summaryrefslogtreecommitdiff
path: root/src/modules/snow
diff options
context:
space:
mode:
authorVito Caputo <vcaputo@pengaru.com>2019-11-24 21:39:05 -0800
committerVito Caputo <vcaputo@pengaru.com>2019-11-24 21:59:39 -0800
commit7857405e735fcf4542f9022d8f9e00c176f86597 (patch)
treef2de6a16e9a095419fee6fc17c755cdffbd85fc7 /src/modules/snow
parent4d87a91891a2ff29f399edebfcd3590316a14445 (diff)
snow: per-cpu rand_r() seed state
Snow was already threaded, but used a global seed with rand_r() meaning the CPUs were hammering on the same address. There wasn't any locking or atomics, as it isn't terribly critical when generating white noise if the seed access is racy. But the writes still caused cache lines to ping-pong. This commit gives a ~15.5% speedup in my measurements on an i7-2640M. Note without the padded union, using just an array of ints, zero gain is realized. I used a padding of 256 just to have some headroom, x86 is 64 but other CPUs vary, POWER9 is 128 for example.
Diffstat (limited to 'src/modules/snow')
-rw-r--r--src/modules/snow/snow.c47
1 files changed, 38 insertions, 9 deletions
diff --git a/src/modules/snow/snow.c b/src/modules/snow/snow.c
index 8968e47..3e6331a 100644
--- a/src/modules/snow/snow.c
+++ b/src/modules/snow/snow.c
@@ -9,13 +9,37 @@
/* This implements white noise / snow just using rand() */
-/* To avoid the contention around rand() I'm just using rand_r()
- * in an entirely racy fashion with a single seed from the threaded
- * render_fragment(). It should really be per-cpu but the module
- * api doesn't currently send down a cpu identifier to the render
- * function, so TODO in the future add that.
- */
-static int snow_seed;
+typedef union snow_seed_t {
+ char __padding[256]; /* prevent seeds sharing a cache-line */
+ int seed;
+} snow_seed_t;
+
+typedef struct snow_context_t {
+ int foo; /* make the compiler happy */
+ snow_seed_t seeds[];
+} snow_context_t;
+
+
+static void * snow_create_context(unsigned n_cpus)
+{
+ snow_context_t *ctxt;
+
+ ctxt = calloc(1, sizeof(snow_context_t) + n_cpus * sizeof(snow_seed_t));
+ if (!ctxt)
+ return NULL;
+
+ for (unsigned i = 0; i < n_cpus; i++)
+ ctxt->seeds[i].seed = rand();
+
+ return ctxt;
+}
+
+
+static void snow_destroy_context(void *context)
+{
+ free(context);
+}
+
static int snow_fragmenter(void *context, const fb_fragment_t *fragment, unsigned number, fb_fragment_t *res_fragment)
{
@@ -31,9 +55,12 @@ static void snow_prepare_frame(void *context, unsigned n_cpus, fb_fragment_t *fr
static void snow_render_fragment(void *context, unsigned cpu, fb_fragment_t *fragment)
{
+ snow_context_t *ctxt = context;
+ int *seed = &ctxt->seeds[cpu].seed;
+
for (unsigned y = fragment->y; y < fragment->y + fragment->height; y++) {
for (unsigned x = fragment->x; x < fragment->x + fragment->width; x++) {
- uint32_t pixel = rand_r(&snow_seed) % 256;
+ uint32_t pixel = rand_r(seed) % 256;
fb_fragment_put_pixel_unchecked(fragment, x, y, pixel << 16 | pixel << 8 | pixel);
}
@@ -42,10 +69,12 @@ static void snow_render_fragment(void *context, unsigned cpu, fb_fragment_t *fra
rototiller_module_t snow_module = {
+ .create_context = snow_create_context,
+ .destroy_context = snow_destroy_context,
.prepare_frame = snow_prepare_frame,
.render_fragment = snow_render_fragment,
.name = "snow",
- .description = "TV snow / white noise",
+ .description = "TV snow / white noise (threaded)",
.author = "Vito Caputo <vcaputo@pengaru.com>",
.license = "GPLv2",
};
© All Rights Reserved