fix memory allocation regression of muladd introduced in v1.10

[ranocha]

In version v1.10, allocations were introduced in muladd (and fma) for some inputs. The old (good) behavior is

julia -e 'using Pkg; Pkg.activate(temp = true); Pkg.add(name = "DoubleFloats", version = v"1.9.4"); using DoubleFloats; foo() = begin a = Double64(1.0); b = Double64(1.0); c = Double64(-1.0); return muladd(a, b, c); end; foo(); @show @allocated foo()'
[...]
#= none:1 =# @allocated(foo()) = 0

The new (bad) behavior in the current release is

julia -e 'using Pkg; Pkg.activate(temp = true); Pkg.add(name = "DoubleFloats", version = v"1.10.0"); using DoubleFloats; foo() = begin a = Double64(1.0); b = Double64(1.0); c = Double64(-1.0); return muladd(a, b, c); end; foo(); @show @allocated foo()'
[...]
#= none:1 =# @allocated(foo()) = 320

This PR fixes this performance regression. I used Codex to create this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 69.53%. Comparing base (41c8746) to head (c503c27).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #321      +/-   ##
==========================================
+ Coverage   69.45%   69.53%   +0.08%     
==========================================
  Files          57       57              
  Lines        3051     3056       +5     
==========================================
+ Hits         2119     2125       +6     
+ Misses        932      931       -1     

☔ View full report in Codecov by Harness.
📢 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.

[ranocha]

@JeffreySarnoff Here are some additional benchmark results:

On current main:

julia> using BenchmarkTools, DoubleFloats

julia> function foo(x, y, z)
           a = Double64(x)
           b = Double64(y)
           c = Double64(z)
           return muladd(a, b, c)
       end
foo (generic function with 1 method)

julia> foo(1.0, 1.0, -1.0)
0.0

julia> @allocated foo(1.0, 1.0, -1.0)
320

julia> @benchmark foo($(Ref(1.0))[], $(Ref(1.0))[], $(Ref(-1.0))[])
BenchmarkTools.Trial: 10000 samples with 961 evaluations per sample.
 Range (min … max):   85.588 ns … 986.472 ns  ┊ GC (min … max): 0.00% … 89.84%
 Time  (median):      96.732 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   100.251 ns ±  28.831 ns  ┊ GC (mean ± σ):  2.16% ±  6.07%

     ▁▁   ▁  ▁▁▁▁▂▄▇█▇▅▄▄▅▅▃▃▂▂▃▃▂▂▁▁▂▂▁                        ▂
  ██████▇███████████████████████████████████▆▇▆▆▆▅▅▅▅▆▆▅▅▄▄▅▄▄▅ █
  85.6 ns       Histogram: log(frequency) by time        121 ns <

 Memory estimate: 320 bytes, allocs estimate: 4.

With this PR:

julia> using BenchmarkTools, DoubleFloats

julia> function foo(x, y, z)
           a = Double64(x)
           b = Double64(y)
           c = Double64(z)
           return muladd(a, b, c)
       end
foo (generic function with 2 methods)

julia> foo(1.0, 1.0, -1.0)
0.0

julia> @allocated foo(1.0, 1.0, -1.0)
0

julia> @benchmark foo($(Ref(1.0))[], $(Ref(1.0))[], $(Ref(-1.0))[])
BenchmarkTools.Trial: 10000 samples with 996 evaluations per sample.
 Range (min … max):  17.487 ns … 42.630 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     20.164 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   20.400 ns ±  2.042 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

     ▆▅▃▅▂ ▁▁    ▁▅█▆▄▅▄▃▃▁▁▁▂▂   ▂▃▁    ▂▂     ▁▃▂           ▂
  ▄▄████████████▇██████████████▇▅▆████▇▇████▇▇▆▆████▇▇▇▆▆▇▆██ █
  17.5 ns      Histogram: log(frequency) by time      26.5 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

[JeffreySarnoff]

giid

[JeffreySarnoff]

done thank you

…

On Tue, Jul 28, 2026 at 4:36 AM Hendrik Ranocha ***@***.***> wrote: *ranocha* left a comment (JuliaMath/DoubleFloats.jl#321) <#321 (comment)> @JeffreySarnoff <https://github.com/JeffreySarnoff> Here are some additional benchmark results: On current main: julia> using BenchmarkTools, DoubleFloats julia> function foo(x, y, z) a = Double64(x) b = Double64(y) c = Double64(z) return muladd(a, b, c) end foo (generic function with 1 method) julia> foo(1.0, 1.0, -1.0)0.0 julia> @allocated foo(1.0, 1.0, -1.0)320 julia> @benchmark foo($(Ref(1.0))[], $(Ref(1.0))[], $(Ref(-1.0))[]) BenchmarkTools.Trial: 10000 samples with 961 evaluations per sample. Range (min … max): 85.588 ns … 986.472 ns ┊ GC (min … max): 0.00% … 89.84% Time (median): 96.732 ns ┊ GC (median): 0.00% Time (mean ± σ): 100.251 ns ± 28.831 ns ┊ GC (mean ± σ): 2.16% ± 6.07% ▁▁ ▁ ▁▁▁▁▂▄▇█▇▅▄▄▅▅▃▃▂▂▃▃▂▂▁▁▂▂▁ ▂ ██████▇███████████████████████████████████▆▇▆▆▆▅▅▅▅▆▆▅▅▄▄▅▄▄▅ █ 85.6 ns Histogram: log(frequency) by time 121 ns < Memory estimate: 320 bytes, allocs estimate: 4. With this PR: julia> using BenchmarkTools, DoubleFloats julia> function foo(x, y, z) a = Double64(x) b = Double64(y) c = Double64(z) return muladd(a, b, c) end foo (generic function with 2 methods) julia> foo(1.0, 1.0, -1.0)0.0 julia> @allocated foo(1.0, 1.0, -1.0)0 julia> @benchmark foo($(Ref(1.0))[], $(Ref(1.0))[], $(Ref(-1.0))[]) BenchmarkTools.Trial: 10000 samples with 996 evaluations per sample. Range (min … max): 17.487 ns … 42.630 ns ┊ GC (min … max): 0.00% … 0.00% Time (median): 20.164 ns ┊ GC (median): 0.00% Time (mean ± σ): 20.400 ns ± 2.042 ns ┊ GC (mean ± σ): 0.00% ± 0.00% ▆▅▃▅▂ ▁▁ ▁▅█▆▄▅▄▃▃▁▁▁▂▂ ▂▃▁ ▂▂ ▁▃▂ ▂ ▄▄████████████▇██████████████▇▅▆████▇▇████▇▇▆▆████▇▇▇▆▆▇▆██ █ 17.5 ns Histogram: log(frequency) by time 26.5 ns < Memory estimate: 0 bytes, allocs estimate: 0. — Reply to this email directly, view it on GitHub <#321?email_source=notifications&email_token=AAM2VRQLKUFN5NFVE6CNTXL5HBQ2LA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKMJQGE4DIOJSGA2KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-5101849204>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAM2VRQ5EIZSCP6EIQORBV35HBQ2LAVCNFSNUABFKJSXA33TNF2G64TZHMYTCOBZHEZDMNBRHNEXG43VMU5TIOJZGAZDMNJYGE32C5QC> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/AAM2VRRY5IT2ELMNDXVMEIL5HBQ2LA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKMJQGE4DIOJSGA2KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/AAM2VRTLMUEIYPMNS7M3H6T5HBQ2LA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTKMJQGE4DIOJSGA2KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you were mentioned.Message ID: ***@***.***>

[ranocha]

Thanks a lot for ahndling this so quickly!