OptimizeInstruction: Optimize any boolean & (No overlap with boolean's LSB) ==> 0

[xuruiyang2002]

There are so many rules for and, but we still cannot optimize the following one:

(i32.and
 (i32.eqz
  (i32.load $0
   (i32.const 0)
  )
 )
 (i32.const 4)
)

to zero.

Adding rule for add: any boolean & (No overlap with boolean's LSB) ==> 0

Fixes: #7481

[xuruiyang2002]

I've been fuzzing for 60 minutes; and it looks goods enough.

[xuruiyang2002]

Since the PR has been carefully reviewed many times and lots of time has been invested, we should not miss the opportunity to optimize.

Could you please review the code when you have a moment? @kripken

[kripken]

Sorry for the delay.

[kripken]

Thanks!

[kripken]

The fuzzer found a problem here,

(module
 (type $0 (func))
 (type $16 (func (param f64) (result f32)))
 (memory $0 16 17)
 (export "func_28" (func $24))
 (export "func_80" (func $76))
 (func $24 (type $0)
  (local $1 i32)
  (i32.store offset=22 align=2
   (local.tee $1
    (i32.const 1)
   )
   (i32.const 1)
  )
  (drop
   (block (result i32)
    (if
     (i32.const 0)
     (then
      (nop)
     )
     (else
      (i32.store16 offset=2 align=1
       (i32.and
        (local.get $1)
        (i32.const 1)
       )
       (i32.load16_u offset=22
        (i32.const 0)
       )
      )
      (block
       (unreachable)
       (local.tee $1
        (unreachable)
       )
      )
     )
    )
    (i32.const 0)
   )
  )
 )
 (func $76 (type $16) (param $0 f64) (result f32)
  (local $1 f32)
  (local.set $1
   (f32.load offset=4 align=2
    (i32.const 0)
   )
  )
  (local.get $1)
 )
)

$ bin/wasm-opt b.wat --optimize-instructions --fuzz-exec
[fuzz-exec] calling func_28
[trap unreachable]
[fuzz-exec] calling func_80
[fuzz-exec] note result: func_80 => 1.401298464324817e-45
[fuzz-exec] calling func_28
[trap unreachable]
[fuzz-exec] calling func_80
[fuzz-exec] note result: func_80 => 0
[fuzz-exec] comparing func_28
[fuzz-exec] comparing func_80
values not identical! 0 != 1.401298464324817e-45
[fuzz-exec] optimization passes changed results

I haven't investigated yet.

[kripken]

Turns out this just uncovered an existing bug, fix in #7623

[xuruiyang2002]

Turns out this just uncovered an existing bug, fix in #7623

I am glad to hear the PR by myself didn't bring new bug into opt and even uncovered an existing one. 👍

Update: I've understood the bug, and it's somewhat interesting.