X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libqscale.c;h=1ce7e891fdf0564c70a070bc0555a935a4e45e19;hb=acb34b86d96ef0fa62750897ad83a62af806c057;hp=c26a4418e91746bf337d7552cd6dc9fc513ce4b6;hpb=07342f6e7cf83fbf6707ec370c22431e98b62fd7;p=qscale diff --git a/libqscale.c b/libqscale.c index c26a441..1ce7e89 100644 --- a/libqscale.c +++ b/libqscale.c @@ -42,6 +42,13 @@ #define M_PI 3.14159265358979323846264 #endif +#if USE_VERTICAL_SSE || USE_HORIZONTAL_SSE +typedef float v4sf __attribute__((vector_size(16))); +typedef int v4si __attribute__((vector_size(16))); +typedef short v8hi __attribute__((vector_size(16))); +typedef char v16qi __attribute__((vector_size(16))); +#endif + qscale_img *qscale_load_jpeg(const char *filename) { FILE *file = fopen(filename, "rb"); @@ -75,11 +82,11 @@ qscale_img *qscale_load_jpeg_from_stdio(FILE *file) dinfo.raw_data_out = TRUE; jpeg_start_decompress(&dinfo); - /* We do not handle anything but YCbCr images (yet?). */ - if (dinfo.num_components != 3) { + if (dinfo.num_components != 1 && dinfo.num_components != 3) { qscale_destroy(img); return NULL; } + img->num_components = dinfo.num_components; img->width = dinfo.image_width; img->height = dinfo.image_height; @@ -87,30 +94,40 @@ qscale_img *qscale_load_jpeg_from_stdio(FILE *file) img->w0 = dinfo.image_width * dinfo.comp_info[0].h_samp_factor / dinfo.max_h_samp_factor; img->h0 = dinfo.image_height * dinfo.comp_info[0].v_samp_factor / dinfo.max_v_samp_factor; - img->w1 = dinfo.image_width * dinfo.comp_info[1].h_samp_factor / dinfo.max_h_samp_factor; - img->h1 = dinfo.image_height * dinfo.comp_info[1].v_samp_factor / dinfo.max_v_samp_factor; + if (img->num_components == 3) { + img->w1 = dinfo.image_width * dinfo.comp_info[1].h_samp_factor / dinfo.max_h_samp_factor; + img->h1 = dinfo.image_height * dinfo.comp_info[1].v_samp_factor / dinfo.max_v_samp_factor; - img->w2 = dinfo.image_width * dinfo.comp_info[2].h_samp_factor / dinfo.max_h_samp_factor; - img->h2 = dinfo.image_height * dinfo.comp_info[2].v_samp_factor / dinfo.max_v_samp_factor; + img->w2 = dinfo.image_width * dinfo.comp_info[2].h_samp_factor / dinfo.max_h_samp_factor; + img->h2 = dinfo.image_height * dinfo.comp_info[2].v_samp_factor / dinfo.max_v_samp_factor; + } img->samp_h0 = dinfo.comp_info[0].h_samp_factor; img->samp_v0 = dinfo.comp_info[0].v_samp_factor; - img->samp_h1 = dinfo.comp_info[1].h_samp_factor; - img->samp_v1 = dinfo.comp_info[1].v_samp_factor; + if (img->num_components == 3) { + img->samp_h1 = dinfo.comp_info[1].h_samp_factor; + img->samp_v1 = dinfo.comp_info[1].v_samp_factor; - img->samp_h2 = dinfo.comp_info[2].h_samp_factor; - img->samp_v2 = dinfo.comp_info[2].v_samp_factor; + img->samp_h2 = dinfo.comp_info[2].h_samp_factor; + img->samp_v2 = dinfo.comp_info[2].v_samp_factor; + } img->data_y = (JSAMPLE*)memalign(16, dinfo.comp_info[0].height_in_blocks * dinfo.comp_info[0].width_in_blocks * DCTSIZE * DCTSIZE); - img->data_cb = (JSAMPLE*)memalign(16, dinfo.comp_info[1].height_in_blocks * dinfo.comp_info[1].width_in_blocks * DCTSIZE * DCTSIZE); - img->data_cr = (JSAMPLE*)memalign(16, dinfo.comp_info[2].height_in_blocks * dinfo.comp_info[2].width_in_blocks * DCTSIZE * DCTSIZE); - - if (img->data_y == NULL || img->data_cb == NULL || img->data_cr == NULL) { + if (img->data_y == NULL) { qscale_destroy(img); return NULL; } + if (img->num_components == 3) { + img->data_cb = (JSAMPLE*)memalign(16, dinfo.comp_info[1].height_in_blocks * dinfo.comp_info[1].width_in_blocks * DCTSIZE * DCTSIZE); + img->data_cr = (JSAMPLE*)memalign(16, dinfo.comp_info[2].height_in_blocks * dinfo.comp_info[2].width_in_blocks * DCTSIZE * DCTSIZE); + if (img->data_cb == NULL || img->data_cr == NULL) { + qscale_destroy(img); + return NULL; + } + } + 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; @@ -123,8 +140,10 @@ qscale_img *qscale_load_jpeg_from_stdio(FILE *file) int i; for (i = 0; i < max_lines; ++i) { y_row_ptrs[i] = img->data_y + (i+blocks*DCTSIZE*dinfo.comp_info[0].v_samp_factor) * dinfo.comp_info[0].width_in_blocks * DCTSIZE; - cb_row_ptrs[i] = img->data_cb + (i+blocks*DCTSIZE*dinfo.comp_info[1].v_samp_factor) * dinfo.comp_info[1].width_in_blocks * DCTSIZE; - cr_row_ptrs[i] = img->data_cr + (i+blocks*DCTSIZE*dinfo.comp_info[2].v_samp_factor) * dinfo.comp_info[2].width_in_blocks * DCTSIZE; + if (img->num_components == 3) { + cb_row_ptrs[i] = img->data_cb + (i+blocks*DCTSIZE*dinfo.comp_info[1].v_samp_factor) * dinfo.comp_info[1].width_in_blocks * DCTSIZE; + cr_row_ptrs[i] = img->data_cr + (i+blocks*DCTSIZE*dinfo.comp_info[2].v_samp_factor) * dinfo.comp_info[2].width_in_blocks * DCTSIZE; + } } total_lines += max_lines; @@ -172,12 +191,39 @@ static double lanczos_tap(double x) return sinc(x*M_PI) * sinc(x*M_PI / 3.0); } +static double mitchell_tap(double x) +{ + const double b = 1.0 / 3.0; + const double c = 1.0 / 3.0; + const double p0 = ( 6.0 - 2.0*b ) / 6.0; + const double p2 = (-18.0 + 12.0*b + 6.0*c) / 6.0; + const double p3 = ( 12.0 - 9.0*b - 6.0*c) / 6.0; + const double q0 = ( 8.0*b + 24.0*c) / 6.0; + const double q1 = ( - 12.0*b - 48.0*c) / 6.0; + const double q2 = ( 6.0*b + 30.0*c) / 6.0; + const double q3 = ( - b - 6.0*c) / 6.0; + + if (x < -2.0) { + return 0.0; + } else if (x < -1.0) { + return q0 - x * (q1 - x * (q2 - x * q3)); + } else if (x < 0.0) { + return p0 + x * x * (p2 - x * p3); + } else if (x < 1.0) { + return p0 + x * x * (p2 + x * p3); + } else if (x < 2.0) { + return q0 + x * (q1 + x * (q2 + x * q3)); + } else { + return 0.0; + } +} + struct pix_desc { unsigned start, end; unsigned startcoeff; }; -static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw, unsigned sstride, unsigned dstride) +static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw, unsigned sstride, unsigned dstride, enum qscale_scaling_filter scaling_filter) { struct pix_desc *pd = (struct pix_desc *)malloc(nw * sizeof(struct pix_desc)); int size_coeffs = 8; @@ -185,7 +231,13 @@ static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsi int num_coeffs = 0; int x, y; double sf = (double)w / (double)nw; - double support = (w > nw) ? (3.0 * sf) : (3.0 / sf); + double support; + + if (scaling_filter == LANCZOS) { + support = (w > nw) ? (3.0 * sf) : (3.0 / sf); + } else { /* Mitchell */ + support = (w > nw) ? (2.0 * sf) : (2.0 / sf); + } /* calculate the filter */ for (x = 0; x < nw; ++x) { @@ -222,7 +274,12 @@ static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsi for (sx = start; sx <= end; ++sx) { double nd = (w > nw) ? (sx/sf - x) : (sx - x*sf); - double f = lanczos_tap(nd); + double f; + if (scaling_filter == LANCZOS) { + f = lanczos_tap(nd); + } else { /* Mitchell */ + f = mitchell_tap(nd); + } if (num_coeffs == size_coeffs) { size_coeffs <<= 1; coeffs = (float *)realloc(coeffs, size_coeffs * sizeof(float)); @@ -243,36 +300,26 @@ static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsi unsigned char ch; for (x = 0; x < nw; ++x) { #if USE_HORIZONTAL_SSE + v4sf acc = { 0.0f, 0.0f, 0.0f, 0.0f }; + static const v4sf low = { 0.0f, 0.0f, 0.0f, 0.0f }; + static const v4sf high = { 255.0f, 255.0f, 255.0f, 255.0f }; int result; - float acc; - long tmp; - static const float low = 0.0, high = 255.0; - __asm__ ( - "pxor %1, %1 \n" - "xor %2, %2 \n" - "0: \n" - "movups (%4,%2),%%xmm1 \n" - "movups (%3,%2),%%xmm2 \n" - "mulps %%xmm2,%%xmm1 \n" - "addps %%xmm1,%1 \n" - "add $16,%2 \n" - "dec %5 \n" - "jnz 0b \n" - "haddps %1,%1 \n" - "haddps %1,%1 \n" - "maxss %6,%1 \n" - "minss %7,%1 \n" - "cvtss2si %1,%0 \n" - : "=r" (result), - "=&x" (acc), - "=&r" (tmp) - : "r" (&coeffs[pd[x].startcoeff]), - "r" (&sptr[pd[x].start]), - "r" ((pd[x].end - pd[x].start + 1)/4), - "m" (low), - "m" (high) - : "memory", "xmm1", "xmm2" - ); + int i; + + const float *sptr_xmm = &sptr[pd[x].start]; + const float *coeffptr = &coeffs[pd[x].startcoeff]; + const int filter_len = (pd[x].end - pd[x].start + 1) / 4; + + for (i = 0; i < filter_len; ++i) { + v4sf pixels = __builtin_ia32_loadups(&sptr_xmm[i * 4]); + v4sf coeffs = __builtin_ia32_loadups(&coeffptr[i * 4]); + acc = __builtin_ia32_addps(acc, __builtin_ia32_mulps(pixels, coeffs)); + } + acc = __builtin_ia32_haddps(acc, acc); + acc = __builtin_ia32_haddps(acc, acc); + acc = __builtin_ia32_maxss(acc, low); + acc = __builtin_ia32_minss(acc, high); + result = __builtin_ia32_cvtss2si(acc); *dptr++ = (unsigned char)result; #else @@ -298,9 +345,12 @@ static void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsi *dptr++ = ch; } } + + free(pd); + free(coeffs); } -static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh, unsigned dstride) +static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsigned nh, unsigned dstride, enum qscale_scaling_filter scaling_filter) { struct pix_desc *pd = (struct pix_desc *)malloc(nh * sizeof(struct pix_desc)); int size_coeffs = 8; @@ -308,7 +358,13 @@ static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsi int num_coeffs = 0; int x, y, sy; double sf = (double)h / (double)nh; - double support = (h > nh) ? (3.0 * sf) : (3.0 / sf); + double support; + + if (scaling_filter == LANCZOS) { + support = (h > nh) ? (3.0 * sf) : (3.0 / sf); + } else { /* Mitchell */ + support = (h > nh) ? (2.0 * sf) : (2.0 / sf); + } /* calculate the filter */ for (y = 0; y < nh; ++y) { @@ -329,7 +385,12 @@ static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsi for (sy = start; sy <= end; ++sy) { double nd = (h > nh) ? (sy/sf - y) : (sy - y*sf); - double f = lanczos_tap(nd); + double f; + if (scaling_filter == LANCZOS) { + f = lanczos_tap(nd); + } else { /* Mitchell */ + f = mitchell_tap(nd); + } if (num_coeffs == size_coeffs) { size_coeffs <<= 1; coeffs = (float *)realloc(coeffs, size_coeffs * sizeof(float)); @@ -350,93 +411,51 @@ static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsi float *dptr = npix + x; for (y = 0; y < nh; ++y) { #if USE_VERTICAL_SSE - /* - * xmm0 - xmm3: acc[0..15] - * xmm4: current filter coefficient - * xmm5, xmm6, xmm7: scratchpad - */ - __asm__ ( - /* clear */ - "pxor %%xmm0, %%xmm0 \n" - "pxor %%xmm1, %%xmm1 \n" - "pxor %%xmm2, %%xmm2 \n" - "pxor %%xmm3, %%xmm3 \n" - - /* main loop */ - "0: \n" - - /* a zero is useful during unpacking */ - "pxor %%xmm4, %%xmm4 \n" - - /* fetch all 16 source bytes */ - "movups (%0), %%xmm5 \n" - "prefetcht0 (%0,%3,4) \n" - - /* unpack into words (xmm5, xmm7) */ - "movaps %%xmm5, %%xmm7 \n" - "punpcklbw %%xmm4, %%xmm5 \n" - "punpckhbw %%xmm4, %%xmm7 \n" - - /* unpack xmm5 into dwords (xmm5, xmm6) */ - "movaps %%xmm5, %%xmm6 \n" - "punpcklwd %%xmm4, %%xmm5 \n" - "punpckhwd %%xmm4, %%xmm6 \n" - - /* convert xmm5, xmm6 to floats */ - "cvtdq2ps %%xmm5, %%xmm5 \n" - "cvtdq2ps %%xmm6, %%xmm6 \n" - - /* fetch the coefficient */ - "movss (%2), %%xmm4 \n" - "shufps $0x0, %%xmm4, %%xmm4 \n" - - /* do the muls for xmm5 and xmm6 */ - "mulps %%xmm4, %%xmm5 \n" - "mulps %%xmm4, %%xmm6 \n" - "addps %%xmm5, %%xmm0 \n" - "addps %%xmm6, %%xmm1 \n" - - /* get the zero back again */ - "pxor %%xmm4, %%xmm4 \n" - - /* unpack xmm7 into dwords (xmm7, xmm6) */ - "movaps %%xmm7, %%xmm6 \n" - "punpcklwd %%xmm4, %%xmm7 \n" - "punpckhwd %%xmm4, %%xmm6 \n" - - /* convert xmm7, xmm6 to floats */ - "cvtdq2ps %%xmm7, %%xmm7 \n" - "cvtdq2ps %%xmm6, %%xmm6 \n" - - /* fetch the coefficient */ - "movss (%2), %%xmm4 \n" - "shufps $0x0, %%xmm4, %%xmm4 \n" - - /* do the second set of muls */ - "mulps %%xmm4, %%xmm7 \n" - "mulps %%xmm4, %%xmm6 \n" - "addps %%xmm7, %%xmm2 \n" - "addps %%xmm6, %%xmm3 \n" - - /* move along, and loop */ - "add $4, %2 \n" - "add %3, %0 \n" - "dec %1 \n" - "jnz 0b \n" - - /* store the values */ - "movaps %%xmm0, (%4) \n" - "movaps %%xmm1, 16(%4) \n" - "movaps %%xmm2, 32(%4) \n" - "movaps %%xmm3, 48(%4) \n" - : : - "r" (&sptr[pd[y].start * w]), /* 0: srcptr base */ - "r" (pd[y].end - pd[y].start + 1), /* 1: filter len */ - "r" (&coeffs[pd[y].startcoeff]), /* 2: coeffs base */ - "r" ((long)w), /* 3: stride */ - "r" (dptr) /* 4: dstptr base */ - : "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" - ); + /* A zero is useful during unpacking. */ + static const v4sf zero = { 0.0f, 0.0f, 0.0f, 0.0f }; + const unsigned char *sptr_xmm = &sptr[pd[y].start * w]; + const float *coeffptr = &coeffs[pd[y].startcoeff]; + const int filter_len = pd[y].end - pd[y].start + 1; + int i; + + v4sf acc0 = { 0.0f, 0.0f, 0.0f, 0.0f }; + v4sf acc1 = { 0.0f, 0.0f, 0.0f, 0.0f }; + v4sf acc2 = { 0.0f, 0.0f, 0.0f, 0.0f }; + v4sf acc3 = { 0.0f, 0.0f, 0.0f, 0.0f }; + + for (i = 0; i < filter_len; ++i, ++coeffptr, sptr_xmm += w) { + __builtin_prefetch(sptr_xmm + w, 0); + v16qi src = (v16qi)__builtin_ia32_loadups((float*)sptr_xmm); + + // unpack into words + v8hi src_lo = (v8hi)__builtin_ia32_punpcklbw128(src, (v16qi)zero); + v8hi src_hi = (v8hi)__builtin_ia32_punpckhbw128(src, (v16qi)zero); + + // unpack into dwords, convert to floats + v4si src0_i = (v4si)__builtin_ia32_punpcklwd128(src_lo, (v8hi)zero); + v4si src1_i = (v4si)__builtin_ia32_punpckhwd128(src_lo, (v8hi)zero); + v4si src2_i = (v4si)__builtin_ia32_punpcklwd128(src_hi, (v8hi)zero); + v4si src3_i = (v4si)__builtin_ia32_punpckhwd128(src_hi, (v8hi)zero); + + v4sf src0 = __builtin_ia32_cvtdq2ps(src0_i); + v4sf src1 = __builtin_ia32_cvtdq2ps(src1_i); + v4sf src2 = __builtin_ia32_cvtdq2ps(src2_i); + v4sf src3 = __builtin_ia32_cvtdq2ps(src3_i); + + // fetch the coefficient, and replicate it + v4sf coeff = { *coeffptr, *coeffptr, *coeffptr, *coeffptr }; + + // do the actual muladds + acc0 = __builtin_ia32_addps(acc0, __builtin_ia32_mulps(src0, coeff)); + acc1 = __builtin_ia32_addps(acc1, __builtin_ia32_mulps(src1, coeff)); + acc2 = __builtin_ia32_addps(acc2, __builtin_ia32_mulps(src2, coeff)); + acc3 = __builtin_ia32_addps(acc3, __builtin_ia32_mulps(src3, coeff)); + } + + *(v4sf *)(&dptr[0]) = acc0; + *(v4sf *)(&dptr[4]) = acc1; + *(v4sf *)(&dptr[8]) = acc2; + *(v4sf *)(&dptr[12]) = acc3; #else int i; float acc[CACHE_LINE_FACTOR]; @@ -478,6 +497,9 @@ static void vscale(unsigned char *pix, float *npix, unsigned w, unsigned h, unsi dptr += dstride; } } + + free(pd); + free(coeffs); } qscale_img *qscale_clone(const qscale_img *img) @@ -500,11 +522,11 @@ qscale_img *qscale_clone(const qscale_img *img) } { dst->data_cb = (unsigned char *)malloc(dst->h1 * dstride1); - memcpy(dst->data_cb, img->data_cb, dst->h0 * dstride0); + memcpy(dst->data_cb, img->data_cb, dst->h1 * dstride1); } { - dst->data_cr = (unsigned char *)malloc(dst->h1 * dstride1); - memcpy(dst->data_cr, img->data_cr, dst->h0 * dstride0); + dst->data_cr = (unsigned char *)malloc(dst->h2 * dstride2); + memcpy(dst->data_cr, img->data_cr, dst->h2 * dstride2); } return dst; @@ -519,37 +541,46 @@ qscale_img *qscale_scale(qscale_img *src, unsigned width, unsigned height, unsig dst->width = width; dst->height = height; + dst->num_components = src->num_components; 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; + if (src->num_components == 3) { + 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; + if (src->num_components == 3) { + if (samp_v1 > max_samp_v) + max_samp_v = samp_v1; + if (samp_v2 > max_samp_v) + max_samp_v = samp_v2; + } dst->w0 = width * samp_h0 / max_samp_h; dst->h0 = height * samp_v0 / max_samp_v; - dst->w1 = width * samp_h1 / max_samp_h; - dst->h1 = height * samp_v1 / max_samp_v; + if (src->num_components == 3) { + dst->w1 = width * samp_h1 / max_samp_h; + dst->h1 = height * samp_v1 / max_samp_v; - dst->w2 = width * samp_h2 / max_samp_h; - dst->h2 = height * samp_v2 / max_samp_v; + dst->w2 = width * samp_h2 / max_samp_h; + dst->h2 = height * samp_v2 / max_samp_v; + } dst->samp_h0 = samp_h0; dst->samp_v0 = samp_v0; - dst->samp_h1 = samp_h1; - dst->samp_v1 = samp_v1; + if (src->num_components == 3) { + dst->samp_h1 = samp_h1; + dst->samp_v1 = samp_v1; - dst->samp_h2 = samp_h2; - dst->samp_v2 = samp_v2; + dst->samp_h2 = samp_h2; + dst->samp_v2 = samp_v2; + } unsigned dstride0 = (dst->w0 + DCTSIZE-1) & ~(DCTSIZE-1); unsigned dstride1 = (dst->w1 + DCTSIZE-1) & ~(DCTSIZE-1); @@ -562,24 +593,26 @@ qscale_img *qscale_scale(qscale_img *src, unsigned width, unsigned height, unsig /* FIXME: handle out-of-memory gracefully */ { float *npix = (float*)memalign(16, sstride0 * dst->h0 * sizeof(float)); - vscale(src->data_y, npix, sstride0, src->h0, dst->h0, sstride0); + vscale(src->data_y, npix, sstride0, src->h0, dst->h0, sstride0, scaling_filter); dst->data_y = (unsigned char *)malloc(dst->h0 * dstride0); - hscale(npix, dst->data_y, src->w0, dst->h0, dst->w0, sstride0, dstride0); + hscale(npix, dst->data_y, src->w0, dst->h0, dst->w0, sstride0, dstride0, scaling_filter); free(npix); } - { - float *npix = (float*)memalign(16, sstride1 * dst->h1 * sizeof(float)); - vscale(src->data_cr, npix, sstride1, src->h1, dst->h1, sstride1); - dst->data_cr = (unsigned char *)malloc(dst->h1 * dstride1); - hscale(npix, dst->data_cr, src->w1, dst->h1, dst->w1, sstride1, dstride1); - free(npix); - } - { - float *npix = (float*)memalign(16, sstride2 * dst->h2 * sizeof(float)); - vscale(src->data_cb, npix, sstride2, src->h2, dst->h2, sstride2); - dst->data_cb = (unsigned char *)malloc(dst->h2 * dstride2); - hscale(npix, dst->data_cb, src->w2, dst->h2, dst->w2, sstride2, dstride2); - free(npix); + if (src->num_components == 3) { + { + float *npix = (float*)memalign(16, sstride1 * dst->h1 * sizeof(float)); + vscale(src->data_cr, npix, sstride1, src->h1, dst->h1, sstride1, scaling_filter); + dst->data_cr = (unsigned char *)malloc(dst->h1 * dstride1); + hscale(npix, dst->data_cr, src->w1, dst->h1, dst->w1, sstride1, dstride1, scaling_filter); + free(npix); + } + { + float *npix = (float*)memalign(16, sstride2 * dst->h2 * sizeof(float)); + vscale(src->data_cb, npix, sstride2, src->h2, dst->h2, sstride2, scaling_filter); + dst->data_cb = (unsigned char *)malloc(dst->h2 * dstride2); + hscale(npix, dst->data_cb, src->w2, dst->h2, dst->w2, sstride2, dstride2, scaling_filter); + free(npix); + } } return dst; @@ -605,19 +638,30 @@ int qscale_save_jpeg_to_stdio(const qscale_img *img, FILE *file, unsigned jpeg_q cinfo.err = jpeg_std_error(&jerr); jpeg_create_compress(&cinfo); jpeg_stdio_dest(&cinfo, file); - cinfo.input_components = 3; + cinfo.input_components = img->num_components; jpeg_set_defaults(&cinfo); jpeg_set_quality(&cinfo, jpeg_quality, FALSE); + + if (jpeg_mode == PROGRESSIVE) { + jpeg_simple_progression(&cinfo); + } + cinfo.image_width = img->width; cinfo.image_height = img->height; cinfo.raw_data_in = TRUE; - jpeg_set_colorspace(&cinfo, JCS_YCbCr); + if (img->num_components == 3) { + jpeg_set_colorspace(&cinfo, JCS_YCbCr); + } else { + jpeg_set_colorspace(&cinfo, JCS_GRAYSCALE); + } cinfo.comp_info[0].h_samp_factor = img->samp_h0; cinfo.comp_info[0].v_samp_factor = img->samp_v0; - cinfo.comp_info[1].h_samp_factor = img->samp_h1; - cinfo.comp_info[1].v_samp_factor = img->samp_v1; - cinfo.comp_info[2].h_samp_factor = img->samp_h2; - cinfo.comp_info[2].v_samp_factor = img->samp_v2; + if (img->num_components == 3) { + cinfo.comp_info[1].h_samp_factor = img->samp_h1; + cinfo.comp_info[1].v_samp_factor = img->samp_v1; + cinfo.comp_info[2].h_samp_factor = img->samp_h2; + cinfo.comp_info[2].v_samp_factor = img->samp_v2; + } jpeg_start_compress(&cinfo, TRUE); unsigned dstride0 = (img->w0 + DCTSIZE-1) & ~(DCTSIZE-1); @@ -641,17 +685,20 @@ int qscale_save_jpeg_to_stdio(const qscale_img *img, FILE *file, unsigned jpeg_q if (yline > img->h0 - 1) yline = img->h0 - 1; - int cbline = i + blocks*DCTSIZE*cinfo.comp_info[1].v_samp_factor; - if (cbline > img->h1 - 1) - cbline = img->h1 - 1; + y_row_ptrs[i] = img->data_y + yline * dstride0; + + if (img->num_components == 3) { + int cbline = i + blocks*DCTSIZE*cinfo.comp_info[1].v_samp_factor; + if (cbline > img->h1 - 1) + cbline = img->h1 - 1; - int crline = i + blocks*DCTSIZE*cinfo.comp_info[2].v_samp_factor; - if (crline > img->h2 - 1) - crline = img->h2 - 1; + int crline = i + blocks*DCTSIZE*cinfo.comp_info[2].v_samp_factor; + if (crline > img->h2 - 1) + crline = img->h2 - 1; - y_row_ptrs[i] = img->data_y + yline * dstride0; - cb_row_ptrs[i] = img->data_cb + cbline * dstride1; - cr_row_ptrs[i] = img->data_cr + crline * dstride2; + cb_row_ptrs[i] = img->data_cb + cbline * dstride1; + cr_row_ptrs[i] = img->data_cr + crline * dstride2; + } } total_lines += max_lines;