From c33cfc680827b26642b80a5f4543a85c45e377ab Mon Sep 17 00:00:00 2001 From: arpi Date: Sun, 2 Feb 2003 02:09:23 +0000 Subject: Based on the discussion in the other thread I made a new patch to vf_eq2.c that essentially combines vf_eq.c and vf_eq2.c. patch by Hampa Hug based on R/G/B gamma patch by Daniel Moreno some modifications (credits, buffer allocation, etc) by me git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@9214 b3059339-0415-0410-9bf9-f77b7e298cf2 --- libmpcodecs/vf_eq2.c | 352 +++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 286 insertions(+), 66 deletions(-) (limited to 'libmpcodecs/vf_eq2.c') diff --git a/libmpcodecs/vf_eq2.c b/libmpcodecs/vf_eq2.c index 57d13023f0..919b2ad8c9 100644 --- a/libmpcodecs/vf_eq2.c +++ b/libmpcodecs/vf_eq2.c @@ -1,10 +1,11 @@ /* * vf_eq2.c * - * LUT-based software equalizer (brightness, contrast, gamma) - * - * Hampa Hug + * Software equalizer (brightness, contrast, gamma, saturation) * + * Hampa Hug (original LUT gamma/contrast/brightness filter) + * Daniel Moreno (saturation, R/G/B gamma support) + * Richard Felker (original MMX contrast/brightness code (vf_eq.c)) */ #include @@ -12,8 +13,9 @@ #include #include -#include "../config.h" -#include "../mp_msg.h" +#include "config.h" +#include "mp_msg.h" +#include "cpudetect.h" #include "img_format.h" #include "mp_image.h" @@ -24,71 +26,159 @@ #endif +/* Per channel parameters */ +typedef struct eq2_param_t { + unsigned char lut[256]; + int lut_clean; + + void (*adjust) (struct eq2_param_t *par, unsigned char *dst, unsigned char *src, + unsigned w, unsigned h, unsigned dstride, unsigned sstride); + + double c; + double b; + double g; +} eq2_param_t; + typedef struct vf_priv_s { - unsigned char *buf; - int buf_w; - int buf_h; + eq2_param_t param[3]; double contrast; - double bright; + double brightness; + double saturation; + double gamma; + double rgamma; + double ggamma; + double bgamma; - unsigned char lut[256]; + unsigned buf_w[3]; + unsigned buf_h[3]; + unsigned char *buf[3]; } vf_eq2_t; static -void create_lut (vf_eq2_t *eq2) +void create_lut (eq2_param_t *par) { unsigned i; - double c, b, g; - double v; + double g, v; - c = eq2->contrast; - b = eq2->bright; - g = eq2->gamma; + g = par->g; if ((g < 0.001) || (g > 1000.0)) { g = 1.0; } - fprintf (stderr, "vf_eq2: c=%.2f b=%.2f g=%.4f\n", c, b, g); - g = 1.0 / g; for (i = 0; i < 256; i++) { v = (double) i / 255.0; - v = c * (v - 0.5) + 0.5 + b; + v = par->c * (v - 0.5) + 0.5 + par->b; if (v <= 0.0) { - eq2->lut[i] = 0; + par->lut[i] = 0; } else { v = pow (v, g); if (v >= 1.0) { - eq2->lut[i] = 255; + par->lut[i] = 255; } else { - /* we divided by 255.0 so now we also multiply by 255.0, not - by 256.0. "+ 0.5" ensures proper rounding */ - eq2->lut[i] = (unsigned char) (255.0 * v + 0.5); + par->lut[i] = (unsigned char) (256.0 * v); + } + } + } + + par->lut_clean = 1; +} + +#ifdef HAVE_MMX +static +void affine_1d_MMX (eq2_param_t *par, unsigned char *dst, unsigned char *src, + unsigned w, unsigned h, unsigned dstride, unsigned sstride) +{ + unsigned i; + int contrast, brightness; + unsigned dstep, sstep; + int pel; + short brvec[4]; + short contvec[4]; + +// printf("\nmmx: src=%p dst=%p w=%d h=%d ds=%d ss=%d\n",src,dst,w,h,dstride,sstride); + + contrast = (int) (par->c * 256 * 16); + brightness = ((int) (100.0 * par->b + 100.0) * 511) / 200 - 128 - contrast / 32; + + brvec[0] = brvec[1] = brvec[2] = brvec[3] = brightness; + contvec[0] = contvec[1] = contvec[2] = contvec[3] = contrast; + + sstep = sstride - w; + dstep = dstride - w; + + while (h-- > 0) { + asm volatile ( + "movq (%5), %%mm3 \n\t" + "movq (%6), %%mm4 \n\t" + "pxor %%mm0, %%mm0 \n\t" + "movl %4, %%eax\n\t" + ".balign 16 \n\t" + "1: \n\t" + "movq (%0), %%mm1 \n\t" + "movq (%0), %%mm2 \n\t" + "punpcklbw %%mm0, %%mm1 \n\t" + "punpckhbw %%mm0, %%mm2 \n\t" + "psllw $4, %%mm1 \n\t" + "psllw $4, %%mm2 \n\t" + "pmulhw %%mm4, %%mm1 \n\t" + "pmulhw %%mm4, %%mm2 \n\t" + "paddw %%mm3, %%mm1 \n\t" + "paddw %%mm3, %%mm2 \n\t" + "packuswb %%mm2, %%mm1 \n\t" + "addl $8, %0 \n\t" + "movq %%mm1, (%1) \n\t" + "addl $8, %1 \n\t" + "decl %%eax \n\t" + "jnz 1b \n\t" + : "=r" (src), "=r" (dst) + : "0" (src), "1" (dst), "r" (w >> 3), "r" (brvec), "r" (contvec) + : "%eax" + ); + + for (i = w & 7; i > 0; i--) { + pel = ((*src++ * contrast) >> 12) + brightness; + if (pel & 768) { + pel = (-pel) >> 31; } + *dst++ = pel; } + + src += sstep; + dst += dstep; } + + asm volatile ( "emms \n\t" ::: "memory" ); } +#endif -/* could inline this */ static -void process (unsigned char *dst, int dstride, unsigned char *src, int sstride, - int w, int h, unsigned char lut[256]) +void apply_lut (eq2_param_t *par, unsigned char *dst, unsigned char *src, + unsigned w, unsigned h, unsigned dstride, unsigned sstride) { - int i, j; + unsigned i, j; + unsigned char *lut; + + if (!par->lut_clean) { + create_lut (par); + } + + lut = par->lut; for (j = 0; j < h; j++) { for (i = 0; i < w; i++) { dst[i] = lut[src[i]]; } + src += sstride; dst += dstride; } @@ -97,34 +187,130 @@ void process (unsigned char *dst, int dstride, unsigned char *src, int sstride, static int put_image (vf_instance_t *vf, mp_image_t *src) { - mp_image_t *dst; - vf_eq2_t *eq2; + unsigned i; + vf_eq2_t *eq2; + mp_image_t *dst; + unsigned long img_n,img_c; eq2 = vf->priv; - if ((eq2->buf == NULL) || (eq2->buf_w != src->stride[0]) || (eq2->buf_h != src->h)) { - eq2->buf = (unsigned char *) realloc (eq2->buf, src->stride[0] * src->h); - eq2->buf_w = src->stride[0]; - eq2->buf_h = src->h; + if ((eq2->buf_w[0] != src->w) || (eq2->buf_h[0] != src->h)) { + eq2->buf_w[0] = src->w; + eq2->buf_h[0] = src->h; + eq2->buf_w[1] = eq2->buf_w[2] = src->w >> src->chroma_x_shift; + eq2->buf_h[1] = eq2->buf_h[2] = src->h >> src->chroma_y_shift; + img_n = eq2->buf_w[0]*eq2->buf_h[0]; + if(src->num_planes>1){ + img_c = eq2->buf_w[1]*eq2->buf_h[1]; + eq2->buf[0] = (unsigned char *) realloc (eq2->buf[0], img_n + 2*img_c); + eq2->buf[1] = eq2->buf[0] + img_n; + eq2->buf[2] = eq2->buf[1] + img_c; + } else + eq2->buf[0] = (unsigned char *) realloc (eq2->buf[0], img_n); } dst = vf_get_image (vf->next, src->imgfmt, MP_IMGTYPE_EXPORT, 0, src->w, src->h); - dst->stride[0] = src->stride[0]; - dst->stride[1] = src->stride[1]; - dst->stride[2] = src->stride[2]; - dst->planes[0] = vf->priv->buf; - dst->planes[1] = src->planes[1]; - dst->planes[2] = src->planes[2]; + for (i = 0; i < ((src->num_planes>1)?3:1); i++) { + if (eq2->param[i].adjust != NULL) { + dst->planes[i] = eq2->buf[i]; + dst->stride[i] = eq2->buf_w[i]; - process ( - dst->planes[0], dst->stride[0], src->planes[0], src->stride[0], - src->w, src->h, eq2->lut - ); + eq2->param[i].adjust (&eq2->param[i], dst->planes[i], src->planes[i], + eq2->buf_w[i], eq2->buf_h[i], dst->stride[i], src->stride[i]); + } + else { + dst->planes[i] = src->planes[i]; + dst->stride[i] = src->stride[i]; + } + } return vf_next_put_image (vf, dst); } +static +void check_values (eq2_param_t *par) +{ + /* yuck! floating point comparisons... */ + + if ((par->c == 1.0) && (par->b == 0.0) && (par->g == 1.0)) { + par->adjust = NULL; + } +#ifdef HAVE_MMX + else if (par->g == 1.0 && gCpuCaps.hasMMX) { + par->adjust = &affine_1d_MMX; + } +#endif + else { + par->adjust = &apply_lut; + } +} + +static +void print_values (vf_eq2_t *eq2) +{ + mp_msg (MSGT_VFILTER, MSGL_V, "vf_eq2: c=%.2f b=%.2f g=%.4f s=%.2f \n", + eq2->contrast, eq2->brightness, eq2->gamma, eq2->saturation + ); +} + +static +void set_contrast (vf_eq2_t *eq2, double c) +{ + eq2->contrast = c; + eq2->param[0].c = c; + eq2->param[0].lut_clean = 0; + check_values (&eq2->param[0]); + print_values (eq2); +} + +static +void set_brightness (vf_eq2_t *eq2, double b) +{ + eq2->brightness = b; + eq2->param[0].b = b; + eq2->param[0].lut_clean = 0; + check_values (&eq2->param[0]); + print_values (eq2); +} + +static +void set_gamma (vf_eq2_t *eq2, double g) +{ + eq2->gamma = g; + + eq2->param[0].g = eq2->gamma * eq2->ggamma; + eq2->param[1].g = sqrt (eq2->bgamma / eq2->ggamma); + eq2->param[2].g = sqrt (eq2->rgamma / eq2->ggamma); + + eq2->param[0].lut_clean = 0; + eq2->param[1].lut_clean = 0; + eq2->param[2].lut_clean = 0; + + check_values (&eq2->param[0]); + check_values (&eq2->param[1]); + check_values (&eq2->param[2]); + + print_values (eq2); +} + +static +void set_saturation (vf_eq2_t *eq2, double s) +{ + eq2->saturation = s; + + eq2->param[1].c = s; + eq2->param[2].c = s; + + eq2->param[1].lut_clean = 0; + eq2->param[2].lut_clean = 0; + + check_values (&eq2->param[1]); + check_values (&eq2->param[2]); + + print_values (eq2); +} + static int control (vf_instance_t *vf, int request, void *data) { @@ -135,18 +321,19 @@ int control (vf_instance_t *vf, int request, void *data) eq = (vf_equalizer_t *) data; if (strcmp (eq->item, "gamma") == 0) { - vf->priv->gamma = exp (log (8.0) * eq->value / 100.0); - create_lut (vf->priv); + set_gamma (vf->priv, exp (log (8.0) * eq->value / 100.0)); return CONTROL_TRUE; } else if (strcmp (eq->item, "contrast") == 0) { - vf->priv->contrast = (1.0 / 100.0) * (eq->value + 100); - create_lut (vf->priv); + set_contrast (vf->priv, (1.0 / 100.0) * (eq->value + 100)); return CONTROL_TRUE; } else if (strcmp (eq->item, "brightness") == 0) { - vf->priv->bright = (1.0 / 100.0) * eq->value; - create_lut (vf->priv); + set_brightness (vf->priv, (1.0 / 100.0) * eq->value); + return CONTROL_TRUE; + } + else if (strcmp (eq->item, "saturation") == 0) { + set_saturation (vf->priv, (double) (eq->value + 100) / 100.0); return CONTROL_TRUE; } break; @@ -162,7 +349,11 @@ int control (vf_instance_t *vf, int request, void *data) return CONTROL_TRUE; } else if (strcmp (eq->item, "brightness") == 0) { - eq->value = (int) (100.0 * vf->priv->bright); + eq->value = (int) (100.0 * vf->priv->brightness); + return CONTROL_TRUE; + } + else if (strcmp (eq->item, "saturation") == 0) { + eq->value = (int) (100.0 * vf->priv->saturation) - 100; return CONTROL_TRUE; } break; @@ -180,10 +371,8 @@ int query_format (vf_instance_t *vf, unsigned fmt) case IMGFMT_YV12: case IMGFMT_I420: case IMGFMT_IYUV: - case IMGFMT_CLPL: case IMGFMT_Y800: case IMGFMT_Y8: - case IMGFMT_NV12: case IMGFMT_444P: case IMGFMT_422P: case IMGFMT_411P: @@ -197,7 +386,7 @@ static void uninit (vf_instance_t *vf) { if (vf->priv != NULL) { - free (vf->priv->buf); + free (vf->priv->buf[0]); free (vf->priv); } } @@ -205,7 +394,9 @@ void uninit (vf_instance_t *vf) static int open (vf_instance_t *vf, char *args) { + unsigned i; vf_eq2_t *eq2; + double par[7]; vf->control = control; vf->query_format = query_format; @@ -215,33 +406,62 @@ int open (vf_instance_t *vf, char *args) vf->priv = (vf_eq2_t *) malloc (sizeof (vf_eq2_t)); eq2 = vf->priv; - eq2->buf = NULL; - eq2->buf_w = 0; - eq2->buf_h = 0; + for (i = 0; i < 3; i++) { + eq2->buf[i] = NULL; + eq2->buf_w[i] = 0; + eq2->buf_h[i] = 0; + + eq2->param[i].adjust = NULL; + eq2->param[i].c = 1.0; + eq2->param[i].b = 0.0; + eq2->param[i].g = 1.0; + eq2->param[i].lut_clean = 0; + } - eq2->gamma = 1.0; eq2->contrast = 1.0; - eq2->bright = 0.0; + eq2->brightness = 0.0; + eq2->saturation = 1.0; + + eq2->gamma = 1.0; + eq2->rgamma = 1.0; + eq2->ggamma = 1.0; + eq2->bgamma = 1.0; if (args != NULL) { + par[0] = 1.0; + par[1] = 1.0; + par[2] = 0.0; + par[3] = 1.0; + par[4] = 1.0; + par[5] = 1.0; + par[6] = 1.0; #ifdef USE_SETLOCALE - setlocale( LC_NUMERIC, "C" ); + setlocale (LC_NUMERIC, "C"); #endif - sscanf (args, "%lf:%lf:%lf", &eq2->gamma, &eq2->contrast, &eq2->bright); + sscanf (args, "%lf:%lf:%lf:%lf:%lf:%lf:%lf", + par, par + 1, par + 2, par + 3, par + 4, par + 5, par + 6 + ); #ifdef USE_SETLOCALE - setlocale( LC_NUMERIC, "" ); + setlocale (LC_NUMERIC, ""); #endif - } - create_lut (eq2); + eq2->rgamma = par[4]; + eq2->ggamma = par[5]; + eq2->bgamma = par[6]; + + set_gamma (eq2, par[0]); + set_contrast (eq2, par[1]); + set_brightness (eq2, par[2]); + set_saturation (eq2, par[3]); + } return 1; } vf_info_t vf_info_eq2 = { - "LUT-based software equalizer", + "Software equalizer", "eq2", - "Hampa Hug", + "Hampa Hug, Daniel Moreno, Richard Felker", "", &open }; -- cgit v1.2.3