TESTING only: enforce larger time span for kernel execution #535
Conversation
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
from
23b5d62 to
db20399
Compare
src/stream/TRIAD-Hip.cpp
Outdated
| }); | ||
|
|
||
| for (RepIndex_type extra_rep = 0; extra_rep < 5; ++extra_rep) { | ||
| temp_count += extra_rep * (extra_rep & 1); |
There was a problem hiding this comment.
We can try adding EMPTY_BODY (https://github.com/LLNL/RAJAPerf/blob/cda42470851fff2b7c8e6a9b5b11ab83f33a5a07/src/basic/EMPTY.hpp#L29) if the compiler optimizes out the loop.
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
from
db20399 to
822ce7f
Compare
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
3 times, most recently
from
17f3f8f to
6cbc640
Compare
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
from
6cbc640 to
04510eb
Compare
src/basic/MAT_MAT_SHARED.hpp
Outdated
| if (Row < N && Col < N) \ | ||
| C[Col + N * Row] = Cs[ty][tx]; | ||
|
|
||
| constexpr int extra_kernel_reps = 5; |
There was a problem hiding this comment.
You'll want to make a local variable of this and put it into the DATA_SETUP macro
There was a problem hiding this comment.
Its probably best to add this as a member variable of KernelBase so the compiler doesn't know how many extra_kernel_reps at compile time.
There was a problem hiding this comment.
Sure, for now I added as you advised in the first comment, and will also try to do what you advised in this comment. Thanks
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
from
b65aef2 to
001d11e
Compare
|
I expect this approach to cause the tests to fail, so don't worry too much about getting the tests to pass. |
amroakmal
force-pushed
the
temp-add-outer-loop-for-kernel
branch
from
001d11e to
dd132de
Compare
Summary