[Intel-gfx] [PATCH i-g-t] i915/gem_exec_parallel: Try to trim runtime
Chris Wilson
chris at chris-wilson.co.uk
Tue Mar 3 14:03:27 UTC 2020
In all likelihood the runtime is consumed by the thread setup, but just
in case it is dominated by the execbuf, make sure that is bounded.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
tests/i915/gem_exec_parallel.c | 28 ++++++++++++++++------------
1 file changed, 16 insertions(+), 12 deletions(-)
diff --git a/tests/i915/gem_exec_parallel.c b/tests/i915/gem_exec_parallel.c
index 0d4d6c628..586f85080 100644
--- a/tests/i915/gem_exec_parallel.c
+++ b/tests/i915/gem_exec_parallel.c
@@ -42,22 +42,26 @@
#define VERIFY 0
+static inline uint32_t hash32(uint32_t val)
+{
+#define GOLDEN_RATIO_32 0x61C88647
+ return val * GOLDEN_RATIO_32;
+}
+
static void check_bo(int fd, uint32_t handle, int pass)
{
- uint32_t *map;
- int i;
+ uint32_t x = hash32(handle * pass) % 1024;
igt_debug("Verifying result (pass=%d, handle=%d)\n", pass, handle);
- map = gem_mmap__cpu(fd, handle, 0, 4096, PROT_READ);
- gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, 0);
- for (i = 0; i < 1024; i++)
- igt_assert_eq(map[i], i);
- munmap(map, 4096);
+ gem_read(fd, handle, x * sizeof(x), &x, sizeof(x));
+ igt_assert_eq_u32(x, x);
}
#define CONTEXTS 0x1
#define FDS 0x2
+#define NUMOBJ 16
+
struct thread {
pthread_t thread;
pthread_mutex_t *mutex;
@@ -132,8 +136,8 @@ static void *thread(void *data)
execbuf.rsvd1 = gem_context_clone_with_engines(fd, 0);
}
- for (i = 0; i < 16; i++) {
- obj[0].handle = t->scratch[i];
+ igt_until_timeout(1) {
+ obj[0].handle = t->scratch[i++ % NUMOBJ];
if (t->flags & FDS)
obj[0].handle = gem_open(fd, obj[0].handle);
@@ -160,7 +164,7 @@ static void all(int fd, struct intel_execution_engine2 *engine, unsigned flags)
struct thread *threads;
unsigned engines[16];
unsigned nengine;
- uint32_t scratch[16], handle[16];
+ uint32_t scratch[NUMOBJ], handle[NUMOBJ];
int go;
int i;
@@ -185,7 +189,7 @@ static void all(int fd, struct intel_execution_engine2 *engine, unsigned flags)
}
igt_require(nengine);
- for (i = 0; i < 16; i++) {
+ for (i = 0; i < NUMOBJ; i++) {
scratch[i] = handle[i] = gem_create(fd, 4096);
if (flags & FDS)
scratch[i] = gem_flink(fd, handle[i]);
@@ -221,7 +225,7 @@ static void all(int fd, struct intel_execution_engine2 *engine, unsigned flags)
for (i = 0; i < 1024; i++)
pthread_join(threads[i].thread, NULL);
- for (i = 0; i < 16; i++) {
+ for (i = 0; i < NUMOBJ; i++) {
check_bo(fd, handle[i], i);
gem_close(fd, handle[i]);
}
--
2.25.1
More information about the Intel-gfx
mailing list