Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
H
Hercules
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Package registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
llvm
Hercules
Commits
d179193d
Commit
d179193d
authored
1 month ago
by
Aaron Councilman
Browse files
Options
Downloads
Patches
Plain Diff
Parallel tiled cpu schedule
parent
4b1630b2
No related branches found
Branches containing commit
No related tags found
1 merge request
!205
Fork reshape
Pipeline
#201835
canceled
1 month ago
Stage: test
Changes
1
Pipelines
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
juno_samples/matmul/src/cpu.sch
+63
-0
63 additions, 0 deletions
juno_samples/matmul/src/cpu.sch
with
63 additions
and
0 deletions
juno_samples/matmul/src/cpu.sch
0 → 100644
+
63
−
0
View file @
d179193d
macro optimize!(X) {
gvn(X);
phi-elim(X);
dce(X);
ip-sroa(X);
sroa(X);
dce(X);
gvn(X);
phi-elim(X);
dce(X);
}
macro codegen!(X) {
gcm(*);
float-collections(*);
dce(*);
gcm(*);
}
optimize!(*);
fixpoint panic after 20 {
forkify(matmul);
fork-guard-elim(matmul);
}
// Mark the whole loop nest as associative, any order of iterations is equivalent
associative(matmul@outer);
// Tile the outer 2 loops to create 16 parallel threads (each responsible for
// computing one block of the output
let par = matmul@outer \ matmul@inner;
fork-tile[4, 0, false, true](par);
fork-coalesce(par);
fork-interchange[0, 1](par);
fork-interchange[2, 3](par);
fork-interchange[1, 2](par);
let split = fork-split(*);
fork-coalesce(split.matmul.fj0 \ split.matmul.fj2);
parallel-fork(split.matmul.fj0 \ split.matmul.fj2);
// Pull the body of the parallel loop out into its own device function
let body = outline(split.matmul.fj2);
cpu(body);
// Tile the loop nest for cache performance; 16x16x16 tile
fork-tile[16, 0, false, true](body);
fixpoint { fork-coalesce(body); }
fork-interchange[1, 2](body);
fork-interchange[3, 4](body);
fork-interchange[2, 3](body);
optimize!(*);
fork-split(body);
reduce-slf(*);
unforkify(body);
optimize!(*);
codegen!(*);
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment