pixman-bits-image's wide helpers first obtains the 8-bits image, then converts it to float. This destroys all the precision that the wide path was offering.
Fix this by making get_pixel() take a pointer instead of returning a value. Floating point will fill in a argb_t, while the 8-bits path will fill a 32-bits ARGB value. This also requires writing a floating point bilinear interpolator. With this change pixman can use the full floating point precision internally in all paths. Changes since v1: - Make accum and reduce an argument to convolution functions, to remove duplication. Signed-off-by: Maarten Lankhorst <maarten.lankho...@linux.intel.com> --- pixman/pixman-bits-image.c | 414 +++++++++++++++++++++++++++---------- pixman/pixman-inlines.h | 25 +++ 2 files changed, 328 insertions(+), 111 deletions(-) diff --git a/pixman/pixman-bits-image.c b/pixman/pixman-bits-image.c index 9fb91ff5831d..564789e9c685 100644 --- a/pixman/pixman-bits-image.c +++ b/pixman/pixman-bits-image.c @@ -36,43 +36,45 @@ #include "pixman-combine32.h" #include "pixman-inlines.h" -static uint32_t * -_pixman_image_get_scanline_generic_float (pixman_iter_t * iter, - const uint32_t *mask) -{ - pixman_iter_get_scanline_t fetch_32 = iter->data; - uint32_t *buffer = iter->buffer; - - fetch_32 (iter, NULL); +/* Fetch functions */ - pixman_expand_to_float ((argb_t *)buffer, buffer, PIXMAN_a8r8g8b8, iter->width); +static force_inline void +fetch_pixel_no_alpha_32 (bits_image_t *image, + int x, int y, pixman_bool_t check_bounds, + void *out) +{ + uint32_t *ret = out; - return iter->buffer; + if (check_bounds && + (x < 0 || x >= image->width || y < 0 || y >= image->height)) + *ret = 0; + else + *ret = image->fetch_pixel_32 (image, x, y); } -/* Fetch functions */ - -static force_inline uint32_t -fetch_pixel_no_alpha (bits_image_t *image, - int x, int y, pixman_bool_t check_bounds) +static force_inline void +fetch_pixel_no_alpha_float (bits_image_t *image, + int x, int y, pixman_bool_t check_bounds, + void *out) { + argb_t *ret = out; + if (check_bounds && (x < 0 || x >= image->width || y < 0 || y >= image->height)) - { - return 0; - } - - return image->fetch_pixel_32 (image, x, y); + ret->a = ret->r = ret->g = ret->b = 0.f; + else + *ret = image->fetch_pixel_float (image, x, y); } -typedef uint32_t (* get_pixel_t) (bits_image_t *image, - int x, int y, pixman_bool_t check_bounds); +typedef void (* get_pixel_t) (bits_image_t *image, + int x, int y, pixman_bool_t check_bounds, void *out); -static force_inline uint32_t +static force_inline void bits_image_fetch_pixel_nearest (bits_image_t *image, pixman_fixed_t x, pixman_fixed_t y, - get_pixel_t get_pixel) + get_pixel_t get_pixel, + void *out) { int x0 = pixman_fixed_to_int (x - pixman_fixed_e); int y0 = pixman_fixed_to_int (y - pixman_fixed_e); @@ -82,19 +84,20 @@ bits_image_fetch_pixel_nearest (bits_image_t *image, repeat (image->common.repeat, &x0, image->width); repeat (image->common.repeat, &y0, image->height); - return get_pixel (image, x0, y0, FALSE); + get_pixel (image, x0, y0, FALSE, out); } else { - return get_pixel (image, x0, y0, TRUE); + get_pixel (image, x0, y0, TRUE, out); } } -static force_inline uint32_t -bits_image_fetch_pixel_bilinear (bits_image_t *image, - pixman_fixed_t x, - pixman_fixed_t y, - get_pixel_t get_pixel) +static force_inline void +bits_image_fetch_pixel_bilinear_32 (bits_image_t *image, + pixman_fixed_t x, + pixman_fixed_t y, + get_pixel_t get_pixel, + void *out) { pixman_repeat_t repeat_mode = image->common.repeat; int width = image->width; @@ -102,6 +105,7 @@ bits_image_fetch_pixel_bilinear (bits_image_t *image, int x1, y1, x2, y2; uint32_t tl, tr, bl, br; int32_t distx, disty; + uint32_t *ret = out; x1 = x - pixman_fixed_1 / 2; y1 = y - pixman_fixed_1 / 2; @@ -121,27 +125,140 @@ bits_image_fetch_pixel_bilinear (bits_image_t *image, repeat (repeat_mode, &x2, width); repeat (repeat_mode, &y2, height); - tl = get_pixel (image, x1, y1, FALSE); - bl = get_pixel (image, x1, y2, FALSE); - tr = get_pixel (image, x2, y1, FALSE); - br = get_pixel (image, x2, y2, FALSE); + get_pixel (image, x1, y1, FALSE, &tl); + get_pixel (image, x2, y1, FALSE, &tr); + get_pixel (image, x1, y2, FALSE, &bl); + get_pixel (image, x2, y2, FALSE, &br); } else { - tl = get_pixel (image, x1, y1, TRUE); - tr = get_pixel (image, x2, y1, TRUE); - bl = get_pixel (image, x1, y2, TRUE); - br = get_pixel (image, x2, y2, TRUE); + get_pixel (image, x1, y1, TRUE, &tl); + get_pixel (image, x2, y1, TRUE, &tr); + get_pixel (image, x1, y2, TRUE, &bl); + get_pixel (image, x2, y2, TRUE, &br); } - return bilinear_interpolation (tl, tr, bl, br, distx, disty); + *ret = bilinear_interpolation (tl, tr, bl, br, distx, disty); } -static force_inline uint32_t +static force_inline void +bits_image_fetch_pixel_bilinear_float (bits_image_t *image, + pixman_fixed_t x, + pixman_fixed_t y, + get_pixel_t get_pixel, + void *out) +{ + pixman_repeat_t repeat_mode = image->common.repeat; + int width = image->width; + int height = image->height; + int x1, y1, x2, y2; + argb_t tl, tr, bl, br; + float distx, disty; + argb_t *ret = out; + + x1 = x - pixman_fixed_1 / 2; + y1 = y - pixman_fixed_1 / 2; + + distx = ((float)pixman_fixed_fraction(x1)) / 65536.f; + disty = ((float)pixman_fixed_fraction(y1)) / 65536.f; + + x1 = pixman_fixed_to_int (x1); + y1 = pixman_fixed_to_int (y1); + x2 = x1 + 1; + y2 = y1 + 1; + + if (repeat_mode != PIXMAN_REPEAT_NONE) + { + repeat (repeat_mode, &x1, width); + repeat (repeat_mode, &y1, height); + repeat (repeat_mode, &x2, width); + repeat (repeat_mode, &y2, height); + + get_pixel (image, x1, y1, FALSE, &tl); + get_pixel (image, x2, y1, FALSE, &tr); + get_pixel (image, x1, y2, FALSE, &bl); + get_pixel (image, x2, y2, FALSE, &br); + } + else + { + get_pixel (image, x1, y1, TRUE, &tl); + get_pixel (image, x2, y1, TRUE, &tr); + get_pixel (image, x1, y2, TRUE, &bl); + get_pixel (image, x2, y2, TRUE, &br); + } + + *ret = bilinear_interpolation_float (tl, tr, bl, br, distx, disty); +} + +static force_inline void accum_32(int *satot, int *srtot, + int *sgtot, int *sbtot, + const void *p, pixman_fixed_t f) +{ + uint32_t pixel = *(uint32_t *)p; + + *srtot += (int)RED_8 (pixel) * f; + *sgtot += (int)GREEN_8 (pixel) * f; + *sbtot += (int)BLUE_8 (pixel) * f; + *satot += (int)ALPHA_8 (pixel) * f; +} + +static force_inline void reduce_32(int satot, int srtot, + int sgtot, int sbtot, void *p) +{ + uint32_t *ret = p; + + satot = (satot + 0x8000) >> 16; + srtot = (srtot + 0x8000) >> 16; + sgtot = (sgtot + 0x8000) >> 16; + sbtot = (sbtot + 0x8000) >> 16; + + satot = CLIP (satot, 0, 0xff); + srtot = CLIP (srtot, 0, 0xff); + sgtot = CLIP (sgtot, 0, 0xff); + sbtot = CLIP (sbtot, 0, 0xff); + + *ret = ((satot << 24) | (srtot << 16) | (sgtot << 8) | (sbtot)); +} + +static force_inline void accum_float(int *satot, int *srtot, + int *sgtot, int *sbtot, + const void *p, pixman_fixed_t f) +{ + const argb_t *pixel = p; + + *satot += pixel->a * f; + *srtot += pixel->r * f; + *sgtot += pixel->g * f; + *sbtot += pixel->b * f; +} + +static force_inline void reduce_float(int satot, int srtot, + int sgtot, int sbtot, + void *p) +{ + argb_t *ret = p; + + ret->a = CLIP (satot / 65536.f, 0.f, 1.f); + ret->r = CLIP (srtot / 65536.f, 0.f, 1.f); + ret->g = CLIP (sgtot / 65536.f, 0.f, 1.f); + ret->b = CLIP (sbtot / 65536.f, 0.f, 1.f); +} + +typedef void (* accumulate_pixel_t) (int *satot, int *srtot, + int *sgtot, int *sbtot, + const void *pixel, pixman_fixed_t f); + +typedef void (* reduce_pixel_t) (int satot, int srtot, + int sgtot, int sbtot, void *out); + +static force_inline void bits_image_fetch_pixel_convolution (bits_image_t *image, pixman_fixed_t x, pixman_fixed_t y, - get_pixel_t get_pixel) + get_pixel_t get_pixel, + void *out, + accumulate_pixel_t accum, + reduce_pixel_t reduce) { pixman_fixed_t *params = image->common.filter_params; int x_off = (params[0] - pixman_fixed_1) >> 1; @@ -174,48 +291,39 @@ bits_image_fetch_pixel_convolution (bits_image_t *image, if (f) { - uint32_t pixel; + /* Must be big enough to hold a argb_t */ + argb_t pixel; if (repeat_mode != PIXMAN_REPEAT_NONE) { repeat (repeat_mode, &rx, width); repeat (repeat_mode, &ry, height); - pixel = get_pixel (image, rx, ry, FALSE); + get_pixel (image, rx, ry, FALSE, &pixel); } else { - pixel = get_pixel (image, rx, ry, TRUE); + get_pixel (image, rx, ry, TRUE, &pixel); } - srtot += (int)RED_8 (pixel) * f; - sgtot += (int)GREEN_8 (pixel) * f; - sbtot += (int)BLUE_8 (pixel) * f; - satot += (int)ALPHA_8 (pixel) * f; + accum (&satot, &srtot, &sgtot, &sbtot, &pixel, f); } params++; } } - satot = (satot + 0x8000) >> 16; - srtot = (srtot + 0x8000) >> 16; - sgtot = (sgtot + 0x8000) >> 16; - sbtot = (sbtot + 0x8000) >> 16; - - satot = CLIP (satot, 0, 0xff); - srtot = CLIP (srtot, 0, 0xff); - sgtot = CLIP (sgtot, 0, 0xff); - sbtot = CLIP (sbtot, 0, 0xff); - - return ((satot << 24) | (srtot << 16) | (sgtot << 8) | (sbtot)); + reduce (satot, srtot, sgtot, sbtot, out); } -static uint32_t -bits_image_fetch_pixel_separable_convolution (bits_image_t *image, - pixman_fixed_t x, - pixman_fixed_t y, - get_pixel_t get_pixel) +static void +bits_image_fetch_pixel_separable_convolution (bits_image_t *image, + pixman_fixed_t x, + pixman_fixed_t y, + get_pixel_t get_pixel, + void *out, + accumulate_pixel_t accum, + reduce_pixel_t reduce) { pixman_fixed_t *params = image->common.filter_params; pixman_repeat_t repeat_mode = image->common.repeat; @@ -270,82 +378,91 @@ bits_image_fetch_pixel_separable_convolution (bits_image_t *image, if (fx) { + /* Must be big enough to hold a argb_t */ + argb_t pixel; pixman_fixed_t f; - uint32_t pixel; if (repeat_mode != PIXMAN_REPEAT_NONE) { repeat (repeat_mode, &rx, width); repeat (repeat_mode, &ry, height); - pixel = get_pixel (image, rx, ry, FALSE); + get_pixel (image, rx, ry, FALSE, &pixel); } else { - pixel = get_pixel (image, rx, ry, TRUE); + get_pixel (image, rx, ry, TRUE, &pixel); } f = (fy * fx + 0x8000) >> 16; - srtot += (int)RED_8 (pixel) * f; - sgtot += (int)GREEN_8 (pixel) * f; - sbtot += (int)BLUE_8 (pixel) * f; - satot += (int)ALPHA_8 (pixel) * f; + accum(&satot, &srtot, &sgtot, &sbtot, &pixel, f); } } } } - satot = (satot + 0x8000) >> 16; - srtot = (srtot + 0x8000) >> 16; - sgtot = (sgtot + 0x8000) >> 16; - sbtot = (sbtot + 0x8000) >> 16; - satot = CLIP (satot, 0, 0xff); - srtot = CLIP (srtot, 0, 0xff); - sgtot = CLIP (sgtot, 0, 0xff); - sbtot = CLIP (sbtot, 0, 0xff); - - return ((satot << 24) | (srtot << 16) | (sgtot << 8) | (sbtot)); + reduce(satot, srtot, sgtot, sbtot, out); } -static force_inline uint32_t -bits_image_fetch_pixel_filtered (bits_image_t *image, +static force_inline void +bits_image_fetch_pixel_filtered (bits_image_t *image, + pixman_bool_t wide, pixman_fixed_t x, pixman_fixed_t y, - get_pixel_t get_pixel) + get_pixel_t get_pixel, + void *out) { switch (image->common.filter) { case PIXMAN_FILTER_NEAREST: case PIXMAN_FILTER_FAST: - return bits_image_fetch_pixel_nearest (image, x, y, get_pixel); + bits_image_fetch_pixel_nearest (image, x, y, get_pixel, out); break; case PIXMAN_FILTER_BILINEAR: case PIXMAN_FILTER_GOOD: case PIXMAN_FILTER_BEST: - return bits_image_fetch_pixel_bilinear (image, x, y, get_pixel); + if (wide) + bits_image_fetch_pixel_bilinear_float (image, x, y, get_pixel, out); + else + bits_image_fetch_pixel_bilinear_32 (image, x, y, get_pixel, out); break; case PIXMAN_FILTER_CONVOLUTION: - return bits_image_fetch_pixel_convolution (image, x, y, get_pixel); + if (wide) + bits_image_fetch_pixel_convolution (image, x, y, + get_pixel, out, + accum_float, + reduce_float); + else + bits_image_fetch_pixel_convolution (image, x, y, + get_pixel, out, + accum_32, reduce_32); break; case PIXMAN_FILTER_SEPARABLE_CONVOLUTION: - return bits_image_fetch_pixel_separable_convolution (image, x, y, get_pixel); + if (wide) + bits_image_fetch_pixel_separable_convolution (image, x, y, + get_pixel, out, + accum_float, + reduce_float); + else + bits_image_fetch_pixel_separable_convolution (image, x, y, + get_pixel, out, + accum_32, reduce_32); break; default: break; } - - return 0; } static uint32_t * -bits_image_fetch_affine_no_alpha (pixman_iter_t * iter, - const uint32_t * mask) +__bits_image_fetch_affine_no_alpha (pixman_iter_t * iter, + pixman_bool_t wide, + const uint32_t * mask) { pixman_image_t *image = iter->image; int offset = iter->x; @@ -357,6 +474,8 @@ bits_image_fetch_affine_no_alpha (pixman_iter_t * iter, pixman_fixed_t ux, uy; pixman_vector_t v; int i; + get_pixel_t get_pixel = + wide ? fetch_pixel_no_alpha_float : fetch_pixel_no_alpha_32; /* reference point is the center of the pixel */ v.vector[0] = pixman_int_to_fixed (offset) + pixman_fixed_1 / 2; @@ -384,27 +503,45 @@ bits_image_fetch_affine_no_alpha (pixman_iter_t * iter, { if (!mask || mask[i]) { - buffer[i] = bits_image_fetch_pixel_filtered ( - &image->bits, x, y, fetch_pixel_no_alpha); + bits_image_fetch_pixel_filtered ( + &image->bits, wide, x, y, get_pixel, buffer); } x += ux; y += uy; + buffer += wide ? 4 : 1; } - return buffer; + return iter->buffer; +} + +static uint32_t * +bits_image_fetch_affine_no_alpha_32 (pixman_iter_t *iter, + const uint32_t *mask) +{ + return __bits_image_fetch_affine_no_alpha(iter, FALSE, mask); +} + +static uint32_t * +bits_image_fetch_affine_no_alpha_float (pixman_iter_t *iter, + const uint32_t *mask) +{ + return __bits_image_fetch_affine_no_alpha(iter, TRUE, mask); } /* General fetcher */ -static force_inline uint32_t -fetch_pixel_general (bits_image_t *image, int x, int y, pixman_bool_t check_bounds) +static force_inline void +fetch_pixel_general_32 (bits_image_t *image, + int x, int y, pixman_bool_t check_bounds, + void *out) { - uint32_t pixel; + uint32_t pixel, *ret = out; if (check_bounds && (x < 0 || x >= image->width || y < 0 || y >= image->height)) { - return 0; + *ret = 0; + return; } pixel = image->fetch_pixel_32 (image, x, y); @@ -433,18 +570,59 @@ fetch_pixel_general (bits_image_t *image, int x, int y, pixman_bool_t check_boun pixel |= (pixel_a << 24); } - return pixel; + *ret = pixel; +} + +static force_inline void +fetch_pixel_general_float (bits_image_t *image, + int x, int y, pixman_bool_t check_bounds, + void *out) +{ + argb_t *ret = out; + + if (check_bounds && + (x < 0 || x >= image->width || y < 0 || y >= image->height)) + { + ret->a = ret->r = ret->g = ret->b = 0; + return; + } + + *ret = image->fetch_pixel_float (image, x, y); + + if (image->common.alpha_map) + { + x -= image->common.alpha_origin_x; + y -= image->common.alpha_origin_y; + + if (x < 0 || x >= image->common.alpha_map->width || + y < 0 || y >= image->common.alpha_map->height) + { + ret->a = 0.f; + } + else + { + argb_t alpha; + + alpha = image->common.alpha_map->fetch_pixel_float ( + image->common.alpha_map, x, y); + + ret->a = alpha.a; + } + } } static uint32_t * -bits_image_fetch_general (pixman_iter_t *iter, - const uint32_t *mask) +__bits_image_fetch_general (pixman_iter_t *iter, + pixman_bool_t wide, + const uint32_t *mask) { pixman_image_t *image = iter->image; int offset = iter->x; int line = iter->y++; int width = iter->width; uint32_t * buffer = iter->buffer; + get_pixel_t get_pixel = + wide ? fetch_pixel_general_float : fetch_pixel_general_32; pixman_fixed_t x, y, w; pixman_fixed_t ux, uy, uw; @@ -493,16 +671,31 @@ bits_image_fetch_general (pixman_iter_t *iter, y0 = 0; } - buffer[i] = bits_image_fetch_pixel_filtered ( - &image->bits, x0, y0, fetch_pixel_general); + bits_image_fetch_pixel_filtered ( + &image->bits, wide, x0, y0, get_pixel, buffer); } x += ux; y += uy; w += uw; + buffer += wide ? 4 : 1; } - return buffer; + return iter->buffer; +} + +static uint32_t * +bits_image_fetch_general_32 (pixman_iter_t *iter, + const uint32_t *mask) +{ + return __bits_image_fetch_general(iter, FALSE, mask); +} + +static uint32_t * +bits_image_fetch_general_float (pixman_iter_t *iter, + const uint32_t *mask) +{ + return __bits_image_fetch_general(iter, TRUE, mask); } static void @@ -703,15 +896,15 @@ static const fetcher_info_t fetcher_info[] = /* Affine, no alpha */ { PIXMAN_any, (FAST_PATH_NO_ALPHA_MAP | FAST_PATH_HAS_TRANSFORM | FAST_PATH_AFFINE_TRANSFORM), - bits_image_fetch_affine_no_alpha, - _pixman_image_get_scanline_generic_float + bits_image_fetch_affine_no_alpha_32, + bits_image_fetch_affine_no_alpha_float, }, /* General */ { PIXMAN_any, 0, - bits_image_fetch_general, - _pixman_image_get_scanline_generic_float + bits_image_fetch_general_32, + bits_image_fetch_general_float, }, { PIXMAN_null }, @@ -741,7 +934,6 @@ _pixman_bits_image_src_iter_init (pixman_image_t *image, pixman_iter_t *iter) } else { - iter->data = info->get_scanline_32; iter->get_scanline = info->get_scanline_float; } return; diff --git a/pixman/pixman-inlines.h b/pixman/pixman-inlines.h index 1c8441d6dabe..332e208140a0 100644 --- a/pixman/pixman-inlines.h +++ b/pixman/pixman-inlines.h @@ -222,6 +222,31 @@ bilinear_interpolation (uint32_t tl, uint32_t tr, #endif #endif // BILINEAR_INTERPOLATION_BITS <= 4 +static force_inline argb_t +bilinear_interpolation_float (argb_t tl, argb_t tr, + argb_t bl, argb_t br, + float distx, float disty) +{ + float distxy, distxiy, distixy, distixiy; + argb_t r; + + distxy = distx * disty; + distxiy = distx - (1.f - distxy); + distixy = (1.f - distx) * disty; + distixiy = (1.f - distx) * (1.f - disty); + + r.a = tl.a * distixiy + tr.a * distxiy + + bl.a * distixy + br.a * distxy; + r.r = tl.r * distixiy + tr.r * distxiy + + bl.r * distixy + br.r * distxy; + r.g = tl.g * distixiy + tr.g * distxiy + + bl.g * distixy + br.g * distxy; + r.b = tl.b * distixiy + tr.b * distxiy + + bl.b * distixy + br.b * distxy; + + return r; +} + /* * For each scanline fetched from source image with PAD repeat: * - calculate how many pixels need to be padded on the left side -- 2.20.1 _______________________________________________ Pixman mailing list Pixman@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/pixman