Merge pull request #466 from willow-ahrens/wma/fastmath

bug fixes and performance improvements
finch-tensor · Mar 14, 2024 · 9d0ff9d · 9d0ff9d · willow-ahrens · Mar 14, 2024
2 parents c00f9e5 + fd10d93
commit 9d0ff9d
Show file tree

Hide file tree

Showing 6 changed files with 10 additions and 10 deletions.
diff --git a/Project.toml b/Project.toml
@@ -1,7 +1,7 @@
 name = "Finch"
 uuid = "9177782c-1635-4eb9-9bfb-d9dfa25e6bce"
 authors = ["Willow Ahrens"]
-version = "0.6.16"
+version = "0.6.17"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"

diff --git a/docs/src/interactive.ipynb b/docs/src/interactive.ipynb
@@ -49,11 +49,11 @@
    "file_extension": ".jl",
    "mimetype": "application/julia",
    "name": "julia",
-   "version": "1.10.2"
+   "version": "1.6.7"
   },
   "kernelspec": {
-   "name": "julia-1.10",
-   "display_name": "Julia 1.10.2",
+   "name": "julia-1.6",
+   "display_name": "Julia 1.6.7",
    "language": "julia"
   }
  },

diff --git a/src/execute.jl b/src/execute.jl
@@ -63,7 +63,7 @@ execute(ex) = execute(ex, NamedTuple())
         code = execute_code(:ex, ex; virtualize(:opts, opts, ctx)...)
         quote
             # try
-                @inbounds begin
+                @inbounds @fastmath begin
                     $(code |> unblock)
                 end
             # catch
@@ -229,7 +229,7 @@ function finch_kernel(fname, args, prgm; algebra = DefaultAlgebra(), mode = safe
     end |> pretty |> unresolve |> dataflow |> unquote_literals
     arg_defs = map(((key, val),) -> :($key::$(maybe_typeof(val))), args)
     striplines(:(function $fname($(arg_defs...))
-        @inbounds $(striplines(unblock(code)))
+        @inbounds @fastmath $(striplines(unblock(code)))
     end))
 end
 

diff --git a/src/util/limits.jl b/src/util/limits.jl
@@ -193,6 +193,6 @@ for S in limit_types
     end
 end
 
-Base.promote_rule(::Type{Limit{T}}, ::Type{Limit{S}}) where {T, S} = Limit(promote_type(T, S))
+Base.promote_rule(::Type{Limit{T}}, ::Type{Limit{S}}) where {T, S} = Limit{promote_type(T, S)}
 Base.convert(::Type{Limit{T}}, i::Limit) where {T} = Limit{T}(convert(T, i.val), i.sign)
 Base.hash(x::Limit, h::UInt) = hash(typeof(x), hash(x.val, hash(x.sign, h)))
diff --git a/test/reference32/issues/issue288_concordize_double_let.jl b/test/reference32/issues/issue288_concordize_double_let.jl
@@ -24,10 +24,10 @@ begin
             for i_9 = 1:A_mode1_stop
                 val = X[i_9, j_5]
                 for l_6 = 1:A_mode2_stop
+                    val_2 = A[i_9, l_6, k_6]
                     phase_stop = min(i_9, l_6 + -1)
                     if phase_stop >= i_9
-                        for s_4 = i_9:phase_stop
-                            val_2 = A[i_9, l_6, k_6]
+                        for s_5 = i_9:phase_stop
                             C[i_9, j_5, k_6] = val_2 * val + C[i_9, j_5, k_6]
                             C[i_9, j_5, k_6] = val * val_2 + C[i_9, j_5, k_6]
                         end

diff --git a/test/reference64/issues/short_circuit_bfs.jl b/test/reference64/issues/short_circuit_bfs.jl
@@ -53,7 +53,7 @@ begin
                 if y_lvl_i == phase_stop_2 && A_lvl_2_i == phase_stop_2
                     A_lvl_3_val = A_lvl_2_val[A_lvl_2_q]
                     y_lvl_2_val = y_lvl_val[y_lvl_q]
-                    p_val = (Finch.Chooser{0}())(p_val, ifelse(y_lvl_2_val && A_lvl_3_val, phase_stop_2, 0))
+                    p_val = (Finch.Chooser{0}())(p_val, (Core).ifelse(y_lvl_2_val && A_lvl_3_val, phase_stop_2, 0))
                     y_lvl_q += 1
                     A_lvl_2_q += 1
                 elseif A_lvl_2_i == phase_stop_2