Optimize Jacobi update second step vertex #49
Merged
Conversation
This file contains 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
balancap
force-pushed
the
optimize-jacobi-update-second-step-vertex
branch
3 times, most recently
from
dc61601
to
6ab4883
Compare
balancap
force-pushed
the
optimize-jacobi-update-second-step-vertex
branch
from
6ab4883
to
6551867
Compare
balancap
changed the title
The second step of the Jacobi algorithm is an usual "sparse" access update pattern, which can not be simply optimized with a `rpt` loop. The loop is intrinsically limited by the number of load/store operations. Nevertheless, by using `aop` outer product intrinsic + unrolling of 2 steps, the bundling of operations in the loop can be massively improved, leading to 40% decrease of cycle count. Additionally, this PR is adding a thin AMP C++ abstraction, allowing a simple implementation between IPU hardware and IPU model.
balancap
force-pushed
the
optimize-jacobi-update-second-step-vertex
branch
from
6551867
to
1335dbb
Compare
balancap
added a commit
that referenced
this pull request
Oct 18, 2023
The recent improvement in PR #49 introduced a regression in Jacobi `eigh`, raising an error when size % 4 == 2. This is due to the partial loop unrolling in Jacobi second update stage. This PR is fixing the issue by passing explicitely the offset and size of the workload to the vertex.
balancap
added a commit
that referenced
this pull request
Oct 18, 2023
The recent improvement in PR #49 introduced a regression in Jacobi `eigh`, raising an error when size % 4 == 2. This is due to the partial loop unrolling in Jacobi second update stage. This PR is fixing the issue by passing explicitely the offset and size of the workload to the vertex.
balancap
added a commit
that referenced
this pull request
Oct 19, 2023
The recent improvement in PR #49 introduced a regression in Jacobi `eigh`, raising an error when size % 4 == 2. This is due to the partial loop unrolling in Jacobi second update stage. This PR is fixing the issue by passing explicitely the offset and size of the workload to the vertex.
balancap
added a commit
that referenced
this pull request
Oct 19, 2023
The recent improvement in PR #49 introduced a regression in Jacobi `eigh`, raising an error when size % 4 == 2. This is due to the partial loop unrolling in Jacobi second update stage. This PR is fixing the issue by passing explicitely the offset and size of the workload to the vertex.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
The second step of the Jacobi algorithm is an usual "sparse" access
update pattern, which can not be simply optimized with a
rptloop.The loop is intrinsically limited by the number of load/store
operations. Nevertheless, by using
aopouter product intrinsic +unrolling of 2 steps, the bundling of operations in the loop can be
massively improved, leading to 40% decrease of cycle count.
Additionally, this PR is adding a thin AMP C++ abstraction, allowing
a simple implementation between IPU hardware and IPU model.