Skip to content

compilation: stop pinning aiecc to one thread #185

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
atassis wants to merge 1 commit into
base: devel
Choose a base branch
from
+11 −2
Open

compilation: stop pinning aiecc to one thread #185

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 11 additions & 2 deletions iron/common/compilation/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -556,6 +556,15 @@ def link_files_from(directory: Path) -> None:
link_files_from(build_dir / get_kernel_dir())


# aiecc's own default. "1" here made every design's per-core compiles serial: on
# the encoder-MHA design (24 cores) aiecc costs 7.7 s at -j1 and 6.0 s at -j0, and
# nothing above 8 helps. Safe because -j does not change what aiecc produces --
# measured on that design, insts.bin and all 24 per-core ELFs are byte-identical
# between -j1 and -j16, and input_with_addresses.mlir differs only in the work-dir
# path it embeds, which two runs at the SAME -j also differ in.
Comment on lines +559 to +564

@andrej andrej Sep 8, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
# aiecc's own default. "1" here made every design's per-core compiles serial: on
# the encoder-MHA design (24 cores) aiecc costs 7.7 s at -j1 and 6.0 s at -j0, and
# nothing above 8 helps. Safe because -j does not change what aiecc produces --
# measured on that design, insts.bin and all 24 per-core ELFs are byte-identical
# between -j1 and -j16, and input_with_addresses.mlir differs only in the work-dir
# path it embeds, which two runs at the SAME -j also differ in.
# Default parallelism setting used when environment variable AIECC_JOBS is not set.
# 0 means auto-detect.

_AIECC_DEFAULT_JOBS = "0"


class AieccCompilationRule(CompilationRule):
def __init__(self, use_chess=False, *args, **kwargs):
self.use_chess = use_chess
Expand All @@ -574,7 +583,7 @@ def compile(self, graph):
mlir_source = artifact.mlir_input
work_dir = _aiecc_work_dir(mlir_source.filename)
options = [
f"-j{os.environ.get('AIECC_JOBS', '1')}",
f"-j{os.environ.get('AIECC_JOBS', _AIECC_DEFAULT_JOBS)}",
"--expand-load-pdis",
"--get-scratchpad-parameters",
] + artifact.extra_flags
Expand Down Expand Up @@ -623,7 +632,7 @@ def compile(self, graph):
commands = []
# Now we know for each mlir source if we need to generate an xclbin, an insts.bin or both for it
for mlir_source in mlir_sources:
options = [f"-j{os.environ.get('AIECC_JOBS', '1')}"]
options = [f"-j{os.environ.get('AIECC_JOBS', _AIECC_DEFAULT_JOBS)}"]
xclbin_path = None
insts_path = None
do_compile_xclbin = mlir_source in mlir_sources_to_xclbins
Expand Down
Loading