+/*
+ * qscale: Quick, high-quality JPEG-to-JPEG scaler.
+ * Copyright (C) 2008 Steinar H. Gunderson <sgunderson@bigfoot.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
+ */
+
#include <stdio.h>
#include <malloc.h>
#include <math.h>
#include <string.h>
#include <stdlib.h>
#include "jpeglib.h"
+#include "libqscale.h"
+/* The number of pixels to process at a time when scaling vertically. */
#define CACHE_LINE_FACTOR 16
+/* Whether to use SSE for horizontal scaling or not (requires SSE3). */
+#define USE_HORIZONTAL_SSE 0
+
+/* Whether to use SSE for vertical scaling or not (requires only SSE1). */
+#define USE_VERTICAL_SSE 0
+
+#if USE_VERTICAL_SSE
+#undef CACHE_LINE_FACTOR
+#define CACHE_LINE_FACTOR 16
+#endif
+
#ifndef M_PI
#define M_PI 3.14159265358979323846264
#endif
end = w - 1;
}
+#if USE_HORIZONTAL_SSE
/* round up so we get a multiple of four for the SSE code */
int num = (end - start + 1);
if (num % 4 != 0) {
end += 4 - (num % 4);
}
}
+#endif
pd[x].start = start;
pd[x].end = end;
unsigned char *dptr = npix + y*dstride;
unsigned char ch;
for (x = 0; x < nw; ++x) {
+#if USE_HORIZONTAL_SSE
int result;
float acc;
long tmp;
: "memory", "xmm1", "xmm2"
);
-#if 0
+ *dptr++ = (unsigned char)result;
+#else
+ float acc = 0.0;
+ float *cf = &coeffs[pd[x].startcoeff];
+ unsigned sx;
+
+ for (sx = pd[x].start; sx <= pd[x].end; ++sx) {
+ acc += sptr[sx] * *cf++;
+ }
+
if (acc < 0.0)
ch = 0;
else if (acc > 255.0)
ch = (unsigned char)acc;
*dptr++ = ch;
#endif
- *dptr++ = (unsigned char)result;
}
ch = dptr[-1];
for ( ; x < dstride; ++x) {
unsigned char *sptr = pix + x;
float *dptr = npix + x;
for (y = 0; y < nh; ++y) {
-#if 0
- int i;
- float acc[CACHE_LINE_FACTOR];
- for (i = 0; i < CACHE_LINE_FACTOR; ++i)
- acc[i] = 0.0;
- float *cf = &coeffs[pd[y].startcoeff];
- unsigned sy;
-
- for (sy = pd[y].start; sy <= pd[y].end; ++sy) {
- for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
- acc[i] += sptr[sy * w + i] * *cf;
- }
- ++cf;
- }
-
- for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
- dptr[i] = acc[i];
- }
-#else
+#if USE_VERTICAL_SSE
/*
* xmm0 - xmm3: acc[0..15]
* xmm4: current filter coefficient
"r" (dptr) /* 4: dstptr base */
: "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
);
+#else
+ int i;
+ float acc[CACHE_LINE_FACTOR];
+ for (i = 0; i < CACHE_LINE_FACTOR; ++i)
+ acc[i] = 0.0;
+ float *cf = &coeffs[pd[y].startcoeff];
+ unsigned sy;
+
+ for (sy = pd[y].start; sy <= pd[y].end; ++sy) {
+ for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
+ acc[i] += sptr[sy * w + i] * *cf;
+ }
+ ++cf;
+ }
+
+ for (i = 0; i < CACHE_LINE_FACTOR; ++i) {
+ dptr[i] = acc[i];
+ }
#endif
dptr += dstride;
}
int main(int argc, char **argv)
{
+ /* user-settable parameters */
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 jpeg_quality = 85;
+ /* end */
+
+ unsigned max_samp_h, max_samp_v;
+ max_samp_h = samp_h0;
+ if (samp_h1 > max_samp_h)
+ max_samp_h = samp_h1;
+ if (samp_h2 > max_samp_h)
+ max_samp_h = samp_h2;
+
+ max_samp_v = samp_v0;
+ if (samp_v1 > max_samp_v)
+ max_samp_v = samp_v1;
+ if (samp_v2 > max_samp_v)
+ max_samp_v = samp_v2;
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);
+ unsigned dstride0 = (nw0 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned dstride1 = (nw1 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned dstride2 = (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);
-
- unsigned w0 = dinfo.image_width * dinfo.comp_info[0].h_samp_factor / dinfo.max_h_samp_factor;
- unsigned h0 = dinfo.image_height * dinfo.comp_info[0].v_samp_factor / dinfo.max_v_samp_factor;
-
- unsigned w1 = dinfo.image_width * dinfo.comp_info[1].h_samp_factor / dinfo.max_h_samp_factor;
- unsigned h1 = dinfo.image_height * dinfo.comp_info[1].v_samp_factor / dinfo.max_v_samp_factor;
-
- unsigned w2 = dinfo.image_width * dinfo.comp_info[2].h_samp_factor / dinfo.max_h_samp_factor;
- unsigned h2 = dinfo.image_height * dinfo.comp_info[2].v_samp_factor / dinfo.max_v_samp_factor;
-
- 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*)memalign(16, dinfo.comp_info[0].height_in_blocks * dinfo.comp_info[0].width_in_blocks * DCTSIZE * DCTSIZE);
- JSAMPLE *data_cb = (JSAMPLE*)memalign(16, dinfo.comp_info[1].height_in_blocks * dinfo.comp_info[1].width_in_blocks * DCTSIZE * DCTSIZE);
- JSAMPLE *data_cr = (JSAMPLE*)memalign(16, dinfo.comp_info[2].height_in_blocks * dinfo.comp_info[2].width_in_blocks * DCTSIZE * DCTSIZE);
+ qscale_img *img = qscale_load_jpeg_from_stdio(stdin);
+ unsigned sstride0 = (img->w0 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned sstride1 = (img->w1 + DCTSIZE-1) & ~(DCTSIZE-1);
+ unsigned sstride2 = (img->w2 + DCTSIZE-1) & ~(DCTSIZE-1);
+
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 (jpeg_read_raw_data(&dinfo, ptrs, max_lines) == 0)
- break;
- }
-
{
- float *npix = (float*)memalign(16, dinfo.comp_info[0].width_in_blocks * DCTSIZE * nh0 * sizeof(float));
- vscale(data_y, npix, dinfo.comp_info[0].width_in_blocks * DCTSIZE, h0, nh0, dinfo.comp_info[0].width_in_blocks * DCTSIZE);
- data_ny = (unsigned char *)malloc(nw0 * stride0);
- hscale(npix, data_ny, w0, nh0, nw0, dinfo.comp_info[0].width_in_blocks * DCTSIZE, stride0);
+ float *npix = (float*)memalign(16, sstride0 * nh0 * sizeof(float));
+ vscale(img->data_y, npix, sstride0, img->h0, nh0, sstride0);
+ data_ny = (unsigned char *)malloc(nh0 * dstride0);
+ hscale(npix, data_ny, img->w0, nh0, nw0, sstride0, dstride0);
free(npix);
}
{
- float *npix = (float*)memalign(16, dinfo.comp_info[1].width_in_blocks * DCTSIZE * nh1 * sizeof(float));
- vscale(data_cr, npix, dinfo.comp_info[1].width_in_blocks * DCTSIZE, h1, nh1, dinfo.comp_info[1].width_in_blocks * DCTSIZE);
- data_ncr = (unsigned char *)malloc(nw1 * stride1);
- hscale(npix, data_ncr, w1, nh1, nw1, dinfo.comp_info[1].width_in_blocks * DCTSIZE, stride1);
+ float *npix = (float*)memalign(16, sstride1 * nh1 * sizeof(float));
+ vscale(img->data_cr, npix, sstride1, img->h1, nh1, sstride1);
+ data_ncr = (unsigned char *)malloc(nh1 * dstride1);
+ hscale(npix, data_ncr, img->w1, nh1, nw1, sstride1, dstride1);
free(npix);
}
{
- float *npix = (float*)memalign(16, dinfo.comp_info[2].width_in_blocks * DCTSIZE * nh2 * sizeof(float));
- vscale(data_cb, npix, dinfo.comp_info[2].width_in_blocks * DCTSIZE, h2, nh2, dinfo.comp_info[2].width_in_blocks * DCTSIZE);
- data_ncb = (unsigned char *)malloc(nw2 * stride2);
- hscale(npix, data_ncb, w2, nh2, nw2, dinfo.comp_info[2].width_in_blocks * DCTSIZE, stride2);
+ float *npix = (float*)memalign(16, sstride2 * nh2 * sizeof(float));
+ vscale(img->data_cb, npix, sstride2, img->h2, nh2, sstride2);
+ data_ncb = (unsigned char *)malloc(nh2 * dstride2);
+ hscale(npix, data_ncb, img->w2, nh2, nw2, sstride2, dstride2);
free(npix);
}
- jpeg_destroy_decompress(&dinfo);
struct jpeg_compress_struct cinfo;
+ struct jpeg_error_mgr jerr;
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);
+ jpeg_set_quality(&cinfo, jpeg_quality, FALSE);
cinfo.image_width = nominal_w;
cinfo.image_height = nominal_h;
cinfo.raw_data_in = TRUE;
cinfo.comp_info[2].v_samp_factor = samp_v2;
jpeg_start_compress(&cinfo, TRUE);
- total_lines = 0;
- blocks = 0;
+ int total_lines = 0;
+ int blocks = 0;
while (total_lines < cinfo.comp_info[0].height_in_blocks * DCTSIZE) {
unsigned max_lines = cinfo.max_v_samp_factor * DCTSIZE;
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;
+ y_row_ptrs[i] = data_ny + yline * dstride0;
+ cb_row_ptrs[i] = data_ncb + cbline * dstride1;
+ cr_row_ptrs[i] = data_ncr + crline * dstride2;
}
total_lines += max_lines;
jpeg_finish_compress(&cinfo);
jpeg_destroy_compress(&cinfo);
+ qscale_destroy(img);
+
return 0;
}