8000 [GPU][DT] Enable e2e tests for tensor.pack op on GPU · Issue #17186 · iree-org/iree · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

[GPU][DT] Enable e2e tests for tensor.pack op on GPU #17186

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.

Sign up for GitHub

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

Jump to bottom
Closed
Tracked by #17181
, #17608 ,  #17720
...
hanhanW opened this issue Apr 25, 2024 · 2 comments
Closed
Tracked by #17181
, #17608 ,  #17720
...

[GPU][DT] Enable e2e tests for tensor.pack op on GPU #17186

hanhanW opened this issue Apr 25, 2024 · 2 comments
Assignees
@hanhanW

Comments

@hanhanW
Copy link
Contributor
hanhanW commented Apr 25, 2024

Some of them are done on CUDA path, and the rest need to be revisited.

iree/tests/e2e/tensor_ops/BUILD.bazel

Lines 76 to 102 in 1fcb89d

iree_check_single_backend_test_suite(
name = "check_cuda",
srcs = enforce_glob(
# keep sorted
[
"extract_slice.mlir",
"pack.mlir",
"tensor_insert_slice.mlir",
],
include = ["*.mlir"],
exclude = [
"pack_dynamic_inner_tiles.mlir",
"pack_i8.mlir",
"tensor_cast.mlir",
"unpack.mlir",
],
),
driver = "cuda",
tags = [
"noasan",
"nomsan",
"notsan",
"noubsan",
"requires-gpu-nvidia",
],
target_backend = "cuda",
)

We also need to revamp the pipeline, given that we have vector distribution and better pack/unpack vectorization today.
@hanhanW hanhanW self-assigned this Apr 25, 2024
@hanhanW hanhanW mentioned this issue Apr 25, 2024
Open
10 tasks
@hanhanW hanhanW mentioned this issue Jun 6, 2024
Merged
@bjacob bjacob mentioned this issue Jun 7, 2024
Open
13 tasks
hanhanW added a commit that referenced this issue Jun 11, 2024
@hanhanW
[GPU] Enable tensor.pack e2e tests for rocm backend. (#17587)
cda3ccb 
The SplitFullPartialTransferPass and WorkgroupSpecializationPass are no longer needed because we have
much more mature vector lowering. It was added long time ago.

Progress on #17186

---------

Signed-off-by: hanhanW <hanhan0912@gmail.com>
@hanhanW
Copy link
Contributor Author
hanhanW commented Jun 11, 2024

We have it functional. The rest of work is benchmarking. and see if it is reasonable.

@hanhanW hanhanW mentioned this issue Jun 21, 2024
Closed
3 tasks
LLITCHEV pushed a commit to LLITCHEV/iree that referenced this issue Jul 30, 2024
@hanhanW @LLITCHEV
[GPU] Enable tensor.pack e2e tests for rocm backend. (iree-org#17587)
067d9d3 
The SplitFullPartialTransferPass and WorkgroupSpecializationPass are no longer needed because we have
much more mature vector lowering. It was added long time ago.

Progress on iree-org#17186

---------

Signed-off-by: hanhanW <hanhan0912@gmail.com>
Signed-off-by: Lubo Litchev <lubol@google.com>
@hanhanW
Copy link
Contributor Author
hanhanW commented May 22, 2025

It's done:

iree/tests/e2e/linalg/BUILD.bazel

Lines 187 to 213 in bd1a7ec

ROCM_SRCS = enforce_glob(
# keep sorted
[
"pack_i8.mlir",
"unpack.mlir",
],
include = ["*.mlir"],
exclude = [
"conv2d.mlir",
"fp_to_subbyte.mlir",
"index.mlir",
"large_linalg_matmul.mlir",
"narrow_n_matmuls.mlir",
"subbyte_to_fp.mlir",
# https://github.com/llvm/llvm-project/issues/131386 causes
# See bug #20294
"pack.mlir",
"pack_dynamic_inner_tiles.mlir",
],
)
iree_check_single_backend_test_suite(
name = "check_rocm_hip",
srcs = ROCM_SRCS,
driver = "hip",
target_backend = "rocm",
)

There is another new issue that is tracked in #20294

@hanhanW hanhanW closed this as completed May 22, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@hanhanW hanhanW

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@hanhanW
0