[i-g-t 1/1] tests/intel/xe_exec_threads: wait for all submissions to complete

[fei.yang at intel.com]

From: Fei Yang <fei.yang at intel.com>

In test_compute_mode, there is an one second sleep waiting for all
the submissions to complete, but that is not reliable especially
on pre-si platforms where the GPU could be a lot slower. Instead we
should wait for the ufence to make sure the GPU is inactive before
unbinding the BO.

Signed-off-by: Fei Yang <fei.yang at intel.com>
---
 tests/intel/xe_exec_threads.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/tests/intel/xe_exec_threads.c b/tests/intel/xe_exec_threads.c
index 413d6626b..b73d3df2b 100644
--- a/tests/intel/xe_exec_threads.c
+++ b/tests/intel/xe_exec_threads.c
@@ -340,7 +340,7 @@ test_compute_mode(int fd, uint32_t vm, uint64_t addr, uint64_t userptr,
 		xe_exec(fd, &exec);
 
 		if (flags & REBIND && i && !(i & 0x1f)) {
-			for (j = i - 0x20; j <= i; ++j)
+			for (j = i - 0x20; j < i; ++j)
 				xe_wait_ufence(fd, &data[j].exec_sync,
 					       USER_FENCE_VALUE,
 					       exec_queues[e], fence_timeout);
@@ -370,7 +370,7 @@ test_compute_mode(int fd, uint32_t vm, uint64_t addr, uint64_t userptr,
 				 * physical memory on next mmap call triggering
 				 * an invalidate.
 				 */
-				for (j = i == 0x20 ? 0 : i - 0x1f; j <= i; ++j)
+				for (j = i - 0x20; j < i; ++j)
 					xe_wait_ufence(fd, &data[j].exec_sync,
 						       USER_FENCE_VALUE,
 						       exec_queues[e],
@@ -404,16 +404,26 @@ test_compute_mode(int fd, uint32_t vm, uint64_t addr, uint64_t userptr,
 		}
 	}
 
-	j = flags & INVALIDATE ?
-		(flags & RACE ? n_execs / 2 + 1 : n_execs - 1) : 0;
+	j = 0; /* wait for all submissions to complete */
+	if (flags & INVALIDATE)
+		/*
+		 * For !RACE cases xe_wait_ufence has been called in above for-loop
+		 * except the last batch of submissions (0x20 submissions in a batch).
+		 * For RACE cases we will need to wait for all the submissions to
+		 * complete here.
+		 */
+		j = (flags & RACE) ? 0 : (n_execs - 0x20);
+	else if (flags & REBIND)
+		/*
+		 * For REBIND cases xe_wait_ufence has been called in above for-loop
+		 * except the last batch of submissions (0x20 submissions in a batch).
+		 */
+		j = n_execs - 0x20;
+
 	for (i = j; i < n_execs; i++)
 		xe_wait_ufence(fd, &data[i].exec_sync, USER_FENCE_VALUE,
 			       exec_queues[i % n_exec_queues], fence_timeout);
 
-	/* Wait for all execs to complete */
-	if (flags & INVALIDATE)
-		sleep(1);
-
 	sync[0].addr = to_user_pointer(&data[0].vm_sync);
 	xe_vm_unbind_async(fd, vm, 0, 0, addr, bo_size, sync, 1);
 	xe_wait_ufence(fd, &data[0].vm_sync, USER_FENCE_VALUE, 0, fence_timeout);
-- 
2.25.1