/* * High quality image resampling with polyphase filters * Copyright (c) 2001 Fabrice Bellard. * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library 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 * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ /** * @file imgresample.c * High quality image resampling with polyphase filters . */ #include "avcodec.h" #include "dsputil.h" #ifdef USE_FASTMEMCPY #include "fastmemcpy.h" #endif #define NB_COMPONENTS 3 #define PHASE_BITS 4 #define NB_PHASES (1 << PHASE_BITS) #define NB_TAPS 4 #define FCENTER 1 /* index of the center of the filter */ //#define TEST 1 /* Test it */ #define POS_FRAC_BITS 16 #define POS_FRAC (1 << POS_FRAC_BITS) /* 6 bits precision is needed for MMX */ #define FILTER_BITS 8 #define LINE_BUF_HEIGHT (NB_TAPS * 4) struct ImgReSampleContext { int iwidth, iheight, owidth, oheight, topBand, bottomBand, leftBand, rightBand; int h_incr, v_incr; int16_t h_filters[NB_PHASES][NB_TAPS] __align8; /* horizontal filters */ int16_t v_filters[NB_PHASES][NB_TAPS] __align8; /* vertical filters */ uint8_t *line_buf; }; static inline int get_phase(int pos) { return ((pos) >> (POS_FRAC_BITS - PHASE_BITS)) & ((1 << PHASE_BITS) - 1); } /* This function must be optimized */ static void h_resample_fast(uint8_t *dst, int dst_width, const uint8_t *src, int src_width, int src_start, int src_incr, int16_t *filters) { int src_pos, phase, sum, i; const uint8_t *s; int16_t *filter; src_pos = src_start; for(i=0;i<dst_width;i++) { #ifdef TEST /* test */ if ((src_pos >> POS_FRAC_BITS) < 0 || (src_pos >> POS_FRAC_BITS) > (src_width - NB_TAPS)) av_abort(); #endif s = src + (src_pos >> POS_FRAC_BITS); phase = get_phase(src_pos); filter = filters + phase * NB_TAPS; #if NB_TAPS == 4 sum = s[0] * filter[0] + s[1] * filter[1] + s[2] * filter[2] + s[3] * filter[3]; #else { int j; sum = 0; for(j=0;j<NB_TAPS;j++) sum += s[j] * filter[j]; } #endif sum = sum >> FILTER_BITS; if (sum < 0) sum = 0; else if (sum > 255) sum = 255; dst[0] = sum; src_pos += src_incr; dst++; } } /* This function must be optimized */ static void v_resample(uint8_t *dst, int dst_width, const uint8_t *src, int wrap, int16_t *filter) { int sum, i; const uint8_t *s; s = src; for(i=0;i<dst_width;i++) { #if NB_TAPS == 4 sum = s[0 * wrap] * filter[0] + s[1 * wrap] * filter[1] + s[2 * wrap] * filter[2] + s[3 * wrap] * filter[3]; #else { int j; uint8_t *s1 = s; sum = 0; for(j=0;j<NB_TAPS;j++) { sum += s1[0] * filter[j]; s1 += wrap; } } #endif sum = sum >> FILTER_BITS; if (sum < 0) sum = 0; else if (sum > 255) sum = 255; dst[0] = sum; dst++; s++; } } #ifdef HAVE_MMX #include "i386/mmx.h" #define FILTER4(reg) \ {\ s = src + (src_pos >> POS_FRAC_BITS);\ phase = get_phase(src_pos);\ filter = filters + phase * NB_TAPS;\ movq_m2r(*s, reg);\ punpcklbw_r2r(mm7, reg);\ movq_m2r(*filter, mm6);\ pmaddwd_r2r(reg, mm6);\ movq_r2r(mm6, reg);\ psrlq_i2r(32, reg);\ paddd_r2r(mm6, reg);\ psrad_i2r(FILTER_BITS, reg);\ src_pos += src_incr;\ } #define DUMP(reg) movq_r2m(reg, tmp); printf(#reg "=%016Lx\n", tmp.uq); /* XXX: do four pixels at a time */ static void h_resample_fast4_mmx(uint8_t *dst, int dst_width, const uint8_t *src, int src_width, int src_start, int src_incr, int16_t *filters) { int src_pos, phase; const uint8_t *s; int16_t *filter; mmx_t tmp; src_pos = src_start; pxor_r2r(mm7, mm7); while (dst_width >= 4) { FILTER4(mm0); FILTER4(mm1); FILTER4(mm2); FILTER4(mm3); packuswb_r2r(mm7, mm0); packuswb_r2r(mm7, mm1); packuswb_r2r(mm7, mm3); packuswb_r2r(mm7, mm2); movq_r2m(mm0, tmp); dst[0] = tmp.ub[0]; movq_r2m(mm1, tmp); dst[1] = tmp.ub[0]; movq_r2m(mm2, tmp); dst[2] = tmp.ub[0]; movq_r2m(mm3, tmp); dst[3] = tmp.ub[0]; dst += 4; dst_width -= 4; } while (dst_width > 0) { FILTER4(mm0); packuswb_r2r(mm7, mm0); movq_r2m(mm0, tmp); dst[0] = tmp.ub[0]; dst++; dst_width--; } emms(); } static void v_resample4_mmx(uint8_t *dst, int dst_width, const uint8_t *src, int wrap, int16_t *filter) { int sum, i, v; const uint8_t *s; mmx_t tmp; mmx_t coefs[4]; for(i=0;i<4;i++) { v = filter[i]; coefs[i].uw[0] = v; coefs[i].uw[1] = v; coefs[i].uw[2] = v; coefs[i].uw[3] = v; } pxor_r2r(mm7, mm7); s = src; while (dst_width >= 4) { movq_m2r(s[0 * wrap], mm0); punpcklbw_r2r(mm7, mm0); movq_m2r(s[1 * wrap], mm1); punpcklbw_r2r(mm7, mm1); movq_m2r(s[2 * wrap], mm2); punpcklbw_r2r(mm7, mm2); movq_m2r(s[3 * wrap], mm3); punpcklbw_r2r(mm7, mm3); pmullw_m2r(coefs[0], mm0); pmullw_m2r(coefs[1], mm1); pmullw_m2r(coefs[2], mm2); pmullw_m2r(coefs[3], mm3); paddw_r2r(mm1, mm0); paddw_r2r(mm3, mm2); paddw_r2r(mm2, mm0); psraw_i2r(FILTER_BITS, mm0); packuswb_r2r(mm7, mm0); movq_r2m(mm0, tmp); *(uint32_t *)dst = tmp.ud[0]; dst += 4; s += 4; dst_width -= 4; } while (dst_width > 0) { sum = s[0 * wrap] * filter[0] + s[1 * wrap] * filter[1] + s[2 * wrap] * filter[2] + s[3 * wrap] * filter[3]; sum = sum >> FILTER_BITS; if (sum < 0) sum = 0; else if (sum > 255) sum = 255; dst[0] = sum; dst++; s++; dst_width--; } emms(); } #endif #ifdef HAVE_ALTIVEC typedef union { vector unsigned char v; unsigned char c[16]; } vec_uc_t; typedef union { vector signed short v; signed short s[8]; } vec_ss_t; void v_resample16_altivec(uint8_t *dst, int dst_width, const uint8_t *src, int wrap, int16_t *filter) { int sum, i; const uint8_t *s; vector unsigned char *tv, tmp, dstv, zero; vec_ss_t srchv[4], srclv[4], fv[4]; vector signed short zeros, sumhv, sumlv; s = src; for(i=0;i<4;i++) { /* The vec_madds later on does an implicit >>15 on the result. Since FILTER_BITS is 8, and we have 15 bits of magnitude in a signed short, we have just enough bits to pre-shift our filter constants <<7 to compensate for vec_madds. */ fv[i].s[0] = filter[i] << (15-FILTER_BITS); fv[i].v = vec_splat(fv[i].v, 0); } zero = vec_splat_u8(0); zeros = vec_splat_s16(0); /* When we're resampling, we'd ideally like both our input buffers, and output buffers to be 16-byte aligned, so we can do both aligned reads and writes. Sadly we can't always have this at the moment, so we opt for aligned writes, as unaligned writes have a huge overhead. To do this, do enough scalar resamples to get dst 16-byte aligned. */ i = (-(int)dst) & 0xf; while(i>0) { sum = s[0 * wrap] * filter[0] + s[1 * wrap] * filter[1] + s[2 * wrap] * filter[2] + s[3 * wrap] * filter[3]; sum = sum >> FILTER_BITS; if (sum<0) sum = 0; else if (sum>255) sum=255; dst[0] = sum; dst++; s++; dst_width--; i--; } /* Do our altivec resampling on 16 pixels at once. */ while(dst_width>=16) { /* Read 16 (potentially unaligned) bytes from each of 4 lines into 4 vectors, and split them into shorts. Interleave the multipy/accumulate for the resample filter with the loads to hide the 3 cycle latency the vec_madds have. */ tv = (vector unsigned char *) &s[0 * wrap]; tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[i * wrap])); srchv[0].v = (vector signed short) vec_mergeh(zero, tmp); srclv[0].v = (vector signed short) vec_mergel(zero, tmp); sumhv = vec_madds(srchv[0].v, fv[0].v, zeros); sumlv = vec_madds(srclv[0].v, fv[0].v, zeros); tv = (vector unsigned char *) &s[1 * wrap]; tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[1 * wrap])); srchv[1].v = (vector signed short) vec_mergeh(zero, tmp); srclv[1].v = (vector signed short) vec_mergel(zero, tmp); sumhv = vec_madds(srchv[1].v, fv[1].v, sumhv); sumlv = vec_madds(srclv[1].v, fv[1].v, sumlv); tv = (vector unsigned char *) &s[2 * wrap]; tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[2 * wrap])); srchv[2].v = (vector signed short) vec_mergeh(zero, tmp); srclv[2].v = (vector signed short) vec_mergel(zero, tmp); sumhv = vec_madds(srchv[2].v, fv[2].v, sumhv); sumlv = vec_madds(srclv[2].v, fv[2].v, sumlv); tv = (vector unsigned char *) &s[3 * wrap]; tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[3 * wrap])); srchv[3].v = (vector signed short) vec_mergeh(zero, tmp); srclv[3].v = (vector signed short) vec_mergel(zero, tmp); sumhv = vec_madds(srchv[3].v, fv[3].v, sumhv); sumlv = vec_madds(srclv[3].v, fv[3].v, sumlv); /* Pack the results into our destination vector, and do an aligned write of that back to memory. */ dstv = vec_packsu(sumhv, sumlv) ; vec_st(dstv, 0, (vector unsigned char *) dst); dst+=16; s+=16; dst_width-=16; } /* If there are any leftover pixels, resample them with the slow scalar method. */ while(dst_width>0) { sum = s[0 * wrap] * filter[0] + s[1 * wrap] * filter[1] + s[2 * wrap] * filter[2] + s[3 * wrap] * filter[3]; sum = sum >> FILTER_BITS; if (sum<0) sum = 0; else if (sum>255) sum=255; dst[0] = sum; dst++; s++; dst_width--; } } #endif /* slow version to handle limit cases. Does not need optimisation */ static void h_resample_slow(uint8_t *dst, int dst_width, const uint8_t *src, int src_width, int src_start, int src_incr, int16_t *filters) { int src_pos, phase, sum, j, v, i; const uint8_t *s, *src_end; int16_t *filter; src_end = src + src_width; src_pos = src_start; for(i=0;i<dst_width;i++) { s = src + (src_pos >> POS_FRAC_BITS); phase = get_phase(src_pos); filter = filters + phase * NB_TAPS; sum = 0; for(j=0;j<NB_TAPS;j++) { if (s < src) v = src[0]; else if (s >= src_end) v = src_end[-1]; else v = s[0]; sum += v * filter[j]; s++; } sum = sum >> FILTER_BITS; if (sum < 0) sum = 0; else if (sum > 255) sum = 255; dst[0] = sum; src_pos += src_incr; dst++; } } static void h_resample(uint8_t *dst, int dst_width, const uint8_t *src, int src_width, int src_start, int src_incr, int16_t *filters) { int n, src_end; if (src_start < 0) { n = (0 - src_start + src_incr - 1) / src_incr; h_resample_slow(dst, n, src, src_width, src_start, src_incr, filters); dst += n; dst_width -= n; src_start += n * src_incr; } src_end = src_start + dst_width * src_incr; if (src_end > ((src_width - NB_TAPS) << POS_FRAC_BITS)) { n = (((src_width - NB_TAPS + 1) << POS_FRAC_BITS) - 1 - src_start) / src_incr; } else { n = dst_width; } #ifdef HAVE_MMX if ((mm_flags & MM_MMX) && NB_TAPS == 4) h_resample_fast4_mmx(dst, n, src, src_width, src_start, src_incr, filters); else #endif h_resample_fast(dst, n, src, src_width, src_start, src_incr, filters); if (n < dst_width) { dst += n; dst_width -= n; src_start += n * src_incr; h_resample_slow(dst, dst_width, src, src_width, src_start, src_incr, filters); } } static void component_resample(ImgReSampleContext *s, uint8_t *output, int owrap, int owidth, int oheight, uint8_t *input, int iwrap, int iwidth, int iheight) { int src_y, src_y1, last_src_y, ring_y, phase_y, y1, y; uint8_t *new_line, *src_line; last_src_y = - FCENTER - 1; /* position of the bottom of the filter in the source image */ src_y = (last_src_y + NB_TAPS) * POS_FRAC; ring_y = NB_TAPS; /* position in ring buffer */ for(y=0;y<oheight;y++) { /* apply horizontal filter on new lines from input if needed */ src_y1 = src_y >> POS_FRAC_BITS; while (last_src_y < src_y1) { if (++ring_y >= LINE_BUF_HEIGHT + NB_TAPS) ring_y = NB_TAPS; last_src_y++; /* handle limit conditions : replicate line (slightly inefficient because we filter multiple times) */ y1 = last_src_y; if (y1 < 0) { y1 = 0; } else if (y1 >= iheight) { y1 = iheight - 1; } src_line = input + y1 * iwrap; new_line = s->line_buf + ring_y * owidth; /* apply filter and handle limit cases correctly */ h_resample(new_line, owidth, src_line, iwidth, - FCENTER * POS_FRAC, s->h_incr, &s->h_filters[0][0]); /* handle ring buffer wraping */ if (ring_y >= LINE_BUF_HEIGHT) { memcpy(s->line_buf + (ring_y - LINE_BUF_HEIGHT) * owidth, new_line, owidth); } } /* apply vertical filter */ phase_y = get_phase(src_y); #ifdef HAVE_MMX /* desactivated MMX because loss of precision */ if ((mm_flags & MM_MMX) && NB_TAPS == 4 && 0) v_resample4_mmx(output, owidth, s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth, &s->v_filters[phase_y][0]); else #endif #ifdef HAVE_ALTIVEC if ((mm_flags & MM_ALTIVEC) && NB_TAPS == 4 && FILTER_BITS <= 6) v_resample16_altivec(output, owidth, s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth, &s->v_filters[phase_y][0]); else #endif v_resample(output, owidth, s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth, &s->v_filters[phase_y][0]); src_y += s->v_incr; output += owrap; } } /* XXX: the following filter is quite naive, but it seems to suffice for 4 taps */ static void build_filter(int16_t *filter, float factor) { int ph, i, v; float x, y, tab[NB_TAPS], norm, mult; /* if upsampling, only need to interpolate, no filter */ if (factor > 1.0) factor = 1.0; for(ph=0;ph<NB_PHASES;ph++) { norm = 0; for(i=0;i<NB_TAPS;i++) { x = M_PI * ((float)(i - FCENTER) - (float)ph / NB_PHASES) * factor; if (x == 0) y = 1.0; else y = sin(x) / x; tab[i] = y; norm += y; } /* normalize so that an uniform color remains the same */ mult = (float)(1 << FILTER_BITS) / norm; for(i=0;i<NB_TAPS;i++) { v = (int)(tab[i] * mult); filter[ph * NB_TAPS + i] = v; } } } ImgReSampleContext *img_resample_init(int owidth, int oheight, int iwidth, int iheight) { return img_resample_full_init(owidth, oheight, iwidth, iheight, 0, 0, 0, 0); } ImgReSampleContext *img_resample_full_init(int owidth, int oheight, int iwidth, int iheight, int topBand, int bottomBand, int leftBand, int rightBand) { ImgReSampleContext *s; s = av_mallocz(sizeof(ImgReSampleContext)); if (!s) return NULL; s->line_buf = av_mallocz(owidth * (LINE_BUF_HEIGHT + NB_TAPS)); if (!s->line_buf) goto fail; s->owidth = owidth; s->oheight = oheight; s->iwidth = iwidth; s->iheight = iheight; s->topBand = topBand; s->bottomBand = bottomBand; s->leftBand = leftBand; s->rightBand = rightBand; s->h_incr = ((iwidth - leftBand - rightBand) * POS_FRAC) / owidth; s->v_incr = ((iheight - topBand - bottomBand) * POS_FRAC) / oheight; build_filter(&s->h_filters[0][0], (float) owidth / (float) (iwidth - leftBand - rightBand)); build_filter(&s->v_filters[0][0], (float) oheight / (float) (iheight - topBand - bottomBand)); return s; fail: av_free(s); return NULL; } void img_resample(ImgReSampleContext *s, AVPicture *output, const AVPicture *input) { int i, shift; for(i=0;i<3;i++) { shift = (i == 0) ? 0 : 1; component_resample(s, output->data[i], output->linesize[i], s->owidth >> shift, s->oheight >> shift, input->data[i] + (input->linesize[i] * (s->topBand >> shift)) + (s->leftBand >> shift), input->linesize[i], ((s->iwidth - s->leftBand - s->rightBand) >> shift), (s->iheight - s->topBand - s->bottomBand) >> shift); } } void img_resample_close(ImgReSampleContext *s) { av_free(s->line_buf); av_free(s); } #ifdef TEST void *av_mallocz(int size) { void *ptr; ptr = malloc(size); memset(ptr, 0, size); return ptr; } void av_free(void *ptr) { /* XXX: this test should not be needed on most libcs */ if (ptr) free(ptr); } /* input */ #define XSIZE 256 #define YSIZE 256 uint8_t img[XSIZE * YSIZE]; /* output */ #define XSIZE1 512 #define YSIZE1 512 uint8_t img1[XSIZE1 * YSIZE1]; uint8_t img2[XSIZE1 * YSIZE1]; void save_pgm(const char *filename, uint8_t *img, int xsize, int ysize) { FILE *f; f=fopen(filename,"w"); fprintf(f,"P5\n%d %d\n%d\n", xsize, ysize, 255); fwrite(img,1, xsize * ysize,f); fclose(f); } static void dump_filter(int16_t *filter) { int i, ph; for(ph=0;ph<NB_PHASES;ph++) { printf("%2d: ", ph); for(i=0;i<NB_TAPS;i++) { printf(" %5.2f", filter[ph * NB_TAPS + i] / 256.0); } printf("\n"); } } #ifdef HAVE_MMX int mm_flags; #endif int main(int argc, char **argv) { int x, y, v, i, xsize, ysize; ImgReSampleContext *s; float fact, factors[] = { 1/2.0, 3.0/4.0, 1.0, 4.0/3.0, 16.0/9.0, 2.0 }; char buf[256]; /* build test image */ for(y=0;y<YSIZE;y++) { for(x=0;x<XSIZE;x++) { if (x < XSIZE/2 && y < YSIZE/2) { if (x < XSIZE/4 && y < YSIZE/4) { if ((x % 10) <= 6 && (y % 10) <= 6) v = 0xff; else v = 0x00; } else if (x < XSIZE/4) { if (x & 1) v = 0xff; else v = 0; } else if (y < XSIZE/4) { if (y & 1) v = 0xff; else v = 0; } else { if (y < YSIZE*3/8) { if ((y+x) & 1) v = 0xff; else v = 0; } else { if (((x+3) % 4) <= 1 && ((y+3) % 4) <= 1) v = 0xff; else v = 0x00; } } } else if (x < XSIZE/2) { v = ((x - (XSIZE/2)) * 255) / (XSIZE/2); } else if (y < XSIZE/2) { v = ((y - (XSIZE/2)) * 255) / (XSIZE/2); } else { v = ((x + y - XSIZE) * 255) / XSIZE; } img[(YSIZE - y) * XSIZE + (XSIZE - x)] = v; } } save_pgm("/tmp/in.pgm", img, XSIZE, YSIZE); for(i=0;i<sizeof(factors)/sizeof(float);i++) { fact = factors[i]; xsize = (int)(XSIZE * fact); ysize = (int)((YSIZE - 100) * fact); s = img_resample_full_init(xsize, ysize, XSIZE, YSIZE, 50 ,50, 0, 0); printf("Factor=%0.2f\n", fact); dump_filter(&s->h_filters[0][0]); component_resample(s, img1, xsize, xsize, ysize, img + 50 * XSIZE, XSIZE, XSIZE, YSIZE - 100); img_resample_close(s); sprintf(buf, "/tmp/out%d.pgm", i); save_pgm(buf, img1, xsize, ysize); } /* mmx test */ #ifdef HAVE_MMX printf("MMX test\n"); fact = 0.72; xsize = (int)(XSIZE * fact); ysize = (int)(YSIZE * fact); mm_flags = MM_MMX; s = img_resample_init(xsize, ysize, XSIZE, YSIZE); component_resample(s, img1, xsize, xsize, ysize, img, XSIZE, XSIZE, YSIZE); mm_flags = 0; s = img_resample_init(xsize, ysize, XSIZE, YSIZE); component_resample(s, img2, xsize, xsize, ysize, img, XSIZE, XSIZE, YSIZE); if (memcmp(img1, img2, xsize * ysize) != 0) { fprintf(stderr, "mmx error\n"); exit(1); } printf("MMX OK\n"); #endif return 0; } #endif