Add wasmtime hot-blocks subcommand

[fitzgen]

This commit adds a new wasmtime hot-blocks subcommand that profiles Wasm
execution using perf and identifies the hottest basic blocks in the compiled
Wasm code.

Example output:

$ cargo run -- hot-blocks -F 999 -p 100 tests/all/cli_tests/fib.wat
`fib` :: block 0x16 :: 81.62% total samples

 [Samples]   [Assembly]                 [CLIF]                       [Wasm]
    14.07%   add     esi, 1             -                            -
    21.29%   lea     ecx, [rax + rdi]   v14 = iadd.i32 v13, v12      i32.add
    13.29%   mov     rdi, rax           "                            "
    17.64%   mov     rax, rcx           "                            "
    15.33%   jmp     0xe                jump block3(v18, v13, v14)   br $loop

`fib` :: block 0x0 :: 18.33% total samples

 [Samples]   [Assembly]         [CLIF]                     [Wasm]
    18.33%   cmp     esi, edx   brif v11, block2, block5   br_if $break
             jae     0x2a       "                          "

`wasm[0]::function[1]` :: block 0x37 :: 0.04% total samples

 [Samples]   [Assembly]       [CLIF]             [Wasm]
     0.04%   add     ebx, 1   -                  -
             jmp     0x6b     jump block3(v14)   br $loop

Depends on #13073

[cfallin]

This looks reasonable overall -- with my level of review on all of the output-parsing code calibrated at "this is a useful tool for developers" rather than "production-ready robustness".

A thought below but otherwise I'm happy to see this in-tree for its usefulness. Thanks!

[alexcrichton]

Nice!

Stray thought on this: could debuginfo factor into this as well during rendering? For example at minimum this could print the hex offset in wasm which could then be passed to wasm-tools addr2line, but at most it could run addr2line here in-process and print some information. If it's not easy to add though probably not worth it

[fitzgen]

Stray thought on this: could debuginfo factor into this as well during rendering? For example at minimum this could print the hex offset in wasm which could then be passed to wasm-tools addr2line, but at most it could run addr2line here in-process and print some information. If it's not easy to add though probably not worth it

Sure, I can investigate this in a follow up. Are you imagining, when we have debuginfo available, output that loks something like this?

 [Samples]   [Assembly]                 [CLIF]                       [Wasm]
    14.07%   add     esi, 1             -                            -
    21.29%   lea     ecx, [rax + rdi]   v14 = iadd.i32 v13, v12      0x12 fib fib.rs:12:4

[fitzgen]

Hm or maybe something like objdump's mixed source and asm disassembly? That way we could still see the actual Wasm instructions.

[alexcrichton]

I was just thinking of source locations for now, so no need to read files on disk and intermingle in source code just yet. What you sketched sounds reasoanble to me!