mem effects

[wsmoses]

No description provided.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.75%. Comparing base (8040a3e) to head (99fbaeb).
⚠️ Report is 13 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1666      +/-   ##
==========================================
- Coverage   68.82%   67.75%   -1.07%     
==========================================
  Files         103      108       +5     
  Lines       11596    11783     +187     
==========================================
+ Hits         7981     7984       +3     
- Misses       3615     3799     +184     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[giordano]

I'm on the phone, but I think this entire for loop until the final return can be written as

any(at -> String(at) == "write", attr)

[wsmoses]

problem is that attr doesn't define an iterator

[giordano]

It has indices but isn't iterable? That sounds unfortunate, the basic iteration interface is a single function: https://docs.julialang.org/en/v1/manual/interfaces/#man-interface-iteration

[wsmoses]

yeah I mean ironically the definition of that is in this repo (attribute.jl)

[jumerckx]

I can push this if you want?

diff --git a/src/TracedUtils.jl b/src/TracedUtils.jl
index 64619741..7d6d4de1 100644
--- a/src/TracedUtils.jl
+++ b/src/TracedUtils.jl
@@ -278,12 +278,7 @@ function is_pure(func)
     if attr isa Nothing
         return false
     end
-    for i in 1:length(attr)
-        sa = Base.String(attr[i])
-        if sa == "write"
-            return false
-        end
-    end
+    any(at -> String(at) == "write", attr) && return false
     return true
 end
 
diff --git a/src/mlir/IR/Attribute.jl b/src/mlir/IR/Attribute.jl
index 1321c3e9..ec12785d 100644
--- a/src/mlir/IR/Attribute.jl
+++ b/src/mlir/IR/Attribute.jl
@@ -802,6 +802,14 @@ function Base.getindex(attr::Attribute, i)
     end
 end
 
+function Base.iterate(attr::Attribute, state=1)
+    if state > length(attr)
+        nothing
+    else
+        (attr[state], state + 1)
+    end
+end
+
 function Base.getindex(attr::Attribute)
     @assert isdenseelements(attr) "attribute $(attr) is not a dense elements attribute"
     @assert issplat(attr) "attribute $(attr) is not splatted (more than one different elements)"

[wsmoses]

go for it!

[github-actions]

[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

	Reactant.TracedUtils.is_pure(func3)
	Reactant.TracedUtils.is_pure(func3),

[github-actions]

[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

	function codegen_xla_call(flatten_names, nresults, is_sharded::Bool, ndevices::Int, is_pure::Bool)
	function codegen_xla_call(
	flatten_names, nresults, is_sharded::Bool, ndevices::Int, is_pure::Bool
	)

[github-actions]

[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

	flatten_arg_names, length(linear_results), mlir_fn_res.is_sharded, ndevices, mlir_fn_res.is_pure
	flatten_arg_names,
	length(linear_results),
	mlir_fn_res.is_sharded,
	ndevices,
	mlir_fn_res.is_pure,

[romanlee]

Minor: but not necessary to actually calculate this here since it's recalculated in compile_mlir! I think. Could just pass false instead.

[wsmoses]

the mlir func may go out of scope so we do need to precompute

[romanlee]

Oh I see

[romanlee]

Looks good from my side besides the comments above

[giordano]

This broke building the docs https://github.com/EnzymeAD/Reactant.jl/actions/runs/17802463542/job/50605626751

[wsmoses]

oh let me fix that in post, I thought docs failure was expected from some ssl issue but thats clearly not