[Matrix] Add comparison operator tests - #1447
Open
farzonl wants to merge 6 commits into
Open
Conversation
farzonl
added
the
test-all
Member
Author
|
So I don't have to come back and remove a bunch of xfails lets wait on this PR until llvm/llvm-project#216791 merges |
Member
Author
|
the tests that have 64/16 bit requires are not as exhaustive with operator testing as the 32bit tests. |
kmpeng
self-requested a review
August 19, 2026 17:22
Member
Author
|
Failures to investigate: DXC MetalClang DXC MetalThe issue in Metal seems to be offloader specific. We don't have a uint32 Format. |
farzonl
removed
the
test-all
farzonl
force-pushed
the
task/issue-1446
branch
from
August 20, 2026 16:58
69504ae to
61d2f34
Compare
resolves llvm#1446 cover cmp operators across matrix sizes and base types
farzonl
force-pushed
the
task/issue-1446
branch
from
August 20, 2026 17:40
61d2f34 to
b054c3a
Compare
kmpeng
reviewed
Aug 20, 2026
…are doing elementwise compares
kmpeng
approved these changes
Aug 24, 2026
Icohedron
reviewed
Aug 25, 2026
Icohedron
left a comment
Icohedron
left a comment
Contributor
There was a problem hiding this comment.
I don't see why the 32-bit tests have to be the only comprehensive ones.
The 32-bit tests aren't even that much bigger than the other tests in line count either.
e.g. the fp16 test is 72 lines while the fp32 test is 105 lines
|
|
||
| [numthreads(1, 1, 1)] | ||
| void main() { | ||
| half4x3 HalfEQLHS = HalfIn[0], HalfEQRHS = HalfIn[1]; |
Contributor
There was a problem hiding this comment.
[SUGGESTION] You should do the same as commit 8ebdcbd and make all the tests use unique matrix elements to confirm that the compares are actually done elementwise. Otherwise the comparisons collapse to single fcmp instructions.
| ... | ||
| #--- end | ||
|
|
||
| # Unimplemented https://github.com/microsoft/DirectXShaderCompiler/issues/8796 |
Contributor
There was a problem hiding this comment.
[QUESTION] I'm wondering why these tests don't carry the same platform exclusion as the bool-matrix tests.
matrix_bool_and_operator.test, matrix_bool_and_operator_thread_group.test and
matrix_bool_or_operator.test all end with:
# Note: Metal does not support boolean matrix types. For
# Vulkan KosmicKrisp and MoltenVK the output is always False.
# UNSUPPORTED: Metal || KosmicKrisp || MoltenVK
Every test in this PR produces boolNxM values from the comparison and indexes them.
Member
Author
There was a problem hiding this comment.
The failures don't happen when you produce a boolean matrix only when you try and do an operation on them. At least for metal. I have not tested the mac vulkan runtimes.
| @@ -0,0 +1,71 @@ | |||
| #--- source.hlsl | |||
|
|
|||
| RWBuffer<int64_t> Int64In : register(u0); | |||
Contributor
There was a problem hiding this comment.
[ISSUE] RWBuffer<int64_t> has a known issue in Lavapipe.
The existing typed 64-bit test, test/Feature/TypedBuffer/64bit-scalar.test has this XFAIL:
# Driver bug https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15866
# XFAIL: Lavapipe
So either switch to a RWStructuredBuffer<uint64_t> or add the same XFAIL.
| Buffers: | ||
| - Name: UintIn | ||
| Format: UInt32 | ||
| Data: [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 ] |
Contributor
There was a problem hiding this comment.
[SUGGESTION] The int and uint tests should exercise values where the sign bit is actually set. As-is, there is no observable difference if the compiler used signed/unsigned operations where it should have been unsigned/signed.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
resolves #1446
cover cmp operators across matrix sizes and base types
assisted by co-pilot MAI model