Add ComponentTypeInterpretation for joint matrix type #1835
Conversation
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
MrSidims
force-pushed
the
matrix-type-interpret
branch
from
742d425 to
49726cb
Compare
Contributor
Author
|
@asudarsa @vmaksimo @jcranmer-intel please take a look Most of it should be re-written during switch to opaque pointers (see intel/llvm@7707bb0 ), but the switch has several dependencies still, so to unblock BE development I'd prefer to merge it as is and to create opaque pointers adjustments on top of this patch. |
Contributor
Author
|
Actually, it's a breaking change already. Need to think how to avoid it. So merge together with opaque pointers should be the way to go. UPD: changing translation for bf16 is already a breaking change. So left only tf32. |
MrSidims
force-pushed
the
matrix-type-interpret
branch
from
49726cb to
12c0d47
Compare
It specifies how to interpret 'Component Type' when components of a joint matrix are storages for values of different types, for example float for TF32, unsigned short for bfloat16. Spec update: intel/llvm#8175 Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
Signed-off-by: Sidorov, Dmitry <[email protected]>
MrSidims
force-pushed
the
matrix-type-interpret
branch
from
12c0d47 to
b730243
Compare
Signed-off-by: Sidorov, Dmitry <[email protected]>
Contributor
Author
|
@asudarsa @vmaksimo @jcranmer-intel please take a look |
Signed-off-by: Sidorov, Dmitry <[email protected]>
vmaksimo
reviewed
Feb 24, 2023
Signed-off-by: Sidorov, Dmitry <[email protected]>
Contributor
Author
Contributor
Author
vmaksimo
approved these changes
Mar 7, 2023
| break; | ||
| case internal::InternalJointMatrixCTI::PackedInt2: | ||
| case internal::InternalJointMatrixCTI::PackedInt4: | ||
| // Do nothing just now |
fda0
pushed a commit
to fda0/SPIRV-LLVM-Translator
that referenced
this pull request
May 22, 2025
KhronosGroup#1835) It specifies how to interpret 'Component Type' when components of a joint matrix are storages for values of different types, for example float for TF32, unsigned short for bfloat16. At this point only tf32 type interpretation is added during SPIR-V generation. Adding it to bf16 is a breaking change and requires adaptation across drivers. Spec update: intel/llvm#8175 Signed-off-by: Sidorov, Dmitry [email protected] (cherry picked from commit b7c5218)
bokrzesi
added a commit
to bokrzesi/SPIRV-LLVM-Translator
that referenced
this pull request
Jun 10, 2025
…trix type (KhronosGroup#1835)" This reverts commit dea14c5.
bokrzesi
added a commit
to bokrzesi/SPIRV-LLVM-Translator
that referenced
this pull request
Jun 10, 2025
…trix type (KhronosGroup#1835)" This reverts commit dea14c5.
bokrzesi
added a commit
to bokrzesi/SPIRV-LLVM-Translator
that referenced
this pull request
Oct 20, 2025
Original message: Add ComponentTypeInterpretation for joint matrix type (KhronosGroup#1835) It specifies how to interpret 'Component Type' when components of a joint matrix are storages for values of different types, for example float for TF32, unsigned short for bfloat16. At this point only tf32 type interpretation is added during SPIR-V generation. Adding it to bf16 is a breaking change and requires adaptation across drivers. Spec update: intel/llvm#8175 Signed-off-by: Sidorov, Dmitry [email protected]
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
It specifies how to interpret 'Component Type' when components of a joint matrix are storages for values of different types, for example float for TF32, unsigned short for bfloat16.
At this point only tf32 type interpretation is added during SPIR-V generation. Adding it to bf16 is a breaking change and
requires adaptation across drivers.
Spec update:
intel/llvm#8175
Signed-off-by: Sidorov, Dmitry [email protected]