bool
sna_transform_is_integer_translation(const PictTransform *t, int16_t *tx, int16_t *ty)
{
if (t == NULL) {
*tx = *ty = 0;
return true;
}
if (t->matrix[0][0] != IntToxFixed(1) ||
t->matrix[0][1] != 0 ||
t->matrix[1][0] != 0 ||
t->matrix[1][1] != IntToxFixed(1) ||
t->matrix[2][0] != 0 ||
t->matrix[2][1] != 0 ||
t->matrix[2][2] != IntToxFixed(1))
return false;
if (pixman_fixed_fraction(t->matrix[0][2]) ||
pixman_fixed_fraction(t->matrix[1][2]))
return false;
*tx = pixman_fixed_to_int(t->matrix[0][2]);
*ty = pixman_fixed_to_int(t->matrix[1][2]);
return true;
}
bool
sna_transform_is_imprecise_integer_translation(const PictTransform *t,
int filter, bool precise,
int16_t *tx, int16_t *ty)
{
if (t == NULL) {
DBG(("%s: no transform\n", __FUNCTION__));
*tx = *ty = 0;
return true;
}
DBG(("%s: FilterNearest?=%d, precise?=%d, transform=[%f %f %f, %f %f %f, %f %f %f]\n",
__FUNCTION__, filter==PictFilterNearest, precise,
t->matrix[0][0]/65536., t->matrix[0][1]/65536., t->matrix[0][2]/65536.,
t->matrix[1][0]/65536., t->matrix[1][1]/65536., t->matrix[1][2]/65536.,
t->matrix[2][0]/65536., t->matrix[2][1]/65536., t->matrix[2][2]/65536.));
if (t->matrix[0][0] != IntToxFixed(1) ||
t->matrix[0][1] != 0 ||
t->matrix[1][0] != 0 ||
t->matrix[1][1] != IntToxFixed(1) ||
t->matrix[2][0] != 0 ||
t->matrix[2][1] != 0 ||
t->matrix[2][2] != IntToxFixed(1)) {
DBG(("%s: not unity scaling\n", __FUNCTION__));
return false;
}
if (filter != PictFilterNearest) {
if (precise) {
if (pixman_fixed_fraction(t->matrix[0][2]) ||
pixman_fixed_fraction(t->matrix[1][2])) {
DBG(("%s: precise, fractional translation\n", __FUNCTION__));
return false;
}
} else {
int f;
f = pixman_fixed_fraction(t->matrix[0][2]);
if (f > IntToxFixed(1)/4 && f < IntToxFixed(3)/4) {
DBG(("%s: imprecise, fractional translation X: %x\n", __FUNCTION__, f));
return false;
}
f = pixman_fixed_fraction(t->matrix[1][2]);
if (f > IntToxFixed(1)/4 && f < IntToxFixed(3)/4) {
DBG(("%s: imprecise, fractional translation Y: %x\n", __FUNCTION__, f));
return false;
}
}
}
*tx = pixman_fixed_to_int(t->matrix[0][2] + IntToxFixed(1)/2);
*ty = pixman_fixed_to_int(t->matrix[1][2] + IntToxFixed(1)/2);
return true;
}
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)
{
pixman_repeat_t repeat_mode = image->common.repeat;
int width = image->width;
int height = image->height;
int x1, y1, x2, y2;
uint32_t tl, tr, bl, br;
int32_t distx, disty;
x1 = x - pixman_fixed_1 / 2;
y1 = y - pixman_fixed_1 / 2;
distx = pixman_fixed_to_bilinear_weight (x1);
disty = pixman_fixed_to_bilinear_weight (y1);
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);
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);
}
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);
}
return bilinear_interpolation (tl, tr, bl, br, distx, disty);
}
PIXMAN_EXPORT void
pixman_add_traps (pixman_image_t * image,
int16_t x_off,
int16_t y_off,
int ntrap,
pixman_trap_t * traps)
{
int bpp;
int width;
int height;
pixman_fixed_t x_off_fixed;
pixman_fixed_t y_off_fixed;
pixman_edge_t l, r;
pixman_fixed_t t, b;
_pixman_image_validate (image);
width = image->bits.width;
height = image->bits.height;
bpp = PIXMAN_FORMAT_BPP (image->bits.format);
x_off_fixed = pixman_int_to_fixed (x_off);
y_off_fixed = pixman_int_to_fixed (y_off);
while (ntrap--)
{
t = traps->top.y + y_off_fixed;
if (t < 0)
t = 0;
t = pixman_sample_ceil_y (t, bpp);
b = traps->bot.y + y_off_fixed;
if (pixman_fixed_to_int (b) >= height)
b = pixman_int_to_fixed (height) - 1;
b = pixman_sample_floor_y (b, bpp);
if (b >= t)
{
/* initialize edge walkers */
pixman_edge_init (&l, bpp, t,
traps->top.l + x_off_fixed,
traps->top.y + y_off_fixed,
traps->bot.l + x_off_fixed,
traps->bot.y + y_off_fixed);
pixman_edge_init (&r, bpp, t,
traps->top.r + x_off_fixed,
traps->top.y + y_off_fixed,
traps->bot.r + x_off_fixed,
traps->bot.y + y_off_fixed);
pixman_rasterize_edges (image, &l, &r, t, b);
}
traps++;
}
}
static force_inline uint32_t
bits_image_fetch_pixel_nearest (bits_image_t *image,
pixman_fixed_t x,
pixman_fixed_t y,
get_pixel_t get_pixel)
{
int x0 = pixman_fixed_to_int (x - pixman_fixed_e);
int y0 = pixman_fixed_to_int (y - pixman_fixed_e);
if (image->common.repeat != PIXMAN_REPEAT_NONE)
{
repeat (image->common.repeat, &x0, image->width);
repeat (image->common.repeat, &y0, image->height);
return get_pixel (image, x0, y0, FALSE);
}
else
{
return get_pixel (image, x0, y0, TRUE);
}
}
PIXMAN_EXPORT void
pixman_rasterize_trapezoid (pixman_image_t * image,
const pixman_trapezoid_t *trap,
int x_off,
int y_off)
{
int bpp;
int width;
int height;
pixman_fixed_t x_off_fixed;
pixman_fixed_t y_off_fixed;
pixman_edge_t l, r;
pixman_fixed_t t, b;
return_if_fail (image->type == BITS);
_pixman_image_validate (image);
if (!pixman_trapezoid_valid (trap))
return;
width = image->bits.width;
height = image->bits.height;
bpp = PIXMAN_FORMAT_BPP (image->bits.format);
x_off_fixed = pixman_int_to_fixed (x_off);
y_off_fixed = pixman_int_to_fixed (y_off);
t = trap->top + y_off_fixed;
if (t < 0)
t = 0;
t = pixman_sample_ceil_y (t, bpp);
b = trap->bottom + y_off_fixed;
if (pixman_fixed_to_int (b) >= height)
b = pixman_int_to_fixed (height) - 1;
b = pixman_sample_floor_y (b, bpp);
if (b >= t)
{
/* initialize edge walkers */
pixman_line_fixed_edge_init (&l, bpp, t, &trap->left, x_off, y_off);
pixman_line_fixed_edge_init (&r, bpp, t, &trap->right, x_off, y_off);
pixman_rasterize_edges (image, &l, &r, t, b);
}
}
PIXMAN_EXPORT pixman_fixed_t
pixman_sample_ceil_y (pixman_fixed_t y, int n)
{
pixman_fixed_t f = pixman_fixed_frac (y);
pixman_fixed_t i = pixman_fixed_floor (y);
f = DIV (f - Y_FRAC_FIRST (n) + (STEP_Y_SMALL (n) - pixman_fixed_e), STEP_Y_SMALL (n)) * STEP_Y_SMALL (n) +
Y_FRAC_FIRST (n);
if (f > Y_FRAC_LAST (n))
{
if (pixman_fixed_to_int (i) == 0x7fff)
{
f = 0xffff; /* saturate */
}
else
{
f = Y_FRAC_FIRST (n);
i += pixman_fixed_1;
}
}
return (i | f);
}
/*
* Compute the largest value strictly less than y which is on a
* grid row.
*/
PIXMAN_EXPORT pixman_fixed_t
pixman_sample_floor_y (pixman_fixed_t y,
int n)
{
pixman_fixed_t f = pixman_fixed_frac (y);
pixman_fixed_t i = pixman_fixed_floor (y);
f = DIV (f - pixman_fixed_e - Y_FRAC_FIRST (n), STEP_Y_SMALL (n)) * STEP_Y_SMALL (n) +
Y_FRAC_FIRST (n);
if (f < Y_FRAC_FIRST (n))
{
if (pixman_fixed_to_int (i) == 0x8000)
{
f = 0; /* saturate */
}
else
{
f = Y_FRAC_LAST (n);
i -= pixman_fixed_1;
}
}
return (i | f);
}
static force_inline void
bits_image_fetch_bilinear_affine (pixman_image_t * image,
int offset,
int line,
int width,
uint32_t * buffer,
const uint32_t * mask,
convert_pixel_t convert_pixel,
pixman_format_code_t format,
pixman_repeat_t repeat_mode)
{
pixman_fixed_t x, y;
pixman_fixed_t ux, uy;
pixman_vector_t v;
bits_image_t *bits = &image->bits;
int i;
/* reference point is the center of the pixel */
v.vector[0] = pixman_int_to_fixed (offset) + pixman_fixed_1 / 2;
v.vector[1] = pixman_int_to_fixed (line) + pixman_fixed_1 / 2;
v.vector[2] = pixman_fixed_1;
if (!pixman_transform_point_3d (image->common.transform, &v))
return;
ux = image->common.transform->matrix[0][0];
uy = image->common.transform->matrix[1][0];
x = v.vector[0];
y = v.vector[1];
for (i = 0; i < width; ++i)
{
int x1, y1, x2, y2;
uint32_t tl, tr, bl, br;
int32_t distx, disty;
int width = image->bits.width;
int height = image->bits.height;
const uint8_t *row1;
const uint8_t *row2;
if (mask && !mask[i])
goto next;
x1 = x - pixman_fixed_1 / 2;
y1 = y - pixman_fixed_1 / 2;
distx = pixman_fixed_to_bilinear_weight (x1);
disty = pixman_fixed_to_bilinear_weight (y1);
y1 = pixman_fixed_to_int (y1);
y2 = y1 + 1;
x1 = pixman_fixed_to_int (x1);
x2 = x1 + 1;
if (repeat_mode != PIXMAN_REPEAT_NONE)
{
uint32_t mask;
mask = PIXMAN_FORMAT_A (format)? 0 : 0xff000000;
repeat (repeat_mode, &x1, width);
repeat (repeat_mode, &y1, height);
repeat (repeat_mode, &x2, width);
repeat (repeat_mode, &y2, height);
row1 = (uint8_t *)bits->bits + bits->rowstride * 4 * y1;
row2 = (uint8_t *)bits->bits + bits->rowstride * 4 * y2;
tl = convert_pixel (row1, x1) | mask;
tr = convert_pixel (row1, x2) | mask;
bl = convert_pixel (row2, x1) | mask;
br = convert_pixel (row2, x2) | mask;
}
else
{
uint32_t mask1, mask2;
int bpp;
/* Note: PIXMAN_FORMAT_BPP() returns an unsigned value,
* which means if you use it in expressions, those
* expressions become unsigned themselves. Since
* the variables below can be negative in some cases,
* that will lead to crashes on 64 bit architectures.
*
* So this line makes sure bpp is signed
*/
bpp = PIXMAN_FORMAT_BPP (format);
if (x1 >= width || x2 < 0 || y1 >= height || y2 < 0)
{
buffer[i] = 0;
goto next;
}
if (y2 == 0)
{
row1 = zero;
mask1 = 0;
}
else
{
row1 = (uint8_t *)bits->bits + bits->rowstride * 4 * y1;
row1 += bpp / 8 * x1;
mask1 = PIXMAN_FORMAT_A (format)? 0 : 0xff000000;
}
if (y1 == height - 1)
{
row2 = zero;
mask2 = 0;
}
else
{
row2 = (uint8_t *)bits->bits + bits->rowstride * 4 * y2;
row2 += bpp / 8 * x1;
mask2 = PIXMAN_FORMAT_A (format)? 0 : 0xff000000;
}
if (x2 == 0)
{
tl = 0;
bl = 0;
}
else
{
tl = convert_pixel (row1, 0) | mask1;
bl = convert_pixel (row2, 0) | mask2;
}
if (x1 == width - 1)
{
tr = 0;
br = 0;
}
else
{
tr = convert_pixel (row1, 1) | mask1;
br = convert_pixel (row2, 1) | mask2;
}
}
buffer[i] = bilinear_interpolation (
tl, tr, bl, br, distx, disty);
next:
x += ux;
y += uy;
}
}
int
main (int argc, char *argv[])
{
bench_info_t binfo;
pixman_filter_t filter = PIXMAN_FILTER_NEAREST;
pixman_format_code_t src_format = PIXMAN_a8r8g8b8;
pixman_format_code_t mask_format = 0;
pixman_format_code_t dest_format = PIXMAN_a8r8g8b8;
pixman_box32_t dest_box = { 0, 0, WIDTH, HEIGHT };
box_48_16_t transformed = { 0 };
int32_t xmin, ymin, xmax, ymax;
uint32_t *src, *mask, *dest;
binfo.op = PIXMAN_OP_SRC;
binfo.mask_image = NULL;
pixman_transform_init_identity (&binfo.transform);
++argv;
if (*argv && (*argv)[0] == '-' && (*argv)[1] == 'n')
{
filter = PIXMAN_FILTER_NEAREST;
++argv;
--argc;
}
if (*argv && (*argv)[0] == '-' && (*argv)[1] == 'b')
{
filter = PIXMAN_FILTER_BILINEAR;
++argv;
--argc;
}
if (argc == 1 ||
!parse_arguments (argc, argv, &binfo.transform, &binfo.op,
&src_format, &mask_format, &dest_format))
{
printf ("Usage: affine-bench [-n] [-b] axx [axy] [ayx] [ayy] [combine type]\n");
printf (" [src format] [mask format] [dest format]\n");
printf (" -n : nearest scaling (default)\n");
printf (" -b : bilinear scaling\n");
printf (" axx : x_out:x_in factor\n");
printf (" axy : x_out:y_in factor (default 0)\n");
printf (" ayx : y_out:x_in factor (default 0)\n");
printf (" ayy : y_out:y_in factor (default 1)\n");
printf (" combine type : src, over, in etc (default src)\n");
printf (" src format : a8r8g8b8, r5g6b5 etc (default a8r8g8b8)\n");
printf (" mask format : as for src format, but no mask used if omitted\n");
printf (" dest format : as for src format (default a8r8g8b8)\n");
printf ("The output is a single number in megapixels/second.\n");
return EXIT_FAILURE;
}
/* Compute required extents for source and mask image so they qualify
* for COVER fast paths and get the flags in pixman.c:analyze_extent().
* These computations are for FAST_PATH_SAMPLES_COVER_CLIP_BILINEAR,
* but at the same time they also allow COVER_CLIP_NEAREST.
*/
compute_transformed_extents (&binfo.transform, &dest_box, &transformed);
xmin = pixman_fixed_to_int (transformed.x1 - pixman_fixed_1 / 2);
ymin = pixman_fixed_to_int (transformed.y1 - pixman_fixed_1 / 2);
xmax = pixman_fixed_to_int (transformed.x2 + pixman_fixed_1 / 2);
ymax = pixman_fixed_to_int (transformed.y2 + pixman_fixed_1 / 2);
/* Note:
* The upper limits can be reduced to the following when fetchers
* are guaranteed to not access pixels with zero weight. This concerns
* particularly all bilinear samplers.
*
* xmax = pixman_fixed_to_int (transformed.x2 + pixman_fixed_1 / 2 - pixman_fixed_e);
* ymax = pixman_fixed_to_int (transformed.y2 + pixman_fixed_1 / 2 - pixman_fixed_e);
* This is equivalent to subtracting 0.5 and rounding up, rather than
* subtracting 0.5, rounding down and adding 1.
*/
binfo.src_x = -xmin;
binfo.src_y = -ymin;
/* Always over-allocate width by 64 pixels for all src, mask and dst,
* so that we can iterate over an x-offset 0..63 in bench ().
* This is similar to lowlevel-blt-bench, which uses the same method
* to hit different cacheline misalignments.
*/
create_image (xmax - xmin + 64, ymax - ymin + 1, src_format, filter,
&src, &binfo.src_image);
if (mask_format)
{
create_image (xmax - xmin + 64, ymax - ymin + 1, mask_format, filter,
&mask, &binfo.mask_image);
if ((PIXMAN_FORMAT_R(mask_format) ||
PIXMAN_FORMAT_G(mask_format) ||
PIXMAN_FORMAT_B(mask_format)))
{
pixman_image_set_component_alpha (binfo.mask_image, 1);
}
}
create_image (WIDTH + 64, HEIGHT, dest_format, filter,
&dest, &binfo.dest_image);
run_benchmark (&binfo);
return EXIT_SUCCESS;
}
/*
* pixman_composite_trapezoids()
*
* All the trapezoids are conceptually rendered to an infinitely big image.
* The (0, 0) coordinates of this image are then aligned with the (x, y)
* coordinates of the source image, and then both images are aligned with
* the (x, y) coordinates of the destination. Then, in principle, compositing
* of these three images takes place across the entire destination.
*
* FIXME: However, there is currently a bug, where we restrict this compositing
* to the bounding box of the trapezoids. This is incorrect for operators such
* as SRC and IN where blank source pixels do have an effect on the destination.
*/
PIXMAN_EXPORT void
pixman_composite_trapezoids (pixman_op_t op,
pixman_image_t * src,
pixman_image_t * dst,
pixman_format_code_t mask_format,
int x_src,
int y_src,
int x_dst,
int y_dst,
int n_traps,
const pixman_trapezoid_t * traps)
{
int i;
if (n_traps <= 0)
return;
_pixman_image_validate (src);
_pixman_image_validate (dst);
if (op == PIXMAN_OP_ADD &&
(src->common.flags & FAST_PATH_IS_OPAQUE) &&
(mask_format == dst->common.extended_format_code) &&
!(dst->common.have_clip_region))
{
for (i = 0; i < n_traps; ++i)
{
const pixman_trapezoid_t *trap = &(traps[i]);
if (!pixman_trapezoid_valid (trap))
continue;
pixman_rasterize_trapezoid (dst, trap, x_dst, y_dst);
}
}
else
{
pixman_image_t *tmp;
pixman_box32_t box;
box.x1 = INT32_MAX;
box.y1 = INT32_MAX;
box.x2 = INT32_MIN;
box.y2 = INT32_MIN;
for (i = 0; i < n_traps; ++i)
{
const pixman_trapezoid_t *trap = &(traps[i]);
int y1, y2;
if (!pixman_trapezoid_valid (trap))
continue;
y1 = pixman_fixed_to_int (trap->top);
if (y1 < box.y1)
box.y1 = y1;
y2 = pixman_fixed_to_int (pixman_fixed_ceil (trap->bottom));
if (y2 > box.y2)
box.y2 = y2;
#define EXTEND_MIN(x) \
if (pixman_fixed_to_int ((x)) < box.x1) \
box.x1 = pixman_fixed_to_int ((x));
#define EXTEND_MAX(x) \
if (pixman_fixed_to_int (pixman_fixed_ceil ((x))) > box.x2) \
box.x2 = pixman_fixed_to_int (pixman_fixed_ceil ((x)));
#define EXTEND(x) \
EXTEND_MIN(x); \
EXTEND_MAX(x);
EXTEND(trap->left.p1.x);
EXTEND(trap->left.p2.x);
EXTEND(trap->right.p1.x);
EXTEND(trap->right.p2.x);
}
if (box.x1 >= box.x2 || box.y1 >= box.y2)
return;
tmp = pixman_image_create_bits (
mask_format, box.x2 - box.x1, box.y2 - box.y1, NULL, -1);
for (i = 0; i < n_traps; ++i)
{
const pixman_trapezoid_t *trap = &(traps[i]);
if (!pixman_trapezoid_valid (trap))
continue;
pixman_rasterize_trapezoid (tmp, trap, - box.x1, - box.y1);
}
pixman_image_composite (op, src, tmp, dst,
x_src + box.x1, y_src + box.y1,
0, 0,
x_dst + box.x1, y_dst + box.y1,
box.x2 - box.x1, box.y2 - box.y1);
pixman_image_unref (tmp);
}
}
static void
ssse3_fetch_horizontal (bits_image_t *image, line_t *line,
int y, pixman_fixed_t x, pixman_fixed_t ux, int n)
{
uint32_t *bits = image->bits + y * image->rowstride;
__m128i vx = _mm_set_epi16 (
- (x + 1), x, - (x + 1), x,
- (x + ux + 1), x + ux, - (x + ux + 1), x + ux);
__m128i vux = _mm_set_epi16 (
- 2 * ux, 2 * ux, - 2 * ux, 2 * ux,
- 2 * ux, 2 * ux, - 2 * ux, 2 * ux);
__m128i vaddc = _mm_set_epi16 (1, 0, 1, 0, 1, 0, 1, 0);
__m128i *b = (__m128i *)line->buffer;
__m128i vrl0, vrl1;
while ((n -= 2) >= 0)
{
__m128i vw, vr, s;
vrl1 = _mm_loadl_epi64 (
(__m128i *)(bits + pixman_fixed_to_int (x + ux)));
/* vrl1: R1, L1 */
final_pixel:
vrl0 = _mm_loadl_epi64 (
(__m128i *)(bits + pixman_fixed_to_int (x)));
/* vrl0: R0, L0 */
/* The weights are based on vx which is a vector of
*
* - (x + 1), x, - (x + 1), x,
* - (x + ux + 1), x + ux, - (x + ux + 1), x + ux
*
* so the 16 bit weights end up like this:
*
* iw0, w0, iw0, w0, iw1, w1, iw1, w1
*
* and after shifting and packing, we get these bytes:
*
* iw0, w0, iw0, w0, iw1, w1, iw1, w1,
* iw0, w0, iw0, w0, iw1, w1, iw1, w1,
*
* which means the first and the second input pixel
* have to be interleaved like this:
*
* la0, ra0, lr0, rr0, la1, ra1, lr1, rr1,
* lg0, rg0, lb0, rb0, lg1, rg1, lb1, rb1
*
* before maddubsw can be used.
*/
vw = _mm_add_epi16 (
vaddc, _mm_srli_epi16 (vx, 16 - BILINEAR_INTERPOLATION_BITS));
/* vw: iw0, w0, iw0, w0, iw1, w1, iw1, w1
*/
vw = _mm_packus_epi16 (vw, vw);
/* vw: iw0, w0, iw0, w0, iw1, w1, iw1, w1,
* iw0, w0, iw0, w0, iw1, w1, iw1, w1
*/
vx = _mm_add_epi16 (vx, vux);
x += 2 * ux;
vr = _mm_unpacklo_epi16 (vrl1, vrl0);
/* vr: rar0, rar1, rgb0, rgb1, lar0, lar1, lgb0, lgb1 */
s = _mm_shuffle_epi32 (vr, _MM_SHUFFLE (1, 0, 3, 2));
/* s: lar0, lar1, lgb0, lgb1, rar0, rar1, rgb0, rgb1 */
vr = _mm_unpackhi_epi8 (vr, s);
/* vr: la0, ra0, lr0, rr0, la1, ra1, lr1, rr1,
* lg0, rg0, lb0, rb0, lg1, rg1, lb1, rb1
*/
vr = _mm_maddubs_epi16 (vr, vw);
/* When the weight is 0, the inverse weight is
* 128 which can't be represented in a signed byte.
* As a result maddubsw computes the following:
*
* r = l * -128 + r * 0
*
* rather than the desired
*
* r = l * 128 + r * 0
*
* We fix this by taking the absolute value of the
* result.
*/
vr = _mm_abs_epi16 (vr);
/* vr: A0, R0, A1, R1, G0, B0, G1, B1 */
_mm_store_si128 (b++, vr);
}
if (n == -1)
{
vrl1 = _mm_setzero_si128();
goto final_pixel;
}
line->y = y;
}
static uint32_t *
ssse3_fetch_bilinear_cover (pixman_iter_t *iter, const uint32_t *mask)
{
pixman_fixed_t fx, ux;
bilinear_info_t *info = iter->data;
line_t *line0, *line1;
int y0, y1;
int32_t dist_y;
__m128i vw;
int i;
fx = info->x;
ux = iter->image->common.transform->matrix[0][0];
y0 = pixman_fixed_to_int (info->y);
y1 = y0 + 1;
line0 = &info->lines[y0 & 0x01];
line1 = &info->lines[y1 & 0x01];
if (line0->y != y0)
{
ssse3_fetch_horizontal (
&iter->image->bits, line0, y0, fx, ux, iter->width);
}
if (line1->y != y1)
{
ssse3_fetch_horizontal (
&iter->image->bits, line1, y1, fx, ux, iter->width);
}
dist_y = pixman_fixed_to_bilinear_weight (info->y);
dist_y <<= (16 - BILINEAR_INTERPOLATION_BITS);
vw = _mm_set_epi16 (
dist_y, dist_y, dist_y, dist_y, dist_y, dist_y, dist_y, dist_y);
for (i = 0; i + 3 < iter->width; i += 4)
{
__m128i top0 = _mm_load_si128 ((__m128i *)(line0->buffer + i));
__m128i bot0 = _mm_load_si128 ((__m128i *)(line1->buffer + i));
__m128i top1 = _mm_load_si128 ((__m128i *)(line0->buffer + i + 2));
__m128i bot1 = _mm_load_si128 ((__m128i *)(line1->buffer + i + 2));
__m128i r0, r1, tmp, p;
r0 = _mm_mulhi_epu16 (
_mm_sub_epi16 (bot0, top0), vw);
tmp = _mm_cmplt_epi16 (bot0, top0);
tmp = _mm_and_si128 (tmp, vw);
r0 = _mm_sub_epi16 (r0, tmp);
r0 = _mm_add_epi16 (r0, top0);
r0 = _mm_srli_epi16 (r0, BILINEAR_INTERPOLATION_BITS);
/* r0: A0 R0 A1 R1 G0 B0 G1 B1 */
r0 = _mm_shuffle_epi32 (r0, _MM_SHUFFLE (2, 0, 3, 1));
/* r0: A1 R1 G1 B1 A0 R0 G0 B0 */
r1 = _mm_mulhi_epu16 (
_mm_sub_epi16 (bot1, top1), vw);
tmp = _mm_cmplt_epi16 (bot1, top1);
tmp = _mm_and_si128 (tmp, vw);
r1 = _mm_sub_epi16 (r1, tmp);
r1 = _mm_add_epi16 (r1, top1);
r1 = _mm_srli_epi16 (r1, BILINEAR_INTERPOLATION_BITS);
r1 = _mm_shuffle_epi32 (r1, _MM_SHUFFLE (2, 0, 3, 1));
/* r1: A3 R3 G3 B3 A2 R2 G2 B2 */
p = _mm_packus_epi16 (r0, r1);
_mm_storeu_si128 ((__m128i *)(iter->buffer + i), p);
}
while (i < iter->width)
{
__m128i top0 = _mm_load_si128 ((__m128i *)(line0->buffer + i));
__m128i bot0 = _mm_load_si128 ((__m128i *)(line1->buffer + i));
__m128i r0, tmp, p;
r0 = _mm_mulhi_epu16 (
_mm_sub_epi16 (bot0, top0), vw);
tmp = _mm_cmplt_epi16 (bot0, top0);
tmp = _mm_and_si128 (tmp, vw);
r0 = _mm_sub_epi16 (r0, tmp);
r0 = _mm_add_epi16 (r0, top0);
r0 = _mm_srli_epi16 (r0, BILINEAR_INTERPOLATION_BITS);
/* r0: A0 R0 A1 R1 G0 B0 G1 B1 */
r0 = _mm_shuffle_epi32 (r0, _MM_SHUFFLE (2, 0, 3, 1));
/* r0: A1 R1 G1 B1 A0 R0 G0 B0 */
p = _mm_packus_epi16 (r0, r0);
if (iter->width - i == 1)
{
*(uint32_t *)(iter->buffer + i) = _mm_cvtsi128_si32 (p);
i++;
}
else
{
_mm_storel_epi64 ((__m128i *)(iter->buffer + i), p);
i += 2;
}
}
info->y += iter->image->common.transform->matrix[1][1];
return iter->buffer;
}
static pixman_bool_t
get_trap_extents (pixman_op_t op, pixman_image_t *dest,
const pixman_trapezoid_t *traps, int n_traps,
pixman_box32_t *box)
{
int i;
/* When the operator is such that a zero source has an
* effect on the underlying image, we have to
* composite across the entire destination
*/
if (!zero_src_has_no_effect [op])
{
box->x1 = 0;
box->y1 = 0;
box->x2 = dest->bits.width;
box->y2 = dest->bits.height;
return TRUE;
}
box->x1 = INT32_MAX;
box->y1 = INT32_MAX;
box->x2 = INT32_MIN;
box->y2 = INT32_MIN;
for (i = 0; i < n_traps; ++i)
{
const pixman_trapezoid_t *trap = &(traps[i]);
int y1, y2;
if (!pixman_trapezoid_valid (trap))
continue;
y1 = pixman_fixed_to_int (trap->top);
if (y1 < box->y1)
box->y1 = y1;
y2 = pixman_fixed_to_int (pixman_fixed_ceil (trap->bottom));
if (y2 > box->y2)
box->y2 = y2;
#define EXTEND_MIN(x) \
if (pixman_fixed_to_int ((x)) < box->x1) \
box->x1 = pixman_fixed_to_int ((x));
#define EXTEND_MAX(x) \
if (pixman_fixed_to_int (pixman_fixed_ceil ((x))) > box->x2) \
box->x2 = pixman_fixed_to_int (pixman_fixed_ceil ((x)));
#define EXTEND(x) \
EXTEND_MIN(x); \
EXTEND_MAX(x);
EXTEND(trap->left.p1.x);
EXTEND(trap->left.p2.x);
EXTEND(trap->right.p1.x);
EXTEND(trap->right.p2.x);
}
if (box->x1 >= box->x2 || box->y1 >= box->y2)
return FALSE;
return TRUE;
}
static force_inline void
bits_image_fetch_nearest_affine (pixman_image_t * image,
int offset,
int line,
int width,
uint32_t * buffer,
const uint32_t * mask,
convert_pixel_t convert_pixel,
pixman_format_code_t format,
pixman_repeat_t repeat_mode)
{
pixman_fixed_t x, y;
pixman_fixed_t ux, uy;
pixman_vector_t v;
bits_image_t *bits = &image->bits;
int i;
/* reference point is the center of the pixel */
v.vector[0] = pixman_int_to_fixed (offset) + pixman_fixed_1 / 2;
v.vector[1] = pixman_int_to_fixed (line) + pixman_fixed_1 / 2;
v.vector[2] = pixman_fixed_1;
if (!pixman_transform_point_3d (image->common.transform, &v))
return;
ux = image->common.transform->matrix[0][0];
uy = image->common.transform->matrix[1][0];
x = v.vector[0];
y = v.vector[1];
for (i = 0; i < width; ++i)
{
int width, height, x0, y0;
const uint8_t *row;
if (mask && !mask[i])
goto next;
width = image->bits.width;
height = image->bits.height;
x0 = pixman_fixed_to_int (x - pixman_fixed_e);
y0 = pixman_fixed_to_int (y - pixman_fixed_e);
if (repeat_mode == PIXMAN_REPEAT_NONE &&
(y0 < 0 || y0 >= height || x0 < 0 || x0 >= width))
{
buffer[i] = 0;
}
else
{
uint32_t mask = PIXMAN_FORMAT_A (format)? 0 : 0xff000000;
if (repeat_mode != PIXMAN_REPEAT_NONE)
{
repeat (repeat_mode, &x0, width);
repeat (repeat_mode, &y0, height);
}
row = (uint8_t *)bits->bits + bits->rowstride * 4 * y0;
buffer[i] = convert_pixel (row, x0) | mask;
}
next:
x += ux;
y += uy;
}
}
/*
* We want to detect the case where we add the same value to a long
* span of pixels. The triangles on the end are filled in while we
* count how many sub-pixel scanlines contribute to the middle section.
*
* +--------------------------+
* fill_height =| \ /
* +------------------+
* |================|
* fill_start fill_end
*/
static void
rasterize_edges_8 (pixman_image_t *image,
pixman_edge_t * l,
pixman_edge_t * r,
pixman_fixed_t t,
pixman_fixed_t b)
{
pixman_fixed_t y = t;
uint32_t *line;
int fill_start = -1, fill_end = -1;
int fill_size = 0;
uint32_t *buf = (image)->bits.bits;
int stride = (image)->bits.rowstride;
int width = (image)->bits.width;
line = buf + pixman_fixed_to_int (y) * stride;
for (;;)
{
uint8_t *ap = (uint8_t *) line;
pixman_fixed_t lx, rx;
int lxi, rxi;
/* clip X */
lx = l->x;
if (lx < 0)
lx = 0;
rx = r->x;
if (pixman_fixed_to_int (rx) >= width)
{
/* Use the last pixel of the scanline, covered 100%.
* We can't use the first pixel following the scanline,
* because accessing it could result in a buffer overrun.
*/
rx = pixman_int_to_fixed (width) - 1;
}
/* Skip empty (or backwards) sections */
if (rx > lx)
{
int lxs, rxs;
/* Find pixel bounds for span. */
lxi = pixman_fixed_to_int (lx);
rxi = pixman_fixed_to_int (rx);
/* Sample coverage for edge pixels */
lxs = RENDER_SAMPLES_X (lx, 8);
rxs = RENDER_SAMPLES_X (rx, 8);
/* Add coverage across row */
if (lxi == rxi)
{
WRITE (image, ap + lxi,
clip255 (READ (image, ap + lxi) + rxs - lxs));
}
else
{
WRITE (image, ap + lxi,
clip255 (READ (image, ap + lxi) + N_X_FRAC (8) - lxs));
/* Move forward so that lxi/rxi is the pixel span */
lxi++;
/* Don't bother trying to optimize the fill unless
* the span is longer than 4 pixels. */
if (rxi - lxi > 4)
{
if (fill_start < 0)
{
fill_start = lxi;
fill_end = rxi;
fill_size++;
}
else
{
if (lxi >= fill_end || rxi < fill_start)
{
/* We're beyond what we saved, just fill it */
ADD_SATURATE_8 (ap + fill_start,
fill_size * N_X_FRAC (8),
fill_end - fill_start);
fill_start = lxi;
fill_end = rxi;
fill_size = 1;
}
else
{
/* Update fill_start */
if (lxi > fill_start)
{
ADD_SATURATE_8 (ap + fill_start,
fill_size * N_X_FRAC (8),
lxi - fill_start);
fill_start = lxi;
}
else if (lxi < fill_start)
{
ADD_SATURATE_8 (ap + lxi, N_X_FRAC (8),
fill_start - lxi);
}
/* Update fill_end */
if (rxi < fill_end)
{
ADD_SATURATE_8 (ap + rxi,
fill_size * N_X_FRAC (8),
fill_end - rxi);
fill_end = rxi;
}
else if (fill_end < rxi)
{
ADD_SATURATE_8 (ap + fill_end,
N_X_FRAC (8),
rxi - fill_end);
}
fill_size++;
}
}
}
else
{
ADD_SATURATE_8 (ap + lxi, N_X_FRAC (8), rxi - lxi);
}
WRITE (image, ap + rxi, clip255 (READ (image, ap + rxi) + rxs));
}
}
if (y == b)
{
/* We're done, make sure we clean up any remaining fill. */
if (fill_start != fill_end)
{
if (fill_size == N_Y_FRAC (8))
{
MEMSET_WRAPPED (image, ap + fill_start,
0xff, fill_end - fill_start);
}
else
{
ADD_SATURATE_8 (ap + fill_start, fill_size * N_X_FRAC (8),
fill_end - fill_start);
}
}
break;
}
if (pixman_fixed_frac (y) != Y_FRAC_LAST (8))
{
RENDER_EDGE_STEP_SMALL (l);
RENDER_EDGE_STEP_SMALL (r);
y += STEP_Y_SMALL (8);
}
else
{
RENDER_EDGE_STEP_BIG (l);
RENDER_EDGE_STEP_BIG (r);
y += STEP_Y_BIG (8);
if (fill_start != fill_end)
{
if (fill_size == N_Y_FRAC (8))
{
MEMSET_WRAPPED (image, ap + fill_start,
0xff, fill_end - fill_start);
}
else
{
ADD_SATURATE_8 (ap + fill_start, fill_size * N_X_FRAC (8),
fill_end - fill_start);
}
fill_start = fill_end = -1;
fill_size = 0;
}
line += stride;
}
}
}
static uint32_t *
bits_image_fetch_bilinear_no_repeat_8888 (pixman_iter_t *iter,
const uint32_t *mask)
{
pixman_image_t * ima = iter->image;
int offset = iter->x;
int line = iter->y++;
int width = iter->width;
uint32_t * buffer = iter->buffer;
bits_image_t *bits = &ima->bits;
pixman_fixed_t x_top, x_bottom, x;
pixman_fixed_t ux_top, ux_bottom, ux;
pixman_vector_t v;
uint32_t top_mask, bottom_mask;
uint32_t *top_row;
uint32_t *bottom_row;
uint32_t *end;
uint32_t zero[2] = { 0, 0 };
uint32_t one = 1;
int y, y1, y2;
int disty;
int mask_inc;
int w;
/* reference point is the center of the pixel */
v.vector[0] = pixman_int_to_fixed (offset) + pixman_fixed_1 / 2;
v.vector[1] = pixman_int_to_fixed (line) + pixman_fixed_1 / 2;
v.vector[2] = pixman_fixed_1;
if (!pixman_transform_point_3d (bits->common.transform, &v))
return iter->buffer;
ux = ux_top = ux_bottom = bits->common.transform->matrix[0][0];
x = x_top = x_bottom = v.vector[0] - pixman_fixed_1/2;
y = v.vector[1] - pixman_fixed_1/2;
disty = pixman_fixed_to_bilinear_weight (y);
/* Load the pointers to the first and second lines from the source
* image that bilinear code must read.
*
* The main trick in this code is about the check if any line are
* outside of the image;
*
* When I realize that a line (any one) is outside, I change
* the pointer to a dummy area with zeros. Once I change this, I
* must be sure the pointer will not change, so I set the
* variables to each pointer increments inside the loop.
*/
y1 = pixman_fixed_to_int (y);
y2 = y1 + 1;
if (y1 < 0 || y1 >= bits->height)
{
top_row = zero;
x_top = 0;
ux_top = 0;
}
else
{
top_row = bits->bits + y1 * bits->rowstride;
x_top = x;
ux_top = ux;
}
if (y2 < 0 || y2 >= bits->height)
{
bottom_row = zero;
x_bottom = 0;
ux_bottom = 0;
}
else
{
bottom_row = bits->bits + y2 * bits->rowstride;
x_bottom = x;
ux_bottom = ux;
}
/* Instead of checking whether the operation uses the mast in
* each loop iteration, verify this only once and prepare the
* variables to make the code smaller inside the loop.
*/
if (!mask)
{
mask_inc = 0;
mask = &one;
}
else
{
/* If have a mask, prepare the variables to check it */
mask_inc = 1;
}
/* If both are zero, then the whole thing is zero */
if (top_row == zero && bottom_row == zero)
{
memset (buffer, 0, width * sizeof (uint32_t));
return iter->buffer;
}
else if (bits->format == PIXMAN_x8r8g8b8)
{
if (top_row == zero)
{
top_mask = 0;
bottom_mask = 0xff000000;
}
else if (bottom_row == zero)
{
top_mask = 0xff000000;
bottom_mask = 0;
}
else
{
top_mask = 0xff000000;
bottom_mask = 0xff000000;
}
}
else
{
top_mask = 0;
bottom_mask = 0;
}
end = buffer + width;
/* Zero fill to the left of the image */
while (buffer < end && x < pixman_fixed_minus_1)
{
*buffer++ = 0;
x += ux;
x_top += ux_top;
x_bottom += ux_bottom;
mask += mask_inc;
}
/* Left edge
*/
while (buffer < end && x < 0)
{
uint32_t tr, br;
int32_t distx;
tr = top_row[pixman_fixed_to_int (x_top) + 1] | top_mask;
br = bottom_row[pixman_fixed_to_int (x_bottom) + 1] | bottom_mask;
distx = pixman_fixed_to_bilinear_weight (x);
*buffer++ = bilinear_interpolation (0, tr, 0, br, distx, disty);
x += ux;
x_top += ux_top;
x_bottom += ux_bottom;
mask += mask_inc;
}
/* Main part */
w = pixman_int_to_fixed (bits->width - 1);
while (buffer < end && x < w)
{
if (*mask)
{
uint32_t tl, tr, bl, br;
int32_t distx;
tl = top_row [pixman_fixed_to_int (x_top)] | top_mask;
tr = top_row [pixman_fixed_to_int (x_top) + 1] | top_mask;
bl = bottom_row [pixman_fixed_to_int (x_bottom)] | bottom_mask;
br = bottom_row [pixman_fixed_to_int (x_bottom) + 1] | bottom_mask;
distx = pixman_fixed_to_bilinear_weight (x);
*buffer = bilinear_interpolation (tl, tr, bl, br, distx, disty);
}
buffer++;
x += ux;
x_top += ux_top;
x_bottom += ux_bottom;
mask += mask_inc;
}
/* Right Edge */
w = pixman_int_to_fixed (bits->width);
while (buffer < end && x < w)
{
if (*mask)
{
uint32_t tl, bl;
int32_t distx;
tl = top_row [pixman_fixed_to_int (x_top)] | top_mask;
bl = bottom_row [pixman_fixed_to_int (x_bottom)] | bottom_mask;
distx = pixman_fixed_to_bilinear_weight (x);
*buffer = bilinear_interpolation (tl, 0, bl, 0, distx, disty);
}
buffer++;
x += ux;
x_top += ux_top;
x_bottom += ux_bottom;
mask += mask_inc;
}
/* Zero fill to the left of the image */
while (buffer < end)
*buffer++ = 0;
return iter->buffer;
}
static force_inline uint32_t
bits_image_fetch_pixel_convolution (bits_image_t *image,
pixman_fixed_t x,
pixman_fixed_t y,
get_pixel_t get_pixel)
{
pixman_fixed_t *params = image->common.filter_params;
int x_off = (params[0] - pixman_fixed_1) >> 1;
int y_off = (params[1] - pixman_fixed_1) >> 1;
int32_t cwidth = pixman_fixed_to_int (params[0]);
int32_t cheight = pixman_fixed_to_int (params[1]);
int32_t i, j, x1, x2, y1, y2;
pixman_repeat_t repeat_mode = image->common.repeat;
int width = image->width;
int height = image->height;
int srtot, sgtot, sbtot, satot;
params += 2;
x1 = pixman_fixed_to_int (x - pixman_fixed_e - x_off);
y1 = pixman_fixed_to_int (y - pixman_fixed_e - y_off);
x2 = x1 + cwidth;
y2 = y1 + cheight;
srtot = sgtot = sbtot = satot = 0;
for (i = y1; i < y2; ++i)
{
for (j = x1; j < x2; ++j)
{
int rx = j;
int ry = i;
pixman_fixed_t f = *params;
if (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);
}
else
{
pixel = get_pixel (image, rx, ry, TRUE);
}
srtot += (int)RED_8 (pixel) * f;
sgtot += (int)GREEN_8 (pixel) * f;
sbtot += (int)BLUE_8 (pixel) * f;
satot += (int)ALPHA_8 (pixel) * f;
}
params++;
}
}
satot >>= 16;
srtot >>= 16;
sgtot >>= 16;
sbtot >>= 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));
}
