Run simulations in parallel #347
Conversation
|
I'm sorry to be so annoying for this, but I don't like that we need to specify all of the dasm targets in every test separately, this isn't very easy to understand. Such that all of the rules in the downstream Snakefiles simplify to: |
|
@jorendumoulin that is a great idea! I'll work on it! |
kernels/gemm/Snakefile
Outdated
| module snax_rules: | ||
| snakefile: | ||
| "../../util/snake/snax.smk" | ||
| config: | ||
| config | ||
|
|
||
|
|
||
| use rule * from snax_rules as snax_* | ||
|
|
||
|
|
There was a problem hiding this comment.
one small nit: can we import default_rules in snax_rules such that we don't need two imports here? I feel that would be more intuitive and less code
There was a problem hiding this comment.
That's a very good suggestion! It hadn't crossed my mind yet :) . I'll try to see if that works!
JosseVanDelm
force-pushed
the
Josse/parallel-simulation
branch
from
87a8183 to
576a518
Compare
| # use CONDA_PREFIX to access pixi env | ||
| gen_trace_path = f"{os.environ['CONDA_PREFIX']}/snax-utils/gen_trace.py" | ||
| return { | ||
| "python": "python", |
There was a problem hiding this comment.
is this snax-specific?
| @@ -6,3 +9,13 @@ def get_default_paths() -> dict[str, str]: | |||
| "mlir-translate": "mlir-translate", | |||
| "snax-opt": "snax-opt", | |||
There was a problem hiding this comment.
is this not snax-specific?
kernels/gemm/Snakefile
Outdated
| "gemm.x", | ||
| shell: | ||
| "{config[vltsim]} {input[0]}" | ||
| expand("{file}_aggregated.json", file=files), |
There was a problem hiding this comment.
| expand("{file}_aggregated.json", file=files), | |
| expand("{file}_traces.json", file=files), |
?
This adds --prefix-trace support and several utilities to handle the quite difficult naming scheme provided by the simulation tracer.