *npix2;
-
- myreadline(buf);
- if (strcmp(buf, "P5") != 0) {
- fprintf(stderr, "Error: Not a PGM file\n");
- exit(1);
- }
- myreadline(buf); /* comment */
- myreadline(buf);
+ unsigned nominal_w = atoi(argv[1]);
+ unsigned nominal_h = atoi(argv[2]);
+
+ unsigned samp_h0 = 2, samp_v0 = 2;
+ unsigned samp_h1 = 1, samp_v1 = 1;
+ unsigned samp_h2 = 1, samp_v2 = 1;
+ unsigned max_samp_h = 2, max_samp_v = 2;
+
+ unsigned nw0 = nominal_w * samp_h0 / max_samp_h, nh0 = nominal_h * samp_v0 / max_samp_v;
+ unsigned nw1 = nominal_w * samp_h1 / max_samp_h, nh1 = nominal_h * samp_v1 / max_samp_v;
+ unsigned nw2 = nominal_w * samp_h2 / max_samp_h, nh2 = nominal_h * samp_v2 / max_samp_v;
+
+ unsigned stride0 = (nw0 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned stride1 = (nw1 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned stride2 = (nw2 + DCTSIZE-1) & ~(DCTSIZE-1);
+
+ struct jpeg_decompress_struct dinfo;
+ struct jpeg_error_mgr jerr;
+ dinfo.err = jpeg_std_error(&jerr);
+ jpeg_create_decompress(&dinfo);
+ jpeg_stdio_src(&dinfo, stdin);
+ jpeg_read_header(&dinfo, TRUE);
+ dinfo.raw_data_out = TRUE;
+ jpeg_start_decompress(&dinfo);
+
+ fprintf(stderr, "Scaling using Lanczos filter:\n");
+ fprintf(stderr, " Y component: %ux%u -> %ux%u\n", dinfo.comp_info[0].width_in_blocks * DCTSIZE, dinfo.comp_info[0].height_in_blocks * DCTSIZE, nw0, nh0);
+ fprintf(stderr, " Cb component: %ux%u -> %ux%u\n", dinfo.comp_info[1].width_in_blocks * DCTSIZE, dinfo.comp_info[1].height_in_blocks * DCTSIZE, nw1, nh1);
+ fprintf(stderr, " Cr component: %ux%u -> %ux%u\n", dinfo.comp_info[2].width_in_blocks * DCTSIZE, dinfo.comp_info[2].height_in_blocks * DCTSIZE, nw2, nh2);
+
+ JSAMPLE *data_y = (JSAMPLE*)malloc(dinfo.comp_info[0].height_in_blocks * dinfo.comp_info[0].width_in_blocks * DCTSIZE * DCTSIZE);
+ JSAMPLE *data_cb = (JSAMPLE*)malloc(dinfo.comp_info[1].height_in_blocks * dinfo.comp_info[1].width_in_blocks * DCTSIZE * DCTSIZE);
+ JSAMPLE *data_cr = (JSAMPLE*)malloc(dinfo.comp_info[2].height_in_blocks * dinfo.comp_info[2].width_in_blocks * DCTSIZE * DCTSIZE);
+ JSAMPLE *data_ny, *data_ncb, *data_ncr;
+
+ int total_lines = 0, blocks = 0;
+ while (total_lines < dinfo.comp_info[0].height_in_blocks * DCTSIZE) {
+ unsigned max_lines = dinfo.max_v_samp_factor * DCTSIZE;
+
+ JSAMPROW y_row_ptrs[max_lines];
+ JSAMPROW cb_row_ptrs[max_lines];
+ JSAMPROW cr_row_ptrs[max_lines];
+ JSAMPROW* ptrs[] = { y_row_ptrs, cb_row_ptrs, cr_row_ptrs };
+ int i;
+
+ for (i = 0; i < max_lines; ++i) {
+ y_row_ptrs[i] = data_y + (i+blocks*DCTSIZE*dinfo.comp_info[0].v_samp_factor) * dinfo.comp_info[0].width_in_blocks * DCTSIZE;
+ cb_row_ptrs[i] = data_cb + (i+blocks*DCTSIZE*dinfo.comp_info[1].v_samp_factor) * dinfo.comp_info[1].width_in_blocks * DCTSIZE;
+ cr_row_ptrs[i] = data_cr + (i+blocks*DCTSIZE*dinfo.comp_info[2].v_samp_factor) * dinfo.comp_info[2].width_in_blocks * DCTSIZE;
+ }
+
+ total_lines += max_lines;
+ ++blocks;
- if (sscanf(buf, "%u %u", &w, &h) != 2) {
- fprintf(stderr, "Error: Expected width/height\n");
- exit(1);
+ if (jpeg_read_raw_data(&dinfo, ptrs, max_lines) == 0)
+ break;
}
- myreadline(buf);
- if (strcmp(buf, "255") != 0) {
- fprintf(stderr, "Error: Expected 8-bit\n");
- exit(1);
+ {
+ float *npix = (float*)malloc(dinfo.comp_info[0].width_in_blocks * DCTSIZE * nh0 * sizeof(float));
+ vscale(data_y, npix, dinfo.comp_info[0].width_in_blocks * DCTSIZE, dinfo.comp_info[0].height_in_blocks * DCTSIZE, nh0, dinfo.comp_info[0].width_in_blocks * DCTSIZE);
+ data_ny = (unsigned char *)malloc(nw0 * stride0);
+ hscale(npix, data_ny, dinfo.comp_info[0].width_in_blocks * DCTSIZE, nh0, nw0, stride0);
+ free(npix);
}
-
- pix = (unsigned char *)malloc(w * h);
- if (pix == NULL) {
- perror("malloc");
- exit(1);
+ {
+ float *npix = (float*)malloc(dinfo.comp_info[1].width_in_blocks * DCTSIZE * nh1 * sizeof(float));
+ vscale(data_cr, npix, dinfo.comp_info[1].width_in_blocks * DCTSIZE, dinfo.comp_info[1].height_in_blocks * DCTSIZE, nh1, dinfo.comp_info[1].width_in_blocks * DCTSIZE);
+ data_ncr = (unsigned char *)malloc(nw1 * stride1);
+ hscale(npix, data_ncr, dinfo.comp_info[1].width_in_blocks * DCTSIZE, nh1, nw1, stride1);
+ free(npix);
}
-
- if (fread(pix, w*h, 1, stdin) != 1) {
- fprintf(stderr, "Error: Short read\n");
- exit(1);
+ {
+ float *npix = (float*)malloc(dinfo.comp_info[2].width_in_blocks * DCTSIZE * nh2 * sizeof(float));
+ vscale(data_cb, npix, dinfo.comp_info[2].width_in_blocks * DCTSIZE, dinfo.comp_info[2].height_in_blocks * DCTSIZE, nh2, dinfo.comp_info[2].width_in_blocks * DCTSIZE);
+ data_ncb = (unsigned char *)malloc(nw2 * stride2);
+ hscale(npix, data_ncb, dinfo.comp_info[2].width_in_blocks * DCTSIZE, nh2, nw2, stride2);
+ free(npix);
}
+ jpeg_destroy_decompress(&dinfo);
- npix = (float *)malloc(sizeof(float) * nw * h);
- if (npix == NULL) {
- perror("malloc");
- exit(1);
- }
- npix2 = (unsigned char *)malloc(nw * nh);
- if (npix2 == NULL) {
- perror("malloc");
- exit(1);
- }
-
- hscale(pix, npix, w, h, nw);
- vscale(npix, npix2, nw, h, nh);
-
- printf("P5\n# COMMENT: Made by qscale\n%u %u\n255\n", nw, nh);
- {
- int y, x;
- for (y = 0; y < nh; ++y) {
- for (x = 0; x < nw; ++x) {
- putchar(npix2[y * nw + x]);
- }
+ struct jpeg_compress_struct cinfo;
+ cinfo.err = jpeg_std_error(&jerr);
+ jpeg_create_compress(&cinfo);
+ jpeg_stdio_dest(&cinfo, stdout);
+ cinfo.input_components = 3;
+ jpeg_set_defaults(&cinfo);
+ jpeg_set_quality(&cinfo, 85, FALSE);
+ cinfo.image_width = nominal_w;
+ cinfo.image_height = nominal_h;
+ cinfo.raw_data_in = TRUE;
+ jpeg_set_colorspace(&cinfo, JCS_YCbCr);
+ cinfo.comp_info[0].h_samp_factor = samp_h0;
+ cinfo.comp_info[0].v_samp_factor = samp_v0;
+ cinfo.comp_info[1].h_samp_factor = samp_h1;
+ cinfo.comp_info[1].v_samp_factor = samp_v1;
+ cinfo.comp_info[2].h_samp_factor = samp_h2;
+ cinfo.comp_info[2].v_samp_factor = samp_v2;
+ jpeg_start_compress(&cinfo, TRUE);
+
+ total_lines = 0;
+ blocks = 0;
+ while (total_lines < cinfo.comp_info[0].height_in_blocks * DCTSIZE) {
+ unsigned max_lines = cinfo.max_v_samp_factor * DCTSIZE;
+
+ JSAMPROW y_row_ptrs[max_lines];
+ JSAMPROW cb_row_ptrs[max_lines];
+ JSAMPROW cr_row_ptrs[max_lines];
+ JSAMPROW* ptrs[] = { y_row_ptrs, cb_row_ptrs, cr_row_ptrs };
+ int i;
+
+ for (i = 0; i < max_lines; ++i) {
+ // simple edge extension
+ int yline = i + blocks*DCTSIZE*cinfo.comp_info[0].v_samp_factor;
+ if (yline > nh0 - 1)
+ yline = nh0 - 1;
+
+ int cbline = i + blocks*DCTSIZE*cinfo.comp_info[1].v_samp_factor;
+ if (cbline > nh1 - 1)
+ cbline = nh1 - 1;
+
+ int crline = i + blocks*DCTSIZE*cinfo.comp_info[2].v_samp_factor;
+ if (crline > nh2 - 1)
+ crline = nh2 - 1;
+
+ y_row_ptrs[i] = data_ny + yline * stride0;
+ cb_row_ptrs[i] = data_ncb + cbline * stride1;
+ cr_row_ptrs[i] = data_ncr + crline * stride2;
}
+
+ total_lines += max_lines;
+ ++blocks;
+
+ jpeg_write_raw_data(&cinfo, ptrs, max_lines);
}
+ jpeg_finish_compress(&cinfo);
+ jpeg_destroy_compress(&cinfo);
return 0;
}
+