Elon Musk may not be right about AIs just generating raw assembler code, but I think he may not be far from the right track!
Why do I think Elon might have a point?
Elon Musk's recent tweet has had quite a lot of ridicule from compiler developers and AI naysayers. The problem is it turns out AIs are really quite good at generating and analyzing assembler code.
The bigger problem is that having AIs generate assembler is silly in the same way it is to have humans do it. The reason we use high level languages is because we want to work with higher levels of abstraction. There's only so much complexity we (or AIs) can handle at the same time.
So why do I think AIs can write good assembler? While I've been working with the Menai compiler I've gone through a stack machine VM and now to a virtual register based VM. Not only did Claude understand both, but was able to debug the compiler based on reading both. It also helped me incrementally transition through a Frankenstein's monster VM in which some opcodes were stack based and some were register based.
Its expertise in all these matters came entirely from reading the Menai compiler code. It's not "oh it's been pre-trained", it's the AI reading and understanding very fast. Impressively, this is not a "thinking" version of Claude - this is dumb Claude doing all this!
So in the last 6 weeks Claude has helped me go through many different iterations of a high level language and then do very deep low-level optimizations. It has done the heavy lifting in building what's now a very good and very fast optimizing compiler, but importantly it jumps in and out of every form of representation the compiler can produce. It writes off-the-cuff tests to explore AST, IR, CFG, and bytecode representations, using any of them to help it shape the compiler logic.
This is where I think Elon is sort of right. The AIs won't write assembler, but they will write the tools that will generate extremely high quality machine code.
Reviewing performance changes
It's interesting to revisit tests we've seen before. With that, I now present the profiler data for the Rubik's cube solver (it's an awful solver, but it's an interesting test):
====================================================================================================
PROFILING RESULTS (Top 80 functions, sorted by cumulative)
====================================================================================================
4614251 function calls (4368849 primitive calls) in 0.964 seconds
Ordered by: cumulative time
List reduced from 374 to 80 due to restriction <80>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 0.979 0.979 /Users/dave/github/m6r/humbug/src/menai/menai.py:1060(evaluate)
1 0.000 0.000 0.979 0.979 /Users/dave/github/m6r/humbug/src/menai/menai.py:1043(_evaluate_raw)
1 0.000 0.000 0.862 0.862 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:341(execute)
89116/1 0.270 0.000 0.846 0.846 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:380(_execute_frame)
89115/4 0.079 0.000 0.846 0.212 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:683(_op_call)
1 0.000 0.000 0.116 0.116 /Users/dave/github/m6r/humbug/src/menai/menai_compiler.py:103(compile)
328491 0.076 0.000 0.113 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:547(_op_push)
118858 0.042 0.000 0.058 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:573(_op_enter)
981715 0.049 0.000 0.052 0.000 {built-in method builtins.isinstance}
89116 0.029 0.000 0.045 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:49(__init__)
89116 0.024 0.000 0.034 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:804(_op_return)
468084 0.034 0.000 0.034 0.000 {method 'append' of 'list' objects}
424304 0.030 0.000 0.030 0.000 {method 'pop' of 'list' objects}
2/1 0.000 0.000 0.027 0.027 /Users/dave/github/m6r/humbug/src/menai/menai_compiler.py:77(compile_to_resolved_ast)
2585/1 0.001 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:89(resolve)
3395/4 0.000 0.000 0.026 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:117(<genexpr>)
1 0.000 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:120(_resolve_import)
1 0.000 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai.py:1193(load_module)
41405 0.020 0.000 0.026 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2756(_op_list_append)
601001 0.025 0.000 0.025 0.000 /opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/typing.py:2371(cast)
43125 0.016 0.000 0.023 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2840(_op_list_ref)
28815 0.013 0.000 0.022 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2866(_op_list_null_p)
5350 0.003 0.000 0.022 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2674(_op_dict_set)
1 0.000 0.000 0.020 0.020 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:184(build)
35/1 0.000 0.000 0.020 0.020 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:207(_emit_function_body)
170/1 0.000 0.000 0.020 0.020 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:267(_emit_block)
1643/117 0.001 0.000 0.020 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:286(_emit_instr)
34/30 0.000 0.000 0.019 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:475(_emit_make_closure)
34/30 0.000 0.000 0.019 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:526(_generate_lambda_code_object)
5350 0.007 0.000 0.018 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:441(set)
2 0.000 0.000 0.017 0.009 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:48(optimize)
26075 0.007 0.000 0.017 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2795(_op_list_rest)
1 0.000 0.000 0.016 0.016 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:907(validate_bytecode)
35/1 0.000 0.000 0.016 0.016 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:312(validate)
38535 0.011 0.000 0.016 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2779(_op_list_first)
6 0.000 0.000 0.015 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:82(count)
14465/6 0.007 0.000 0.015 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:123(_walk)
305/6 0.000 0.000 0.015 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:227(_walk_let)
204/180 0.000 0.000 0.015 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:192(_walk_lambda)
96695 0.014 0.000 0.014 0.000 /opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/typing.py:426(inner)
24030 0.010 0.000 0.014 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:1077(_op_integer_neq_p)
2 0.000 0.000 0.013 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:50(optimize)
65127 0.010 0.000 0.013 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:611(_op_jump_if_true)
35 0.003 0.000 0.012 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_cfg_liveness.py:208(allocate_slots)
8939/2 0.007 0.000 0.012 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:54(_inline)
166/2 0.000 0.000 0.012 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:109(_inline_let)
68/60 0.000 0.000 0.011 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:161(_inline_lambda)
2 0.000 0.000 0.011 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:50(optimize)
2 0.003 0.001 0.011 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_lexer.py:62(lex)
176524 0.010 0.000 0.010 0.000 {built-in method builtins.len}
2 0.000 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:67(build)
11566 0.005 0.000 0.009 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2744(_op_list_prepend)
2514/2 0.001 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:118(_parse_expression)
741/2 0.002 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:396(_parse_list)
15/2 0.000 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:433(_parse_let_with_tracking)
15/6 0.000 0.000 0.009 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:500(_parse_let_bindings)
58/39 0.000 0.000 0.009 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:546(_parse_single_binding)
26075 0.007 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:318(rest)
93223 0.008 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:304(_check_and_pack_args)
45948 0.006 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:518(_to_hashable_key)
8458 0.004 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2690(_op_dict_get)
1 0.000 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:117(build)
2192/1 0.001 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:135(_analyze_expression)
742/1 0.001 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:167(_analyze_list)
69/1 0.000 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:293(_analyze_let)
34/30 0.000 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:384(_analyze_lambda)
4838/2 0.005 0.000 0.007 0.004 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:56(_prop)
113/2 0.000 0.000 0.007 0.004 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:111(_prop_let)
2435/1 0.001 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:169(desugar)
13/1 0.000 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:367(_desugar_let)
2/1 0.000 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:395(_desugar_letrec)
68/60 0.000 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:163(_prop_lambda)
35 0.004 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:580(_validate_initialization)
6250 0.003 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:375(__init__)
1275/16 0.001 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:181(_substitute)
26/8 0.000 0.000 0.006 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:257(_substitute_let)
28815 0.004 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:307(is_empty)
66380 0.006 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:510(_op_load_const)
4653/2 0.003 0.000 0.006 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:57(_opt)
88/2 0.000 0.000 0.006 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:99(_opt_let)On 2026-02-28 we'd got the runtime for this test down to 9.6 seconds. Now it's 0.96 seconds. Yes, that is 10x faster!
Same language, same initial codebase. Pretty impressive!
The art of throwing things away
Over the last week, my current token-fueled rampage through compiler land has ended up with me implementing a static single assignment control flow graph analysis. This made it much easier to go to a register-based VM and to collapse a lot of code paths.
The main exercise was all done in a single day. Let's put that in perspective - Claude's initial estimate was this was a couple of months of work. We then reworked the entire bytecode design and got something working.
Along the way we spotted lots of inefficient code generation and wrote a series of CFG passes that resolved the problems. This was all absolutely astonishing.
Here, however, is where things get really interesting. Over dinner last night (I was sitting in a bar!) I had Claude write a lifetime analysis pass that could optimally allocate registers. That took about 3 hours and generated wonderful code. This morning I looked at those earlier passes and asked if they were really doing anything useful anymore.
In 2 hours we had removed 3 passes. Even more bizarrely, Claude decided to deep dive the passes that were generating anything that those passes might have matched and fixed all the root causes.
That last part should really give people pause. To make that work Claude did things like patching the testsuite so it could see which tests triggered any part of those passes. It extracted the tests into custom scripts that dumped out the CFG and the bytecode, looking for what had triggered the phi nodes that were being matched in the passes. Finally it fixed the root cause issues so the unnecessary nodes were never emitted in the first place.
As it went along, it also removed the now-dead passes and refactored. This is not the Claude I started using in late 2024. This is screaming expert territory, except about 100x faster than any human.
Anyway, with this, the Menai compiler is back to 14 passes, and still insanely fast even though it's written in Python!
The power of using AI for optimizing
While I was reading some disassembler output I noticed this:
4018 Function: valid-constraints(1 param)
4019 Instructions: 7
4020 Locals: 2
4021 Constants: 0
4022 Code Objects: 0
4023 ======================================================================
4024 Instructions:
4025 ----------------------------------------------------------------------
4026 0: ENTER 1 ; Store 1 params into locals: 'c'
4027 1: r1 = STRING_P r0
4028 2: JUMP_IF_TRUE r1, @5
4029
4030 3: r1 = LOAD_FALSE ; local[0] = #f
4031 4: JUMP @6
4032
4033 ► 5: r1 = LOAD_TRUE ; local[0] = #t
4034
4035 ► 6: RETURN r1 ; Return value from r1A quick look revealed this came from the following expression:
(lambda (c) (match c ((? string? s) #t) (_ #f)))Having read a lot of assembler over the years, I recognized this could be much simpler. To be fair, however, the compiler had already done a pretty respectable job. This expression is actually pretty silly because it's matching c to see if it's a string, and if it is then returns #t, otherwise the wildcard path causes it to return #f.
Now, I know from a few days ago we'd done some work to handle cases like this but I couldn't remember where that was.
Claude was not deterred. It scanned all the relevant parts of the compiler to track this down to a missed optimization arising from some wrapping inside the IR. This is conversation 2026-03-08-11-12-44-007.conv.
Finding and fixing this took 15 mins, some of which was down to me not being around to approve a tool use. That would have taken hours to find by hand! That 15 minutes also included writing a series of new unit tests to check the expected behaviour.
Here's what the optimized code now looks like:
4018 Function: valid-constraints(1 param)
4019 Instructions: 3
4020 Locals: 2
4021 Constants: 0
4022 Code Objects: 0
4023 ======================================================================
4024 Instructions:
4025 ----------------------------------------------------------------------
4026 0: ENTER 1 ; Store 1 params into locals: 'c'
4027 1: r1 = STRING_P r0
4028 2: RETURN r1 ; Return value from r1
4029 ----------------------------------------------------------------------There is almost no way I'd have bothered chasing this down on my own. Suddenly, however, that's no longer the issue!
And then it's time for another lowering
Building a compiler is a process of successively lowering the code. The key thing is that the more semantic information we have then the more powerful the structural optimizations we can make, but as we lower things we open up all sorts of other possibilities.
One interesting realization was that the CFG is very powerful, but is actually quite limiting in some ways. The phi node structure is quite unfriendly to doing some types of jump optimizations, and we really do want those. What had also happened over the last week is the bytecode generator had ended up getting progressively more complex. This was a pretty strong indication that there was another intermediate structuring that was missing.
With that we slotted in a "vcode" layer that lowers the phi nodes away and really represents the underlying VM. Even without aggressively tuning anything, it's already apparent that this unlocked a new approach to simple tail recursion!
1101 0: ENTER 2 ; Store 2 params into locals: 'date-str', 'calendar'
1103 ► 1: PUSH r0 ; Push param 'date-str' onto call stack
1104 2: PUSH r3 ; Push captured 'add-one-day' onto call stack
1105 3: r5 = CALL 1 ; Call function with 1 arg, result -> r5
1106 4: PUSH r5 ; Push local[0] onto call stack
1107 5: PUSH r1 ; Push param 'calendar' onto call stack
1108 6: PUSH r4 ; Push captured 'is-working-day?' onto call stack
1109 7: r6 = CALL 2 ; Call function with 2 args, result -> r6
1110 8: JUMP_IF_TRUE r6, @11
1112 9: r0 = MOVE r5
1113 10: JUMP @1
1115 ► 11: RETURN r5 ; Return value from r5Previously we'd have pushed r0 and r1 back onto the stack at around instruction 9, and would then have jumped to instruction 0. Now we're not modifying r1 because it never varies, we're doing a register move for r0, but we're now jumping to instruction 1. This reduces 2 PUSH and 1 ENTER opcode into a single MOVE!
This really shows up in our standard Rubik's cube test:
====================================================================================================
PROFILING RESULTS (Top 80 functions, sorted by cumulative)
====================================================================================================
4186180 function calls (3941482 primitive calls) in 0.890 seconds
Ordered by: cumulative time
List reduced from 370 to 80 due to restriction <80>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 0.905 0.905 /Users/dave/github/m6r/humbug/src/menai/menai.py:1060(evaluate)
1 0.000 0.000 0.905 0.905 /Users/dave/github/m6r/humbug/src/menai/menai.py:1043(_evaluate_raw)
1 0.000 0.000 0.781 0.781 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:341(execute)
89116/1 0.261 0.000 0.765 0.765 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:380(_execute_frame)
89115/4 0.080 0.000 0.765 0.191 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:683(_op_call)
1 0.000 0.000 0.124 0.124 /Users/dave/github/m6r/humbug/src/menai/menai_compiler.py:103(compile)
250479 0.059 0.000 0.087 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:547(_op_push)
1000030 0.050 0.000 0.052 0.000 {built-in method builtins.isinstance}
89116 0.028 0.000 0.045 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:49(__init__)
93222 0.029 0.000 0.040 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:573(_op_enter)
89116 0.024 0.000 0.034 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:804(_op_return)
393217 0.028 0.000 0.028 0.000 {method 'append' of 'list' objects}
2/1 0.000 0.000 0.027 0.027 /Users/dave/github/m6r/humbug/src/menai/menai_compiler.py:77(compile_to_resolved_ast)
2585/1 0.001 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:89(resolve)
3395/4 0.000 0.000 0.026 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:117(<genexpr>)
1 0.000 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai_ast_module_resolver.py:120(_resolve_import)
1 0.000 0.000 0.026 0.026 /Users/dave/github/m6r/humbug/src/menai/menai.py:1193(load_module)
41405 0.019 0.000 0.025 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2814(_op_list_append)
346278 0.025 0.000 0.025 0.000 {method 'pop' of 'list' objects}
1 0.000 0.000 0.023 0.023 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:169(build)
5350 0.010 0.000 0.022 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2707(_op_dict_set)
525669 0.021 0.000 0.021 0.000 /opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/typing.py:2371(cast)
2 0.000 0.000 0.018 0.009 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:48(optimize)
28815 0.012 0.000 0.016 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2918(_op_list_null_p)
6 0.000 0.000 0.016 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:82(count)
14465/6 0.008 0.000 0.016 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:123(_walk)
305/6 0.000 0.000 0.016 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:227(_walk_let)
1 0.000 0.000 0.016 0.016 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:907(validate_bytecode)
35/1 0.000 0.000 0.016 0.016 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:312(validate)
43125 0.012 0.000 0.015 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2892(_op_list_ref)
204/180 0.000 0.000 0.015 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_use_counter.py:192(_walk_lambda)
35/1 0.001 0.000 0.015 0.015 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:196(_emit_vcode)
34/30 0.000 0.000 0.015 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:360(_emit_make_closure)
96765 0.015 0.000 0.015 0.000 /opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/typing.py:426(inner)
34/30 0.000 0.000 0.014 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_bytecode_builder.py:389(_emit_lambda)
24030 0.010 0.000 0.014 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:1077(_op_integer_neq_p)
26075 0.010 0.000 0.012 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2851(_op_list_rest)
65127 0.010 0.000 0.012 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:611(_op_jump_if_true)
8939/2 0.007 0.000 0.012 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:54(_inline)
166/2 0.000 0.000 0.012 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:109(_inline_let)
2 0.000 0.000 0.012 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:50(optimize)
68/60 0.000 0.000 0.012 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:161(_inline_lambda)
2 0.000 0.000 0.011 0.006 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:50(optimize)
2 0.003 0.001 0.011 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_lexer.py:62(lex)
140/4 0.000 0.000 0.010 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_cfg_optimization_pass.py:16(optimize)
178444 0.010 0.000 0.010 0.000 {built-in method builtins.len}
140/4 0.001 0.000 0.010 0.002 /Users/dave/github/m6r/humbug/src/menai/menai_cfg_optimization_pass.py:53(_optimize_nested)
2 0.000 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:67(build)
2514/2 0.001 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:118(_parse_expression)
741/2 0.002 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:396(_parse_list)
15/2 0.000 0.000 0.009 0.005 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:433(_parse_let_with_tracking)
15/6 0.000 0.000 0.009 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:500(_parse_let_bindings)
58/39 0.000 0.000 0.009 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ast_builder.py:546(_parse_single_binding)
38535 0.007 0.000 0.009 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2837(_op_list_first)
93223 0.009 0.000 0.009 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:304(_check_and_pack_args)
45948 0.006 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:370(to_hashable_key)
1 0.000 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:117(build)
2192/1 0.001 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:135(_analyze_expression)
742/1 0.001 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:167(_analyze_list)
69/1 0.000 0.000 0.008 0.008 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:293(_analyze_let)
34/30 0.000 0.000 0.008 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_builder.py:384(_analyze_lambda)
2390/1 0.001 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:169(desugar)
13/1 0.000 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:367(_desugar_let)
2/1 0.000 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_ast_desugarer.py:395(_desugar_letrec)
11566 0.006 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2802(_op_list_prepend)
35 0.004 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm_bytecode_validator.py:580(_validate_initialization)
70 0.000 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_cfg_simplify_blocks.py:67(_optimize_function)
1 0.000 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_vcode_builder.py:103(build)
35/1 0.001 0.000 0.007 0.007 /Users/dave/github/m6r/humbug/src/menai/menai_vcode_builder.py:115(_lower_function)
6250 0.004 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_value.py:319(__init__)
1632/117 0.001 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vcode_builder.py:274(_lower_instr)
35 0.002 0.000 0.007 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vcode_allocator.py:80(allocate_slots)
4838/2 0.004 0.000 0.007 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:56(_prop)
113/2 0.000 0.000 0.007 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:111(_prop_let)
1275/16 0.001 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:181(_substitute)
26/8 0.000 0.000 0.006 0.001 /Users/dave/github/m6r/humbug/src/menai/menai_ir_inline_once.py:257(_substitute_let)
8458 0.004 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_vm.py:2743(_op_dict_get)
68/60 0.000 0.000 0.006 0.000 /Users/dave/github/m6r/humbug/src/menai/menai_ir_copy_propagator.py:163(_prop_lambda)
4653/2 0.003 0.000 0.006 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:57(_opt)
88/2 0.000 0.000 0.006 0.003 /Users/dave/github/m6r/humbug/src/menai/menai_ir_optimizer.py:99(_opt_let)That's taking 92.3% of the time from earlier today!