[DLight] Update GEMV rule to support Adreno outer reduction#15730
Merged
Hzfengsy merged 1 commit intoapache:unityfrom Sep 13, 2023
Merged
[DLight] Update GEMV rule to support Adreno outer reduction#15730Hzfengsy merged 1 commit intoapache:unityfrom
Hzfengsy merged 1 commit intoapache:unityfrom
Conversation
spectrometerHBH
approved these changes
Sep 12, 2023
junrushao
approved these changes
Sep 12, 2023
As Adreno has poor shared memory performance, for LLM workloads, we prefer `q4f16_0` instead of `q4f16_1`. This PR adds GEMV rule support for Adreno outer reduction, comparing with `q4f16_1`, we get a performance gain - Prefill(not related with this PR changes): 3636.6501 ms -> 1241.9469 ms - Decode: 211.0834 ms -> 174.9357 ms
Hzfengsy
force-pushed
the
outer_reduction_gemv
branch
from
64e50dc to
3faa86b
Compare
Hzfengsy
pushed a commit
to Hzfengsy/mlc-llm
that referenced
this pull request
Sep 14, 2023
As apache/tvm#15730 merged, it's no need to dispatch pre-tuned kernel anymore. This PR disables the dispatch.
Hzfengsy
pushed a commit
to Hzfengsy/tvm
that referenced
this pull request
Sep 14, 2023
The PR apache#15730 introduced the outer_reduction for adreno gemv. This PR fixes the length issue when applying on dynamic workloads.
Hzfengsy
pushed a commit
to Hzfengsy/mlc-llm
that referenced
this pull request
Sep 14, 2023
As apache/tvm#15730 merged, it's no need to dispatch pre-tuned kernel anymore. This PR disables the dispatch.
MasterJH5574
pushed a commit
to mlc-ai/mlc-llm
that referenced
this pull request
Sep 14, 2023
As apache/tvm#15730 merged, it's no need to dispatch pre-tuned kernel anymore. This PR disables the dispatch.
junrushao
pushed a commit
that referenced
this pull request
Sep 16, 2023
The PR #15730 introduced the outer_reduction for adreno gemv. This PR fixes the length issue when applying on dynamic workloads.
smickey040404
added a commit
to smickey040404/mlc-llm
that referenced
this pull request
Feb 11, 2025
As apache/tvm#15730 merged, it's no need to dispatch pre-tuned kernel anymore. This PR disables the dispatch.
tristankincaid
added a commit
to tristankincaid/mlc-llm
that referenced
this pull request
Feb 16, 2025
As apache/tvm#15730 merged, it's no need to dispatch pre-tuned kernel anymore. This PR disables the dispatch.
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
3 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.
As Adreno has poor shared memory performance, for LLM workloads, we prefer
q4f16_0instead ofq4f16_1.This PR adds GEMV rule support for Adreno outer reduction, comparing with
q4f16_1, we get a performance gain