Add SIMD support for Windows when compiled by Visual C++.

[samyron]

I finally figured out how to build Ruby on Windows. This allowed me to get the SSE2/SIMD code to compile using Visual C++.

The benchmarks were run on a laptop with a Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz (2.59 GHz) running Windows 11 Pro.

Generator

set STAGE=after & ruby -I "lib";"ext" C:/Users/smyro/Development/json/benchmark/encoder2.rb
== Encoding activitypub.json (52595 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after    825.000 i/100ms
Calculating -------------------------------------
              after       8.652k (± 2.0%) i/s  (115.58 μs/i) -     43.725k in   5.055878s

Comparison:
             before :     7174.3 i/s
              after :     8651.7 i/s - 1.21x  faster


== Encoding citm_catalog.json (500298 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after     45.000 i/100ms
Calculating -------------------------------------
              after     450.314 (± 2.9%) i/s    (2.22 ms/i) -      2.295k in   5.100855s

Comparison:
             before :      455.8 i/s
              after :      450.3 i/s - same-ish: difference falls within error


== Encoding twitter.json (466906 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after     84.000 i/100ms
Calculating -------------------------------------
              after     872.990 (± 4.9%) i/s    (1.15 ms/i) -      4.368k in   5.016760s

Comparison:
             before :      742.0 i/s
              after :      873.0 i/s - 1.18x  faster


== Encoding ohai.json (20145 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after      1.154k i/100ms
Calculating -------------------------------------
              after      11.255k (± 1.9%) i/s   (88.85 μs/i) -     56.546k in   5.025850s

Comparison:
             before :    11197.2 i/s
              after :    11255.1 i/s - same-ish: difference falls within error

Parser

set STAGE=after & ruby -I "lib";"ext" C:/Users/smyro/Development/json/benchmark/parser2.rb
== Parsing activitypub.json (58160 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after    530.000 i/100ms
Calculating -------------------------------------
              after       5.563k (± 1.6%) i/s  (179.77 μs/i) -     28.090k in   5.051023s

Comparison:
             before :     4887.4 i/s
              after :     5562.7 i/s - 1.14x  faster


== Parsing twitter.json (567916 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after     45.000 i/100ms
Calculating -------------------------------------
              after     462.771 (± 3.2%) i/s    (2.16 ms/i) -      2.340k in   5.063128s

Comparison:
             before :      355.0 i/s
              after :      462.8 i/s - 1.30x  faster


== Parsing citm_catalog.json (1727030 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after     21.000 i/100ms
Calculating -------------------------------------
              after     217.782 (± 1.8%) i/s    (4.59 ms/i) -      1.092k in   5.016271s

Comparison:
             before :      154.9 i/s
              after :      217.8 i/s - 1.41x  faster


== Parsing ohai.json (32444 bytes)
ruby 4.1.0dev (2026-01-24T14:43:54Z master 1de6133825) +PRISM [x64-mswin64_140]
Warming up --------------------------------------
              after    475.000 i/100ms
Calculating -------------------------------------
              after       4.639k (± 2.9%) i/s  (215.55 μs/i) -     23.275k in   5.022146s

Comparison:
             before :     4209.3 i/s
              after :     4639.2 i/s - 1.10x  faster

[byroot]

So similar to the macOS specific code the other day, I'm a bit on the fence here because it's extra complexity for an environment that is rarely used. And worse than for macOS, it's an environment that I don't have any way to debug other than pushing on CI.

[samyron]

So similar to the macOS specific code the other day, I'm a bit on the fence here because it's extra complexity for an environment that is rarely used. And worse than for macOS, it's an environment that I don't have any way to debug other than pushing on CI.

That's fair. Admittedly, I wouldn't want to sign up to maintain code that I wasn't able to run / debug locally either.

I'm happy to close this. If there is either really strong demand for this at some point or the situation changes, I'd be happy to re-open it.

[byroot]

Sounds good.