Avoid an unnecessary copy of the image buffer.

By pointing Cairo at the mmap-ed file used to create wl_shm buffers, we
can eliminate a memcpy() on every draw, providing a decent speedup
(especially for large window sizes).

This comes at the expense of having to keep track of two Cairo contexts,
one for each of our two buffers used for double buffering. Additionally,
a single memcpy() is still required for initialisation of the second
buffer, so the startup latency isn't affected much.

1 files changed, 3 insertions, 8 deletions

diff --git a/src/surface.c b/src/surface.c
index d02fd4f..30a46cd 100644
--- a/src/surface.c
+++ b/src/surface.c
@@ -8,7 +8,7 @@
 #undef MAX
 #define MAX(a, b) ((a) > (b) ? (a) : (b))
 
-void surface_initialise(
+void surface_init(
 		struct surface *surface,
 		struct wl_shm *wl_shm)
 {
@@ -47,8 +47,6 @@ void surface_initialise(
 
 	log_debug("Created shm file with size %d KiB.\n",
 			surface->shm_pool_size / 1024);
-
-	surface->index = 0;
 }
 
 void surface_destroy(struct surface *surface)
@@ -66,12 +64,9 @@ void surface_draw(
 		struct color *color,
 		struct image *texture)
 {
-	surface->index = !surface->index;
-	int offset = surface->height * surface->stride * surface->index;
-	uint32_t *pixels = (uint32_t *)&surface->shm_pool_data[offset];
-	memcpy(pixels, texture->buffer, surface->height * surface->stride);
-
 	wl_surface_attach(surface->wl_surface, surface->buffers[surface->index], 0, 0);
 	wl_surface_damage_buffer(surface->wl_surface, 0, 0, INT32_MAX, INT32_MAX);
 	wl_surface_commit(surface->wl_surface);
+
+	surface->index = !surface->index;
 }