Submit a chain of spinners across all the engines, using the submit
fence to launch them in parallel.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
lib/igt_dummyload.c | 7 +++++-
lib/igt_dummyload.h | 15 ++++++------
tests/i915/gem_exec_fence.c | 46 +++++++++++++++++++++++++++++++++++++
3 files changed, 60 insertions(+), 8 deletions(-)
diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
index 891e4eaab..4ef79cc4a 100644
--- a/lib/igt_dummyload.c
+++ b/lib/igt_dummyload.c
@@ -326,11 +326,16 @@ emit_recursive_batch(igt_spin_t *spin,
if (opts->flags & IGT_SPIN_FENCE_OUT)
execbuf->flags |= I915_EXEC_FENCE_OUT;
- if (opts->flags & IGT_SPIN_FENCE_IN) {
+ if (opts->flags & IGT_SPIN_FENCE_IN && opts->fence != -1) {
execbuf->flags |= I915_EXEC_FENCE_IN;
execbuf->rsvd2 = opts->fence;
}
+ if (opts->flags & IGT_SPIN_FENCE_SUBMIT && opts->fence != -1) {
+ execbuf->flags |= I915_EXEC_FENCE_SUBMIT;
+ execbuf->rsvd2 = opts->fence;
+ }
+
for (i = 0; i < nengine; i++) {
execbuf->flags &= ~ENGINE_MASK;
execbuf->flags |= flags[i];
diff --git a/lib/igt_dummyload.h b/lib/igt_dummyload.h
index b8baaa6b4..3ece70a50 100644
--- a/lib/igt_dummyload.h
+++ b/lib/igt_dummyload.h
@@ -65,13 +65,14 @@ struct igt_spin_factory {
};
#define IGT_SPIN_FENCE_IN (1 << 0)
-#define IGT_SPIN_FENCE_OUT (1 << 1)
-#define IGT_SPIN_POLL_RUN (1 << 2)
-#define IGT_SPIN_FAST (1 << 3)
-#define IGT_SPIN_NO_PREEMPTION (1 << 4)
-#define IGT_SPIN_INVALID_CS (1 << 5)
-#define IGT_SPIN_USERPTR (1 << 6)
-#define IGT_SPIN_SOFTDEP (1 << 7)
+#define IGT_SPIN_FENCE_SUBMIT (1 << 1)
+#define IGT_SPIN_FENCE_OUT (1 << 2)
+#define IGT_SPIN_POLL_RUN (1 << 3)
+#define IGT_SPIN_FAST (1 << 4)
+#define IGT_SPIN_NO_PREEMPTION (1 << 5)
+#define IGT_SPIN_INVALID_CS (1 << 6)
+#define IGT_SPIN_USERPTR (1 << 7)
+#define IGT_SPIN_SOFTDEP (1 << 8)
igt_spin_t *
__igt_spin_factory(int fd, const struct igt_spin_factory *opts);
diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 6eea050d8..ba454fcdb 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -796,6 +796,47 @@ static void test_concurrent(int i915, const struct intel_execution_engine2 *e)
igt_spin_free(i915, spin);
}
+static void test_submit_chain(int i915)
+{
+ const struct intel_execution_engine2 *e;
+ igt_spin_t *spin, *sn;
+ IGT_LIST_HEAD(list);
+ IGT_CORK_FENCE(cork);
+ int fence;
+
+ /* Check that we can simultaneously launch spinners on each engine */
+
+ fence = igt_cork_plug(&cork, i915);
+ __for_each_physical_engine(i915, e) {
+ spin = igt_spin_new(i915,
+ .engine = e->flags,
+ .fence = fence,
+ .flags = (IGT_SPIN_POLL_RUN |
+ IGT_SPIN_FENCE_OUT |
+ IGT_SPIN_FENCE_SUBMIT));
+
+ fence = spin->out_fence;
+ igt_list_move(&spin->link, &list);
+ }
+
+ /* Nothing shall run until we pop the cork */
+ igt_list_for_each_entry(spin, &list, link) {
+ igt_assert(gem_bo_busy(i915, spin->handle));
+ igt_assert(!igt_spin_has_started(spin));
+ }
+
+ igt_cork_unplug(&cork);
+
+ /* Then everything shall run in parallel */
+ igt_list_for_each_entry_safe(spin, sn, &list, link) {
+ igt_spin_busywait_until_started(spin);
+ igt_spin_end(spin);
+ igt_assert_eq(sync_fence_wait(spin->out_fence, 50), 0);
+ igt_assert_eq(sync_fence_status(spin->out_fence), 1);
+ igt_spin_free(i915, spin);
+ }
+}
+
static uint32_t batch_create(int fd)
{
const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -3032,6 +3073,11 @@ igt_main
}
}
+ igt_subtest("submit-chain") {
+ igt_require(has_submit_fence(i915));
+ test_submit_chain(i915);
+ }
+
igt_fixture {
igt_stop_hang_detector();
}
--
2.29.2
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
