X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=qscale.c;h=d89c8b9463d9a34cc367a22f72f4e4b9f75a7b7d;hb=91eb477176f30904d146d5afb7ec5cbc3b00f735;hp=ee70208c28e5307511324c73f6a3800084cb6ad8;hpb=079f270e1fd5c82ecafb0c473d4196af21a24111;p=qscale diff --git a/qscale.c b/qscale.c index ee70208..d89c8b9 100644 --- a/qscale.c +++ b/qscale.c @@ -1,3 +1,21 @@ +/* + * qscale: Quick, high-quality JPEG-to-JPEG scaler. + * Copyright (C) 2008 Steinar H. Gunderson + * + * 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 #include #include @@ -5,8 +23,20 @@ #include #include "jpeglib.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 1 + +/* Whether to use SSE for vertical scaling or not (requires only SSE1). */ +#define USE_VERTICAL_SSE 1 + +#if USE_VERTICAL_SSE +#undef CACHE_LINE_FACTOR +#define CACHE_LINE_FACTOR 16 +#endif + #ifndef M_PI #define M_PI 3.14159265358979323846264 #endif @@ -66,6 +96,7 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw 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) { @@ -78,6 +109,7 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw end += 4 - (num % 4); } } +#endif pd[x].start = start; pd[x].end = end; @@ -105,6 +137,7 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw 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; @@ -136,7 +169,16 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw : "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) @@ -145,7 +187,6 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw ch = (unsigned char)acc; *dptr++ = ch; #endif - *dptr++ = (unsigned char)result; } ch = dptr[-1]; for ( ; x < dstride; ++x) { @@ -203,25 +244,7 @@ void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh 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 @@ -309,6 +332,24 @@ void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh "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; } @@ -336,13 +377,27 @@ void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh 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; @@ -406,21 +461,21 @@ int main(int argc, char **argv) { 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); + data_ny = (unsigned char *)malloc(nh0 * stride0); hscale(npix, data_ny, w0, nh0, nw0, dinfo.comp_info[0].width_in_blocks * DCTSIZE, stride0); 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); + data_ncr = (unsigned char *)malloc(nh1 * stride1); hscale(npix, data_ncr, w1, nh1, nw1, dinfo.comp_info[1].width_in_blocks * DCTSIZE, stride1); 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); + data_ncb = (unsigned char *)malloc(nh2 * stride2); hscale(npix, data_ncb, w2, nh2, nw2, dinfo.comp_info[2].width_in_blocks * DCTSIZE, stride2); free(npix); } @@ -432,7 +487,7 @@ int main(int argc, char **argv) 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;