Add amd-aie-direct HAL target (3/n)
#420
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
makslevental
requested review from
nirvedhmeshram,
MaheshRavishankar,
yzhang93,
Abhishek-Varma and
jtuyls
as code owners
makslevental
changed the base branch from
makslevental/remove-mlir-aie/aie_r_dep
to
makslevental/remove-mlir-aie/map_object_fifo_to_lib
makslevental
changed the title
amd-aie-direct HAL target (2/n)amd-aie-direct HAL target (3/n)
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
7d93c21 to
5a46312
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/map_object_fifo_to_lib
branch
from
8e65276 to
9b3416a
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
2 times, most recently
from
dfcbfa3 to
a295d90
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
a295d90 to
41c9b5e
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/map_object_fifo_to_lib
branch
2 times, most recently
from
77bead4 to
775351e
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
41c9b5e to
37599a1
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/map_object_fifo_to_lib
branch
5 times, most recently
from
b375fab to
dfea3a6
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
4 times, most recently
from
7fa1376 to
d4dea46
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/map_object_fifo_to_lib
branch
3 times, most recently
from
329e0e7 to
2eda0d0
Compare
Base automatically changed from
makslevental/remove-mlir-aie/map_object_fifo_to_lib
to
main
June 20, 2024 15:47
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
d4dea46 to
7c45ac3
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
15 times, most recently
from
bf23c05 to
3578f9e
Compare
|
@newling @nirvedhmeshram this is now ready for your keen eyes! |
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
3578f9e to
ac58e66
Compare
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/XCLBinGen.h
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.cpp
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Transforms/Passes.h
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETargetDirect.cpp
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETargetDirect.cpp
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETargetDirect.cpp
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/PluginRegistration.cpp
Outdated
Show resolved
Hide resolved
nirvedhmeshram
approved these changes
Jun 24, 2024
newling
approved these changes
Jun 24, 2024
| std::string XCLBinKernelID; | ||
| std::string XCLBinInstanceName; | ||
| bool UseChess = false; | ||
| bool DisableThreading = false; |
There was a problem hiding this comment.
These flags -- do we still need them?
Specifically 'DisableThreading' here is just something we had to throw over the wall because the mlir-aie passes are run on a completely new process (i.e. a new MLIR Context). If the passes are now being run in the same context.
Not a request for change, just some information for you.
|
|
||
| namespace mlir::iree_compiler::AMDAIE { | ||
|
|
||
| // static llvm::cl::opt<std::string> clEnableAMDAIEUkernels( |
There was a problem hiding this comment.
This flag is in AIETarget and so this target should also support it but I can't register the cl::opt (because they're static globals). So I left it here as a reminder to myself.
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETargetDirect.cpp
Outdated
Show resolved
Hide resolved
compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETargetDirect.cpp
Outdated
Show resolved
Hide resolved
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
2 times, most recently
from
52d7a11 to
d4a34e1
Compare
makslevental
force-pushed
the
makslevental/remove-mlir-aie/direct_target
branch
from
d4a34e1 to
3f3cd29
Compare
nirvedhmeshram
added a commit
that referenced
this pull request
Jun 28, 2024
Since #420 added a internal utility to go from AIE to XCLBIN it makes sense for all flows to use that. This utility doesn't add any AIE lowering passes so the `TranslationPassPipeline`s are responsible for that. Hence the passes needed for the IR lowered through the AIR flow to be further lowered to AIE are added in its lowering pipeline. There were some minor additions needed to the utility to get it working e2e for the AIE flow. We no longer need to have a separate `aie2xclbin-disable-threading` as we are passing the existing context and the `mlir-disable-threading` flag gets automatically propagated in this case. And it actually gives an error to try changing it if the passed context is multi-threaded. With this PR the AIE lowering is becoming a part of the IREE `TranslationPassPipeline` so the pack-peel lit test that was just passing the air-to-aie conversion had to be watered down to reflect what can actually be lowered to AIE. Unfortunately, we still also need the mlir-aie wheel as the internal utility is still expecting a mlir-aie install for me_basic.o library [here](https://github.com/nod-ai/iree-amd-aie/blob/main/compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/XCLBinGen.cpp#L302-L304)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR is part of a stack of PRs that refactor the dependency on MLIR-AIE. See #430 for more information.
This PR adds a new target
amd-aiet-direct, directly references AIEVec passes (for eventual lowering of core code to chess LLVM IR) and also (simultaneously) vendorsXCLBinGen.cppfrommlir-aie(second part used to be broken out in #422 but on @newling's suggestion I squashed). Note,XCLBinGenandAIETargetDirectare basically carbon copies of what currently exists exceptAIETargetDirectdoesn't shell out and callsaie2xclbindirectly. Note also thatXCLBinGenwill undergo further revision up to for the chess backend (but that'll have to wait until I get home).To test, I added some lines to
print_ir_aie2xclbin.sh. I'm happy to bike shed this but probably not too hard.