-
Notifications
You must be signed in to change notification settings - Fork 320
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
Update LLVM/MHLO to week of 2022/11/14 #1905
Conversation
I hereby confirm that I abide by the DCO requirements. |
Can one of the admins verify this patch? |
…2f2594fb8' Signed-off-by: Dominik Montada <dominik.montada@amd.com>
2cc2d91
to
1267ad5
Compare
Can one of the admins verify this patch? |
@@ -47,7 +47,7 @@ func.func @test_allocs_not_lowered(%arg0: memref<10x10xf32>, %arg1: memref<10x10 | |||
return %0 : memref<10x10xf32> | |||
} | |||
|
|||
// CHECK: [[MAP:#.+]] = affine_map<(d0, d1) | |||
// CHECK: [[MAP:#.+]] = affine_map<(d0, d1) -> (0, d1 floordiv 64, 0, d0 floordiv 32, d0 mod 32, d1 mod 64) |
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 previous version only matched on the prefix of the map and it happened to work. Now with the update to affine maps, only matching the prefix will match the wrong map. I had to make the CHECK more specific to match the correct map
@jenkins-droid test this please |
@gargaroff Jenkins amd64, ppc64le, and s390x failed due to lit tests for NNPA. I guess they failed because of
|
Can one of the admins verify this patch? |
@tungld Thanks for pointing me to them. I didn't realize that I was building without accelerators. Locally it's passing now, so I hope this one works now. |
Signed-off-by: Dominik Montada <dominik.montada@amd.com>
Signed-off-by: Dominik Montada <dominik.montada@amd.com>
96839bb
to
fb6ebe9
Compare
Can one of the admins verify this patch? |
@jenkins-droid test this please |
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.
LGTM. Thanks for the update!
You're welcome and thanks for the review! |
CMakeLists.txt
Outdated
@@ -151,6 +151,10 @@ add_subdirectory(third_party/rapidcheck) | |||
|
|||
if (ONNX_MLIR_ENABLE_MHLO) | |||
add_subdirectory(third_party/mlir-hlo EXCLUDE_FROM_ALL) | |||
|
|||
# mlir-hlo depends on these paths being visible, but does not propagate them. | |||
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/third_party/mlir-hlo) |
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.
I'd use ONNX_MLIR_SRC_ROOT and ONNX_MLIR_BIN_ROOT. Is there an issue opened against mlir-hlo to propagate the paths going forward?
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.
I have opened tensorflow/mlir-hlo#52, but so far there hasn't been any real reaction to it.
Also I have updated CMake to use the two variables you mentioned instead.
Can one of the admins verify this patch? |
… includes Signed-off-by: Dominik Montada <dominik.montada@amd.com>
114c258
to
aa9b0ed
Compare
Can one of the admins verify this patch? |
Jenkins Linux amd64 Build #8981 [push] Update LLVM/MHLO to week... started at 20:23 |
Jenkins Linux s390x Build #8997 [push] Update LLVM/MHLO to week... started at 21:23 |
Jenkins Linux ppc64le Build #8047 [push] Update LLVM/MHLO to week... started at 21:27 |
Jenkins Linux amd64 Build #8981 [push] Update LLVM/MHLO to week... passed after 2 hr 6 min |
Jenkins Linux ppc64le Build #8047 [push] Update LLVM/MHLO to week... passed after 2 hr 37 min |
Jenkins Linux s390x Build #8997 [push] Update LLVM/MHLO to week... passed after 2 hr 41 min |
Green LLVM commit: e864ac694540342d5e59f59c525c5082f2594fb8
Green MHLO commit: eab364ba2a66bd0613efb94f8a738c1c97aaee92
The commits listed above are the green commits from Nov. 14.
This is the LLVM bump that @ljfitz mentioned in #1624. We would love to kick off a discussion about introducing a regular rotation for bumping to green commits, similar to what torch-mlir is doing.
Changelog:
mlir-hlo changed their directory structure, requiring two
include_directories
directives to make it buildable.Generation of
#map
has changed, requiring large NFC changes to test references.Signed-off-by: Dominik Montada dominik.montada@amd.com