sna: Experiment with a threaded renderer for fallback compositing

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

src/sna/Makefile.am

@@ -34,6 +34,7 @@ AM_CFLAGS += @VALGRIND_CFLAGS@
 endif
 
 noinst_LTLIBRARIES = libsna.la
+libsna_la_LDFLAGS = -pthread
 libsna_la_LIBADD = @UDEV_LIBS@ -lm @DRM_LIBS@ brw/libbrw.la fb/libfb.la
 
 libsna_la_SOURCES = \
@@ -62,6 +63,7 @@ libsna_la_SOURCES = \
 	sna_trapezoids.c \
 	sna_tiling.c \
 	sna_transform.c \
+	sna_threads.c \
 	sna_video.c \
 	sna_video.h \
 	sna_video_overlay.c \

src/sna/sna.h

@@ -835,4 +835,18 @@ inline static bool is_clipped(const RegionRec *r,
 		r->extents.y2 - r->extents.y1 != d->height);
 }
 
+void sna_threads_init(void);
+void sna_image_composite(pixman_op_t        op,
+			 pixman_image_t    *src,
+			 pixman_image_t    *mask,
+			 pixman_image_t    *dst,
+			 int16_t            src_x,
+			 int16_t            src_y,
+			 int16_t            mask_x,
+			 int16_t            mask_y,
+			 int16_t            dst_x,
+			 int16_t            dst_y,
+			 uint16_t           width,
+			 uint16_t           height);
+
 #endif /* _SNA_H */

src/sna/sna_driver.c

@@ -1174,5 +1174,7 @@ Bool sna_init_scrn(ScrnInfoPtr scrn, int entity_num)
 	xf86SetEntityInstanceForScreen(scrn, entity_num,
 				       xf86GetNumEntityInstances(entity_num)-1);
 
+	sna_threads_init();
+
 	return TRUE;
 }

src/sna/sna_render.c

@@ -1465,11 +1465,11 @@ sna_render_picture_approximate_gradient(struct sna *sna,
 		pixman_transform_multiply(&t, picture->transform, &t);
 	pixman_image_set_transform(src, &t);
 
-	pixman_image_composite(PictOpSrc, src, NULL, dst,
-			       x + dx, y + dy,
-			       0, 0,
-			       0, 0,
-			       w2, h2);
+	sna_image_composite(PictOpSrc, src, NULL, dst,
+			    x+dx, y+dy,
+			    0, 0,
+			    0, 0,
+			    w2, h2);
 	free_pixman_pict(picture, src);
 	pixman_image_unref(dst);
 
@@ -1580,11 +1580,11 @@ do_fixup:
 
 	DBG(("%s: compositing tmp=(%d+%d, %d+%d)x(%d, %d)\n",
 	     __FUNCTION__, x, dx, y, dy, w, h));
-	pixman_image_composite(PictOpSrc, src, NULL, dst,
-			       x + dx, y + dy,
-			       0, 0,
-			       0, 0,
-			       w, h);
+	sna_image_composite(PictOpSrc, src, NULL, dst,
+			    x + dx, y + dy,
+			    0, 0,
+			    0, 0,
+			    w, h);
 	free_pixman_pict(picture, src);
 
 	/* Then convert to card format */

src/sna/sna_threads.c

@@ -0,0 +1,236 @@
+/*
+ * Copyright © 2013 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Authors:
+ *    Chris Wilson <chris@chris-wilson.co.uk>
+ *
+ */
+
+#include "sna.h"
+
+#include <unistd.h>
+#include <pthread.h>
+
+static int max_threads = -1;
+
+static struct thread {
+    pthread_t thread;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+
+    void (*func)(void *arg);
+    void *arg;
+} *threads;
+
+static void *__run__(void *arg)
+{
+	struct thread *t = arg;
+
+	pthread_mutex_lock(&t->mutex);
+	while (1) {
+		while (t->func == NULL)
+			pthread_cond_wait(&t->cond, &t->mutex);
+		pthread_mutex_unlock(&t->mutex);
+
+		assert(t->func);
+		t->func(t->arg);
+
+		pthread_mutex_lock(&t->mutex);
+		t->func = NULL;
+		pthread_cond_signal(&t->cond);
+	}
+	pthread_mutex_unlock(&t->mutex);
+
+	return NULL;
+}
+
+void sna_threads_init(void)
+{
+	int n;
+
+	if (max_threads != -1)
+		return;
+
+	max_threads = sysconf (_SC_NPROCESSORS_ONLN) / 2;
+	if (max_threads <= 1)
+		goto bail;
+
+	DBG(("%s: creating a thread pool of %d threads\n",
+	     __func__, max_threads));
+
+	threads = malloc (sizeof(threads[0])*max_threads);
+	if (threads == NULL)
+		goto bail;
+
+	for (n = 0; n < max_threads; n++) {
+		pthread_mutex_init(&threads[n].mutex, NULL);
+		pthread_cond_init(&threads[n].cond, NULL);
+
+		threads[n].func = NULL;
+		if (pthread_create(&threads[n].thread, NULL,
+				   __run__, &threads[n]))
+			goto bail;
+	}
+
+	return;
+
+bail:
+	max_threads = 0;
+}
+
+static void
+threads_run(void (*func)(void *arg), void *arg)
+{
+	int n;
+
+	assert(max_threads > 0);
+
+	for (n = 0; n < max_threads; n++) {
+		if (threads[n].func)
+			continue;
+
+		pthread_mutex_lock(&threads[n].mutex);
+		if (threads[n].func) {
+			pthread_mutex_unlock(&threads[n].mutex);
+			continue;
+		}
+
+		goto execute;
+	}
+
+	n = rand() % max_threads;
+	pthread_mutex_lock(&threads[n].mutex);
+	while (threads[n].func)
+		pthread_cond_wait(&threads[n].cond, &threads[n].mutex);
+
+execute:
+	threads[n].func = func;
+	threads[n].arg = arg;
+	pthread_cond_signal(&threads[n].cond);
+	pthread_mutex_unlock(&threads[n].mutex);
+}
+
+static void threads_wait(void)
+{
+	int n;
+
+	assert(max_threads > 0);
+
+	for (n = 0; n < max_threads; n++) {
+		if (threads[n].func == NULL)
+			continue;
+
+		pthread_mutex_lock(&threads[n].mutex);
+		while (threads[n].func)
+			pthread_cond_wait(&threads[n].cond, &threads[n].mutex);
+		pthread_mutex_unlock(&threads[n].mutex);
+	}
+}
+
+static int
+use_threads (int width, int height, int threshold)
+{
+	int num_threads;
+
+	if (max_threads <= 0)
+		return 1;
+
+	num_threads = height / (128/width + 1) / threshold-1;
+	if (num_threads <= 0)
+		return 1;
+
+	if (num_threads > max_threads)
+		num_threads = max_threads;
+	return num_threads;
+}
+
+struct thread_composite {
+	pixman_image_t *src, *mask, *dst;
+	pixman_op_t op;
+	int16_t src_x, src_y;
+	int16_t mask_x, mask_y;
+	int16_t dst_x, dst_y;
+	uint16_t width, height;
+};
+
+static void thread_composite(void *arg)
+{
+	struct thread_composite *t = arg;
+	pixman_image_composite(t->op, t->src, t->mask, t->dst,
+			       t->src_x, t->src_y,
+			       t->mask_x, t->mask_y,
+			       t->dst_x, t->dst_y,
+			       t->width, t->height);
+}
+
+void sna_image_composite(pixman_op_t        op,
+			 pixman_image_t    *src,
+			 pixman_image_t    *mask,
+			 pixman_image_t    *dst,
+			 int16_t            src_x,
+			 int16_t            src_y,
+			 int16_t            mask_x,
+			 int16_t            mask_y,
+			 int16_t            dst_x,
+			 int16_t            dst_y,
+			 uint16_t           width,
+			 uint16_t           height)
+{
+	int num_threads;
+
+	num_threads = 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;
+
+		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;
+
+			threads_run(thread_composite, &threads[n]);
+
+			y += dy;
+			if (y + dy > dst_y + height)
+				dy = dst_y + height - y;
+		}
+		threads_wait();
+	}
+}