Add Qd8 bf16 qb4w subgraph for fully connected#10818
Open
JakeStevens wants to merge 3 commits into
Open
Conversation
Covers the new fc_type_qd8_bf16_qb4w path (f32 activation -> qdint8 convert -> blockwise int4 weight -> bf16 output, bf16 scales). Only a _BF16 (bf16 scale) variant since there is no qd8_bf16_qb4w_f16_scales create function.
JakeStevens
force-pushed
the
qd8-bf16-qb4w-subgraph-fc
branch
from
July 23, 2026 13:27
bf8a7c1 to
c019fab
Compare
… failure TestStaticB and the dynamic-quant test only skipped on xnn_status_unsupported_hardware; any other CreateRuntime failure fell through and called xnn_reshape_external_value() on a null runtime, segfaulting. This happens for bf16 dynamic-quant fully-connected under XNN_FLAG_NO_INLINED_LHS_PACKING, where CreateRuntime returns xnn_status_unsupported_parameter (the standalone bf16->qdint8 convert has no microkernel). Assert xnn_status_success after the unsupported_hardware skip so an unsupported config fails cleanly instead of dereferencing a null runtime, mirroring the existing guard in the static-weights test below. Repro: build the subgraph fully-connected test and run FullyConnectedQD8BF16QB4W with a bf16 input (bf16 activation -> qdint8 convert); crashes before this change, clean failure after.
JakeStevens
force-pushed
the
qd8-bf16-qb4w-subgraph-fc
branch
from
July 23, 2026 13:29
c019fab to
9e6365e
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Previous PR added support for some microkernels and a f32->bf16 rewrite, but did not add the subgraph, this PR completes an initial FC bring up
Test Plan