-
Notifications
You must be signed in to change notification settings - Fork 1.5k
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
[AMD] Refactor decompose-unsupported-amd-conversions pass #4262
[AMD] Refactor decompose-unsupported-amd-conversions pass #4262
Conversation
This PR depends on #3730 |
@@ -1,8 +1,9 @@ | |||
// RUN: triton-opt %s --split-input-file --decompose-unsupported-amd-conversions=arch=gfx942 | FileCheck %s |
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 wanted to rename this file to test/Conversion/amd/decompose-unsupported-conversions-gfx11.mlir, but git marks it as new file and looses history. So I've left it as-is for now.
bfaa932
to
b23e154
Compare
test/Conversion/amd/decompose-unsupported-conversions-gfx9.mlir
Outdated
Show resolved
Hide resolved
// RUN: triton-opt %s --split-input-file --decompose-unsupported-amd-conversions=arch=gfx942 | FileCheck %s | ||
|
||
// CHECK: #[[DST_ENC:.+]] = #triton_gpu.blocked<{{.*}}> | ||
// CHECK: #[[SRC_ENC:.+]] = #triton_gpu.amd_mfma<{{.*}}> |
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.
Wanna specify out some different pieces between SRC_ENC
and TMP_ENC
and change the CHECK
here to CHECK-DAG
? Right now it relies on solely the ordering to not break..
test/Conversion/amd/decompose-unsupported-conversions-gfx9.mlir
Outdated
Show resolved
Hide resolved
test/Conversion/amd/decompose-unsupported-conversions-gfx9.mlir
Outdated
Show resolved
Hide resolved
third_party/amd/lib/TritonAMDGPUToLLVM/DecomposeUnsupportedConversions.cpp
Outdated
Show resolved
Hide resolved
@binarman ping on this. Can you merge in main and address comments? |
@antiagainst sure, this will be first thing after I finish with small-m dimension FMA optimization. |
This PR: - Simplifying pass code, reusing common code - Introduces supports 3d tensors in mfma -> dot conversion(supported in common code from item above) - Adds more lit tests for decompose-unsupported-amd-conversions pass
b23e154
to
cebb7e9
Compare
This PR introduces: