Conversation
fixes llvm#706 With the following PRs now merged - llvm/llvm-project#172384 - llvm/llvm-project#175809 - llvm/llvm-project#175245 - llvm/llvm-project#175633 We have a working base to test booleans on both DXC and Clang.
farzonl
force-pushed
the
mat_bool_operator_tests
branch
from
30db7a7 to
adb0358
Compare
bob80905
left a comment
bob80905
left a comment
There was a problem hiding this comment.
Looks mostly good. You might want to document your reasoning for why you chose not to add a thread test for the or operator.
| Data: [ 1, 1, 1, 1, 1, 1 ] | ||
| - Name: AndTrueFalseOut | ||
| Format: Int32 | ||
| FillSize: 24 |
There was a problem hiding this comment.
You might consider setting this data to be non-zero. A nit, but it ensures that the buffer is actually being set to 0, and that the program is actually running.
There was a problem hiding this comment.
I agree that we should ensure the buffers are not zero-initialized
There was a problem hiding this comment.
Random thought, should we have a feature where we populate output buffers with random data to catch issues going under the radar due to data being zero-initialized?
inbelic
left a comment
inbelic
left a comment
There was a problem hiding this comment.
LGTM once Josh's comments are addressed.
The tests do seem to overlap a lot, could we not have the same test coverage with a single 2x2 matrix that has a TT, TF, FT, FF combination?
| Data: [ 1, 1, 1, 1, 1, 1 ] | ||
| - Name: AndTrueFalseOut | ||
| Format: Int32 | ||
| FillSize: 24 |
There was a problem hiding this comment.
I agree that we should ensure the buffers are not zero-initialized
| # Bug https://github.com/llvm/offload-test-suite/issues/704 | ||
| # XFAIL: Clang && Intel && Vulkan | ||
|
|
||
| Bug https://github.com/llvm/offload-test-suite/issues/705 |
There was a problem hiding this comment.
| Bug https://github.com/llvm/offload-test-suite/issues/705 | |
| # Bug https://github.com/llvm/offload-test-suite/issues/705 |
| # Note: Metal does not support boolean matrix types. For | ||
| # Vulkan KosmicKrisp and MoltenVK the output is always False. | ||
| # UNSUPPORTED: Darwin |
There was a problem hiding this comment.
These are two different failure modes, so they should probably be two different XFAILs.
I also really don't think we should be using the Darwin feature to XFAIL/UNSUPPORTED anything in this suite - it's really about the APIs and not the host OS. I'll follow up separately to propose we get rid of config.available_features.add(config.offloadtest_os) in lit.cfg.py completely - it really isn't the right way to slice this stuff.
| # Bug https://github.com/llvm/offload-test-suite/issues/704 | ||
| # XFAIL: Clang && Intel && Vulkan | ||
|
|
||
| Bug https://github.com/llvm/offload-test-suite/issues/705 |
There was a problem hiding this comment.
Commented on the issue, but it needs some detail added to it before I'm comfortable using it to mark an XFAIL. It isn't clear to me whether this is really a driver issue or if it's just a clang bug of some kind.
6 participants
fixes #706
With the following PRs now merged
We have a working base to test booleans on both DXC and Clang.