xf86-video-intel: 2 commits - src/sna/sna_accel.c src/sna/sna.h src/sna/sna_trapezoids_imprecise.c
Chris Wilson
ickle at kemper.freedesktop.org
Mon Oct 6 02:26:04 PDT 2014
src/sna/sna.h | 1
src/sna/sna_accel.c | 38 +++++++++----
src/sna/sna_trapezoids_imprecise.c | 108 +++++--------------------------------
3 files changed, 45 insertions(+), 102 deletions(-)
New commits:
commit ec2b9ac81aed0d2dda2948171ca1c260184bf221
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Mon Oct 6 08:18:24 2014 +0100
sna: Avoid the intermediate allocation for PRIME pixmaps
If we attempt to share an unused Pixmap, we first create an ordinary GPU
bo when migrating onto the GPU. Add a flag here to cause the migration
to create a Pixmap suitable for PRIME instead.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna.h b/src/sna/sna.h
index 04ee35c..e035634 100644
--- a/src/sna/sna.h
+++ b/src/sna/sna.h
@@ -709,6 +709,7 @@ sna_pixmap_undo_cow(struct sna *sna, struct sna_pixmap *priv, unsigned flags);
#define __MOVE_DRI 0x80
#define __MOVE_SCANOUT 0x100
#define __MOVE_TILED 0x200
+#define __MOVE_PRIME 0x400
struct sna_pixmap *
sna_pixmap_move_area_to_gpu(PixmapPtr pixmap, const BoxRec *box, unsigned int flags);
diff --git a/src/sna/sna_accel.c b/src/sna/sna_accel.c
index 02ff0c7..a6ba2a0 100644
--- a/src/sna/sna_accel.c
+++ b/src/sna/sna_accel.c
@@ -1093,7 +1093,7 @@ sna_share_pixmap_backing(PixmapPtr pixmap, ScreenPtr slave, void **fd_handle)
DBG(("%s: pixmap=%ld\n", __FUNCTION__, pixmap->drawable.serialNumber));
priv = sna_pixmap_move_to_gpu(pixmap,
- MOVE_READ | MOVE_WRITE | __MOVE_DRI | __MOVE_FORCE);
+ MOVE_READ | MOVE_WRITE | __MOVE_DRI | __MOVE_PRIME | __MOVE_FORCE);
if (priv == NULL)
return FALSE;
@@ -1125,7 +1125,7 @@ sna_share_pixmap_backing(PixmapPtr pixmap, ScreenPtr slave, void **fd_handle)
pixmap->drawable.height,
pixmap->drawable.bitsPerPixel,
I915_TILING_NONE,
- CREATE_GTT_MAP | CREATE_PRIME);
+ CREATE_GTT_MAP | CREATE_PRIME | CREATE_EXACT);
if (bo == NULL) {
DBG(("%s: allocation failed\n", __FUNCTION__));
return FALSE;
@@ -1244,7 +1244,7 @@ sna_create_pixmap_shared(struct sna *sna, ScreenPtr screen,
width, height,
pixmap->drawable.bitsPerPixel,
I915_TILING_NONE,
- CREATE_GTT_MAP | CREATE_PRIME);
+ CREATE_GTT_MAP | CREATE_PRIME | CREATE_EXACT);
if (priv->gpu_bo == NULL) {
free(priv);
FreePixmap(pixmap);
@@ -1921,6 +1921,7 @@ sna_pixmap_undo_cow(struct sna *sna, struct sna_pixmap *priv, unsigned flags)
if (flags & MOVE_READ) {
PixmapPtr pixmap = priv->pixmap;
+ unsigned create, tiling;
BoxRec box;
DBG(("%s: copying cow\n", __FUNCTION__));
@@ -1929,11 +1930,18 @@ sna_pixmap_undo_cow(struct sna *sna, struct sna_pixmap *priv, unsigned flags)
box.x2 = pixmap->drawable.width;
box.y2 = pixmap->drawable.height;
+ if (flags & __MOVE_PRIME) {
+ create|= CREATE_GTT_MAP | CREATE_PRIME | CREATE_EXACT;
+ tiling = I915_TILING_NONE;
+ } else {
+ create = 0;
+ tiling = sna_pixmap_default_tiling(sna, pixmap);
+ }
+
bo = kgem_create_2d(&sna->kgem,
box.x2, box.y2,
pixmap->drawable.bitsPerPixel,
- sna_pixmap_default_tiling(sna, pixmap),
- 0);
+ tiling, create);
if (bo == NULL) {
cow->refcnt++;
DBG(("%s: allocation failed\n", __FUNCTION__));
@@ -4073,6 +4081,10 @@ static bool can_convert_to_gpu(struct sna_pixmap *priv, unsigned flags)
if (priv->cpu_bo->pitch & 63)
return false;
+ if (flags & __MOVE_PRIME)
+ if (priv->cpu_bo->pitch & 255)
+ return false;
+
return true;
}
@@ -4160,10 +4172,16 @@ sna_pixmap_move_to_gpu(PixmapPtr pixmap, unsigned flags)
assert(pixmap->drawable.height > 0);
assert(pixmap->drawable.bitsPerPixel >= 8);
- is_linear = sna_pixmap_default_tiling(sna, pixmap) == I915_TILING_NONE;
- if (is_linear && flags & __MOVE_TILED) {
- DBG(("%s: not creating linear GPU bo\n", __FUNCTION__));
- return NULL;
+ if (flags & __MOVE_PRIME) {
+ assert((flags & __MOVE_TILED) == 0);
+ is_linear = true;
+ } else {
+ is_linear = sna_pixmap_default_tiling(sna, pixmap) == I915_TILING_NONE;
+ if (is_linear && flags & __MOVE_TILED) {
+ DBG(("%s: not creating linear GPU bo\n",
+ __FUNCTION__));
+ return NULL;
+ }
}
if (is_linear &&
@@ -4185,6 +4203,8 @@ sna_pixmap_move_to_gpu(PixmapPtr pixmap, unsigned flags)
unsigned create = 0;
if (flags & MOVE_INPLACE_HINT || (priv->cpu_damage && priv->cpu_bo == NULL))
create = CREATE_GTT_MAP | CREATE_INACTIVE;
+ if (flags & __MOVE_PRIME)
+ create |= CREATE_GTT_MAP | CREATE_PRIME | CREATE_EXACT;
sna_pixmap_alloc_gpu(sna, pixmap, priv, create);
}
commit 649b233f12bdf04c449a813c3a150ac3ccfcab72
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Sun Oct 5 11:06:36 2014 +0100
sna/trapezoids: Remove redundant coverage scaling
Since we do not perform an analytical pass, we can count samples directly.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_trapezoids_imprecise.c b/src/sna/sna_trapezoids_imprecise.c
index b005c97..865a723 100644
--- a/src/sna/sna_trapezoids_imprecise.c
+++ b/src/sna/sna_trapezoids_imprecise.c
@@ -124,7 +124,7 @@ inline static void apply_damage_box(struct sna_composite_op *op, const BoxRec *b
(i) = FAST_SAMPLES_INT(t); \
} while (0)
-#define FAST_SAMPLES_XY (2*FAST_SAMPLES_X*FAST_SAMPLES_Y) /* Unit area on the grid. */
+#define FAST_SAMPLES_XY (FAST_SAMPLES_X*FAST_SAMPLES_Y) /* Unit area on the grid. */
#define AREA_TO_ALPHA(c) ((c) / (float)FAST_SAMPLES_XY)
struct quorem {
@@ -374,14 +374,14 @@ cell_list_add_subspan(struct cell_list *cells, int x1, int x2)
cell = cell_list_find(cells, ix1);
if (ix1 != ix2) {
- cell->uncovered_area += 2*fx1;
+ cell->uncovered_area += fx1;
++cell->covered_height;
cell = cell_list_find(cells, ix2);
- cell->uncovered_area -= 2*fx2;
+ cell->uncovered_area -= fx2;
--cell->covered_height;
} else
- cell->uncovered_area += 2*(fx1-fx2);
+ cell->uncovered_area += (fx1-fx2);
}
inline static void
@@ -399,14 +399,14 @@ cell_list_add_span(struct cell_list *cells, int x1, int x2)
cell = cell_list_find(cells, ix1);
if (ix1 != ix2) {
- cell->uncovered_area += 2*fx1*FAST_SAMPLES_Y;
+ cell->uncovered_area += fx1*FAST_SAMPLES_Y;
cell->covered_height += FAST_SAMPLES_Y;
cell = cell_list_find(cells, ix2);
- cell->uncovered_area -= 2*fx2*FAST_SAMPLES_Y;
+ cell->uncovered_area -= fx2*FAST_SAMPLES_Y;
cell->covered_height -= FAST_SAMPLES_Y;
} else
- cell->uncovered_area += 2*(fx1-fx2)*FAST_SAMPLES_Y;
+ cell->uncovered_area += (fx1-fx2)*FAST_SAMPLES_Y;
}
static void
@@ -1094,7 +1094,7 @@ tor_blt(struct sna *sna,
box.x1 = converter->extents.x1;
/* Form the spans from the coverages and areas. */
- cover = cells->head.covered_height*FAST_SAMPLES_X*2;
+ cover = cells->head.covered_height*FAST_SAMPLES_X;
assert(cover >= 0);
for (cell = cells->head.next; cell != &cells->tail; cell = cell->next) {
int x = cell->x;
@@ -1116,7 +1116,7 @@ tor_blt(struct sna *sna,
span(sna, op, clip, &box, cover);
}
box.x1 = box.x2;
- cover += cell->covered_height*FAST_SAMPLES_X*2;
+ cover += cell->covered_height*FAST_SAMPLES_X;
}
if (cell->uncovered_area) {
@@ -1451,7 +1451,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
buf += 4;
if (dw == 0) {
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
v |= v << 8;
dw = v | v << 16;
@@ -1459,7 +1459,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
cover += (int8_t)(dw & 0xff);
if (cover) {
assert(cover > 0);
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
dw >>= 8;
dw |= v << 24;
@@ -1469,7 +1469,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
cover += (int8_t)(dw & 0xff);
if (cover) {
assert(cover > 0);
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
dw >>= 8;
dw |= v << 24;
@@ -1479,7 +1479,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
cover += (int8_t)(dw & 0xff);
if (cover) {
assert(cover > 0);
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
dw >>= 8;
dw |= v << 24;
@@ -1489,7 +1489,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
cover += (int8_t)(dw & 0xff);
if (cover) {
assert(cover > 0);
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
dw >>= 8;
dw |= v << 24;
@@ -1508,7 +1508,7 @@ inplace_end_subrows(struct active_list *active, uint8_t *row,
cover += *buf++;
assert(cover >= 0);
- v = cover * 256 / (FAST_SAMPLES_X * FAST_SAMPLES_Y);
+ v = cover * 256 / FAST_SAMPLES_XY;
v -= v >> 8;
*row++ = v;
}
@@ -1631,39 +1631,6 @@ static int operator_is_bounded(uint8_t op)
}
}
-static inline bool
-project_trapezoid_onto_grid(const xTrapezoid *in,
- int dx, int dy,
- xTrapezoid *out)
-{
- __DBG(("%s: in: L:(%d, %d), (%d, %d); R:(%d, %d), (%d, %d), [%d, %d]\n",
- __FUNCTION__,
- in->left.p1.x, in->left.p1.y, in->left.p2.x, in->left.p2.y,
- in->right.p1.x, in->right.p1.y, in->right.p2.x, in->right.p2.y,
- in->top, in->bottom));
-
- out->left.p1.x = dx + pixman_fixed_to_fast(in->left.p1.x);
- out->left.p1.y = dy + pixman_fixed_to_fast(in->left.p1.y);
- out->left.p2.x = dx + pixman_fixed_to_fast(in->left.p2.x);
- out->left.p2.y = dy + pixman_fixed_to_fast(in->left.p2.y);
-
- out->right.p1.x = dx + pixman_fixed_to_fast(in->right.p1.x);
- out->right.p1.y = dy + pixman_fixed_to_fast(in->right.p1.y);
- out->right.p2.x = dx + pixman_fixed_to_fast(in->right.p2.x);
- out->right.p2.y = dy + pixman_fixed_to_fast(in->right.p2.y);
-
- out->top = dy + pixman_fixed_to_fast(in->top);
- out->bottom = dy + pixman_fixed_to_fast(in->bottom);
-
- __DBG(("%s: out: L:(%d, %d), (%d, %d); R:(%d, %d), (%d, %d), [%d, %d]\n",
- __FUNCTION__,
- out->left.p1.x, out->left.p1.y, out->left.p2.x, out->left.p2.y,
- out->right.p1.x, out->right.p1.y, out->right.p2.x, out->right.p2.y,
- out->top, out->bottom));
-
- return xTrapezoidValid(out);
-}
-
static span_func_t
choose_span(struct sna_composite_spans_op *tmp,
PicturePtr dst,
@@ -3255,22 +3222,6 @@ imprecise_trapezoid_span_fallback(CARD8 op, PicturePtr src, PicturePtr dst,
return true;
}
-static inline bool
-project_trap_onto_grid(const xTrap *in,
- int dx, int dy,
- xTrap *out)
-{
- out->top.l = dx + pixman_fixed_to_fast(in->top.l);
- out->top.r = dx + pixman_fixed_to_fast(in->top.r);
- out->top.y = dy + pixman_fixed_to_fast(in->top.y);
-
- out->bot.l = dx + pixman_fixed_to_fast(in->bot.l);
- out->bot.r = dx + pixman_fixed_to_fast(in->bot.r);
- out->bot.y = dy + pixman_fixed_to_fast(in->bot.y);
-
- return out->bot.y > out->top.y;
-}
-
bool
trap_span_converter(struct sna *sna,
PicturePtr dst,
@@ -3490,33 +3441,6 @@ trap_mask_converter(struct sna *sna,
return true;
}
-static inline void
-project_point_onto_grid(const xPointFixed *in,
- int dx, int dy,
- xPointFixed *out)
-{
- out->x = dx + pixman_fixed_to_fast(in->x);
- out->y = dy + pixman_fixed_to_fast(in->y);
- __DBG(("%s: (%f, %f) -> (%d, %d)\n",
- __FUNCTION__,
- pixman_fixed_to_double(in->x),
- pixman_fixed_to_double(in->y),
- out->x, out->y));
-}
-
-#if HAS_PIXMAN_TRIANGLES
-static inline bool
-project_triangle_onto_grid(const xTriangle *in,
- int dx, int dy,
- xTriangle *out)
-{
- project_point_onto_grid(&in->p1, dx, dy, &out->p1);
- project_point_onto_grid(&in->p2, dx, dy, &out->p2);
- project_point_onto_grid(&in->p3, dx, dy, &out->p3);
-
- return xTriangleValid(out);
-}
-
bool
triangles_span_converter(struct sna *sna,
CARD8 op, PicturePtr src, PicturePtr dst,
@@ -4001,5 +3925,3 @@ skip:
REGION_UNINIT(NULL, &clip);
return true;
}
-
-#endif
More information about the xorg-commit
mailing list