Bug #20477: RJIT NoMatchingPatternError when initialize Hash literal with more 8 items

• Status: Feedback
• Assignee: k0kubun (Takashi Kokubun)
• Target version: -
• ruby -v: ruby 3.3.1 (2024-04-23 revision c56cd86388) [x86_64-linux]
• Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN

[ukolovda (Dmitry Ukolov)]

When I create Hash with 8 or more items, I got the error:

ruby --rjit -e "a = 1; 1.upto(10) { b={a1: a&1, a2: a&1, a3: a&1, a4: a&1, a5: a&1,  a6: a&1, a7: a&1, a8: a&1} }"

/home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/assembler.rb:124:in `and': [:rax, [:rbx, 128]] (NoMatchingPatternError)
        from /home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/insn_compiler.rb:2348:in `opt_and'
        from /home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/insn_compiler.rb:118:in `compile'
        from /home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/compiler.rb:321:in `compile_block'
        from /home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/compiler.rb:150:in `block in branch_stub_hit'
        from <internal:kernel>:133:in `then'
        from /home/ukolovda/.rvm/rubies/ruby-3.3.1/lib/ruby/3.3.0/ruby_vm/rjit/compiler.rb:149:in `branch_stub_hit'
        from -e:1:in `block in <main>'
        from -e:1:in `upto'
        from -e:1:in `<main>'

Test ruby file:

# test.rb 
# Running: ruby --rjit test.rb
a = 1
1.upto(10) do
    b = { a1: a&1, 
          a2: a&1, 
          a3: a&1, 
          a4: a&1, 
          a5: a&1,  
          a6: a&1, 
          a7: a&1, 
          a8: a&1,
    }
end

(I described it first in #20475, but it is closed now. Current description is more clear and easy reproduce).

[ukolovda (Dmitry Ukolov)]

• Description updated (diff)

[k0kubun (Takashi Kokubun)]

• Copied from Bug #20475: RJIT `and' NoMatchingPatternError added

[k0kubun (Takashi Kokubun)]

• Status changed from Open to Feedback
• Assignee set to k0kubun (Takashi Kokubun)

[k0kubun (Takashi Kokubun)]

Thanks for the easy reproduction steps. As a reminder from the previous ticket,

Given its experimental nature, it's somewhat intended that RJIT's assembler doesn't cover some patterns. It's for developers to try some stuff, so feel free to open a pull request to add support for that instruction encoding, or just switch to --yjit if you just want to use a JIT.

[ukolovda (Dmitry Ukolov)]

Thank you!
I guessed, RJIT will be production solution in some future, sorry.