xf86-video-intel: 2 commits - src/sna/sna.h src/sna/sna_threads.c src/sna/sna_trapezoids.c
Chris Wilson
ickle at kemper.freedesktop.org
Thu Jan 24 11:19:40 PST 2013
src/sna/sna.h | 4 +
src/sna/sna_threads.c | 67 +++++++++++++---------
src/sna/sna_trapezoids.c | 139 ++++++++++++++++++++++++++++++++++++++---------
3 files changed, 159 insertions(+), 51 deletions(-)
New commits:
commit f597b647180c1e7bf83693060f244926191b7462
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Thu Jan 24 18:45:35 2013 +0000
sna: Tidy construction of data for threaded composite
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna_threads.c b/src/sna/sna_threads.c
index 0bd797e..3f70f8f 100644
--- a/src/sna/sna_threads.c
+++ b/src/sna/sna_threads.c
@@ -203,31 +203,49 @@ void sna_image_composite(pixman_op_t op,
dst_x, dst_y,
width, height);
} else {
- struct thread_composite threads[num_threads];
+ struct thread_composite data[num_threads];
int y, dy, n;
+ DBG(("%s: using %d threads for compositing %dx%d\n",
+ __FUNCTION__, num_threads, width, height));
+
y = dst_y;
dy = (height + num_threads - 1) / num_threads;
- for (n = 0; n < num_threads; n++) {
- threads[n].op = op;
- threads[n].src = src;
- threads[n].mask = mask;
- threads[n].dst = dst;
- threads[n].src_x = src_x;
- threads[n].src_y = src_y + y - dst_y;
- threads[n].mask_x = mask_x;
- threads[n].mask_y = mask_y + y - dst_y;
- threads[n].dst_x = dst_x;
- threads[n].dst_y = y;
- threads[n].width = width;
- threads[n].height = dy;
-
- sna_threads_run(thread_composite, &threads[n]);
+ data[0].op = op;
+ data[0].src = src;
+ data[0].mask = mask;
+ data[0].dst = dst;
+ data[0].src_x = src_x;
+ data[0].src_y = src_y;
+ data[0].mask_x = mask_x;
+ data[0].mask_y = mask_y;
+ data[0].dst_x = dst_x;
+ data[0].dst_y = y;
+ data[0].width = width;
+ data[0].height = dy;
+
+ for (n = 0; n < num_threads - 1; n++) {
+ data[n] = data[0];
+ data[n].src_y += y - dst_y;
+ data[n].mask_y += y - dst_y;
+ data[n].dst_y = y;
y += dy;
- if (y + dy > dst_y + height)
- dy = dst_y + height - y;
+
+ sna_threads_run(thread_composite, &data[n]);
}
+
+ if (y + dy > dst_y + height)
+ dy = dst_y + height - y;
+
+ data[n] = data[0];
+ data[n].src_y += y - dst_y;
+ data[n].mask_y += y - dst_y;
+ data[n].dst_y = y;
+ data[n].height = dy;
+
+ sna_threads_run(thread_composite, &data[n]);
+
sna_threads_wait();
}
}
commit 1643c97f8f7b49738b649b5f7d1e574d689d167e
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date: Thu Jan 24 18:24:02 2013 +0000
sna: Use threads for simple mask generation
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
diff --git a/src/sna/sna.h b/src/sna/sna.h
index 389884f..5832c99 100644
--- a/src/sna/sna.h
+++ b/src/sna/sna.h
@@ -845,6 +845,10 @@ inline static bool is_clipped(const RegionRec *r,
}
void sna_threads_init(void);
+int sna_use_threads (int width, int height, int threshold);
+void sna_threads_run(void (*func)(void *arg), void *arg);
+void sna_threads_wait(void);
+
void sna_image_composite(pixman_op_t op,
pixman_image_t *src,
pixman_image_t *mask,
diff --git a/src/sna/sna_threads.c b/src/sna/sna_threads.c
index afa260f..0bd797e 100644
--- a/src/sna/sna_threads.c
+++ b/src/sna/sna_threads.c
@@ -97,8 +97,7 @@ bail:
max_threads = 0;
}
-static void
-threads_run(void (*func)(void *arg), void *arg)
+void sna_threads_run(void (*func)(void *arg), void *arg)
{
int n;
@@ -129,7 +128,7 @@ execute:
pthread_mutex_unlock(&threads[n].mutex);
}
-static void threads_wait(void)
+void sna_threads_wait(void)
{
int n;
@@ -146,8 +145,7 @@ static void threads_wait(void)
}
}
-static int
-use_threads (int width, int height, int threshold)
+int sna_use_threads(int width, int height, int threshold)
{
int num_threads;
@@ -197,14 +195,13 @@ void sna_image_composite(pixman_op_t op,
{
int num_threads;
- num_threads = use_threads(width, height, 16);
+ num_threads = sna_use_threads(width, height, 16);
if (num_threads <= 1) {
pixman_image_composite(op, src, mask, dst,
src_x, src_y,
mask_x, mask_y,
dst_x, dst_y,
width, height);
- return;
} else {
struct thread_composite threads[num_threads];
int y, dy, n;
@@ -225,12 +222,12 @@ void sna_image_composite(pixman_op_t op,
threads[n].width = width;
threads[n].height = dy;
- threads_run(thread_composite, &threads[n]);
+ sna_threads_run(thread_composite, &threads[n]);
y += dy;
if (y + dy > dst_y + height)
dy = dst_y + height - y;
}
- threads_wait();
+ sna_threads_wait();
}
}
diff --git a/src/sna/sna_trapezoids.c b/src/sna/sna_trapezoids.c
index 0b81018..f7a146c 100644
--- a/src/sna/sna_trapezoids.c
+++ b/src/sna/sna_trapezoids.c
@@ -5277,13 +5277,53 @@ trapezoid_span_inplace__x8r8g8b8(CARD8 op,
return true;
}
+struct inplace_thread {
+ xTrapezoid *traps;
+ RegionPtr clip;
+ span_func_t span;
+ struct inplace inplace;
+ BoxRec extents;
+ int dx, dy;
+ int draw_x, draw_y;
+ bool unbounded;
+ int ntrap;
+};
+
+static void inplace_thread(void *arg)
+{
+ struct inplace_thread *thread = arg;
+ struct tor tor;
+ int n;
+
+ if (tor_init(&tor, &thread->extents, 2*thread->ntrap))
+ return;
+
+ for (n = 0; n < thread->ntrap; n++) {
+ xTrapezoid t;
+
+ if (!project_trapezoid_onto_grid(&thread->traps[n], thread->dx, thread->dy, &t))
+ continue;
+
+ if (pixman_fixed_to_int(thread->traps[n].top) >= thread->extents.y2 - thread->draw_y ||
+ pixman_fixed_to_int(thread->traps[n].bottom) < thread->extents.y1 - thread->draw_y)
+ continue;
+
+ tor_add_edge(&tor, &t, &t.left, 1);
+ tor_add_edge(&tor, &t, &t.right, -1);
+ }
+
+ tor_render(NULL, &tor, (void*)&thread->inplace,
+ thread->clip, thread->span, thread->unbounded);
+
+ tor_fini(&tor);
+}
+
static bool
trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
PictFormatPtr maskFormat, INT16 src_x, INT16 src_y,
int ntrap, xTrapezoid *traps,
bool fallback)
{
- struct tor tor;
struct inplace inplace;
span_func_t span;
PixmapPtr pixmap;
@@ -5293,7 +5333,7 @@ trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
bool unbounded;
int16_t dst_x, dst_y;
int dx, dy;
- int n;
+ int num_threads, n;
if (NO_SCAN_CONVERTER)
return false;
@@ -5424,26 +5464,6 @@ trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
region.extents.x1, region.extents.y1,
region.extents.x2, region.extents.y2));
- if (tor_init(&tor, ®ion.extents, 2*ntrap))
- return true;
-
- dx = dst->pDrawable->x * FAST_SAMPLES_X;
- dy = dst->pDrawable->y * FAST_SAMPLES_Y;
-
- for (n = 0; n < ntrap; n++) {
- xTrapezoid t;
-
- if (!project_trapezoid_onto_grid(&traps[n], dx, dy, &t))
- continue;
-
- if (pixman_fixed_to_int(traps[n].top) >= region.extents.y2 - dst->pDrawable->y ||
- pixman_fixed_to_int(traps[n].bottom) < region.extents.y1 - dst->pDrawable->y)
- continue;
-
- tor_add_edge(&tor, &t, &t.left, 1);
- tor_add_edge(&tor, &t, &t.right, -1);
- }
-
if (op == PictOpSrc) {
if (dst->pCompositeClip->data)
span = tor_blt_src_clipped;
@@ -5468,6 +5488,9 @@ trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
op == PictOpSrc ? MOVE_WRITE | MOVE_INPLACE_HINT : MOVE_WRITE | MOVE_READ))
return true;
+ dx = dst->pDrawable->x * FAST_SAMPLES_X;
+ dy = dst->pDrawable->y * FAST_SAMPLES_Y;
+
get_drawable_deltas(dst->pDrawable, pixmap, &dst_x, &dst_y);
inplace.ptr = pixmap->devPrivate.ptr;
@@ -5475,10 +5498,76 @@ trapezoid_span_inplace(CARD8 op, PicturePtr src, PicturePtr dst,
inplace.stride = pixmap->devKind;
inplace.opacity = color >> 24;
- tor_render(NULL, &tor, (void*)&inplace,
- dst->pCompositeClip, span, unbounded);
+ num_threads = sna_use_threads(region.extents.x2 - region.extents.x1,
+ region.extents.y2 - region.extents.y1,
+ 8);
+ if (num_threads == 1) {
+ struct tor tor;
- tor_fini(&tor);
+ if (tor_init(&tor, ®ion.extents, 2*ntrap))
+ return true;
+
+ for (n = 0; n < ntrap; n++) {
+ xTrapezoid t;
+
+ if (!project_trapezoid_onto_grid(&traps[n], dx, dy, &t))
+ continue;
+
+ if (pixman_fixed_to_int(traps[n].top) >= region.extents.y2 - dst->pDrawable->y ||
+ pixman_fixed_to_int(traps[n].bottom) < region.extents.y1 - dst->pDrawable->y)
+ continue;
+
+ tor_add_edge(&tor, &t, &t.left, 1);
+ tor_add_edge(&tor, &t, &t.right, -1);
+ }
+
+ tor_render(NULL, &tor, (void*)&inplace,
+ dst->pCompositeClip, span, unbounded);
+
+ tor_fini(&tor);
+ } else {
+ struct inplace_thread threads[num_threads];
+ int y, h;
+
+ DBG(("%s: using %d threads for inplace compositing %dx%d\n",
+ __FUNCTION__, num_threads,
+ region.extents.x2 - region.extents.x1,
+ region.extents.y2 - region.extents.y1));
+
+ threads[0].traps = traps;
+ threads[0].ntrap = ntrap;
+ threads[0].inplace = inplace;
+ threads[0].extents = region.extents;
+ threads[0].clip = dst->pCompositeClip;
+ threads[0].span = span;
+ threads[0].unbounded = unbounded;
+ threads[0].dx = dx;
+ threads[0].dy = dy;
+ threads[0].draw_x = dst->pDrawable->x;
+ threads[0].draw_y = dst->pDrawable->y;
+
+ y = region.extents.y1;
+ h = region.extents.y2 - region.extents.y1;
+ h = (h + num_threads - 1) / num_threads;
+
+ for (n = 0; n < num_threads - 1; n++) {
+ threads[n] = threads[0];
+ threads[n].extents.y1 = y;
+ threads[n].extents.y2 = y += h;
+
+ sna_threads_run(inplace_thread, &threads[n]);
+ }
+
+ if (y + h > region.extents.y2)
+ h = region.extents.y2 - y;
+
+ threads[n] = threads[0];
+ threads[n].extents.y1 = y;
+ threads[n].extents.y2 = y + h;
+ inplace_thread(&threads[n]);
+
+ sna_threads_wait();
+ }
return true;
}
More information about the xorg-commit
mailing list