-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Refactor convolution test suite #24
Refactor convolution test suite #24
Conversation
2494de4
to
d8f70a1
Compare
Signed-off-by: erman-gurses <[email protected]>
1523ab5
to
fc08f82
Compare
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
@ScottTodd, the CPU part is done. Let me know if you have any input. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One comment for now. Haven't looked at the generate python script or the CI workflow logs yet.
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
mostly LGTM, a few small comments
linalg_ops/convolution/generated/f16_f16_f16/conv2d_f16_f16_f16_medium_calls.mlir
Outdated
Show resolved
Hide resolved
linalg_ops/convolution/generated/f16_f16_f16/conv2d_f16_f16_f16_medium.mlir
Outdated
Show resolved
Hide resolved
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
linalg_ops/convolution/generated/f16_f16_f16/conv2d_f16_f16_f16_large.mlir
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The formats for CPU tests and GPU tests are quite different, but they should be able to use the same source IR. Can we consolidate the test generation into a single format to reduce the number of generated tests?
It would be good to cover the following element type and layout combinations on both CPU and GPU to start:
- Both NCHW and NHWC convs for all the following dtype combinations:
- f16_f16_f16 (with and without winograd)
- f32_f32_f32 (with and without winograd)
- f16_f16_f32 (with and without winograd)
- i8_i8_i32 (without winograd)
EDIT: Btw, I don't mean to block progress if this is just an initial step, but if we are merging these tests, then we should try to share the implementation. If you would prefer to do this as a follow up that is also okay.
Signed-off-by: erman-gurses <[email protected]>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good, thanks!
Signed-off-by: erman-gurses <[email protected]>
595f8ec
to
bb4257b
Compare
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Progress on #2.
This PR refactors, adds, and migrates conv2d e2e tests from the PRs here: iree-org/iree#16849 and iree-org/iree#18125