|
2 | 2 |
|
3 | 3 | // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
|
4 | 4 |
|
5 |
| -// Check that dynamic batching increases batch size |
6 |
| -// RUN: env SYCL_PI_TRACE=2 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CKALL,CKDYNUP %s |
| 5 | +// Check that dynamic batching raises/lowers batch size |
| 6 | +// RUN: env SYCL_PI_TRACE=2 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CKALL,CKDYN %s |
7 | 7 |
|
8 | 8 | // level_zero_dynamic_batch_test.cpp
|
9 | 9 | //
|
10 |
| -// This tests the level zero plugin's kernel dyanmic batch size adjustment |
| 10 | +// This tests the level zero plugin's kernel dynamic batch size adjustment |
11 | 11 | // code.
|
12 | 12 | // It starts out by enqueing 40 kernels before it does a wait, and it does
|
13 | 13 | // this 5 times. That should cause the dynamic batch size adjustment to
|
14 |
| -// raise the batch size up several times. |
| 14 | +// raise the batch size up 3 times. |
15 | 15 | //
|
16 | 16 | // Then the test starts enqueueing only 4 kernels before doing a wait, and
|
17 |
| -// it does that 5 times as well. That should cause the batch size to |
18 |
| -// be lowered, just once to be less than 4. |
| 17 | +// it does that 20 times. That should cause the batch size to |
| 18 | +// be lowered to be less than 4. |
19 | 19 | //
|
20 | 20 | // CKDYN: Raising QueueBatchSize to 5
|
21 | 21 | // CKDYN: Raising QueueBatchSize to 6
|
| 22 | +// CKDYN: Raising QueueBatchSize to 7 |
22 | 23 | // CKDYN-NOT: Raising QueueBatchSize
|
23 | 24 | // CKALL: Test Pass
|
24 | 25 | // CKALL: Test Pass
|
@@ -165,7 +166,7 @@ int main(int argc, char *argv[]) {
|
165 | 166 | validate(Y1, Z7, M * N);
|
166 | 167 | validate(Y1, Z8, M * N);
|
167 | 168 |
|
168 |
| - for (size_t i = 0; i < 5; i++) { |
| 169 | + for (size_t i = 0; i < 20; i++) { |
169 | 170 | q.submit([&](sycl::handler &h) {
|
170 | 171 | h.parallel_for<class u32_copy9>(sycl::range<2>{M, N},
|
171 | 172 | [=](sycl::id<2> it) {
|
|
0 commit comments