1741395 - Add debug printing for the MIR-to-LIR conversion pass (lowering)

Status: NEW

(Core :: JavaScript Engine: JIT, enhancement, P3)

Description

[Julian Seward [:jseward]]

Debugging/changing the MIR-to-LIR conversion pass would be easier,
especially in the presence of the emit-at-uses machinery, if there were
a way to print the LIR generated from each MIR. That would make it
completely obvious which LIRs originate from which MIRs.

Comment 1

[Julian Seward [:jseward]]

Example output from initial patch. This makes it clear

• that MIRs 3 to 6 generate no LIRs, presumably as a result of emit-at-uses
• how phi nodes are handled in LIR generation, which is otherwise something
  of a mystery to the uninitiated

===== BEGIN LIRGenerator::generator =====
  0:wasmparameter
    'f138+=  WasmParameter
  1:wasmparameter
    'f138+=  WasmParameter
  2:wasmparameter
    'f138+=  WasmParameter
  3:constant 0x4d2
  4:wasmbinarybitwise (0:wasmparameter) (1:wasmparameter)
  5:constant 0x0
  6:compare (4:wasmbinarybitwise) (5:constant 0x0)
  7:test (6:compare)
    'f138+=  BitAndAndBranch
  --
  8:constant 0x11d7
  [lowering input #0 for: 11:phi (8:constant 0x11d7) (3:constant 0x4d2)]
    'f178+=  Integer
  9:goto
    'f178+=  Goto
  --
  [lowering input #1 for: 11:phi (8:constant 0x11d7) (3:constant 0x4d2)]
    'f1b8+=  Integer
  10:goto
    'f1b8+=  Goto
  --
  11:phi (8:constant 0x11d7) (3:constant 0x4d2)
    [phis generate no LIRs]
  12:wasmreturn (11:phi) (2:wasmparameter)
    'f1f8+=  WasmReturn
===== END LIRGenerator::generator =====

Comment 2

[Julian Seward [:jseward]]

Attachment: bug1741395-mir2lir-printing.diff-1

WIP patch. Needs multiple #ifdef JS_JITSPEW adding.

Comment 3

[Nicolas B. Pierron [:nbp]]

I think each LIR had a mirRaw pointer?

Comment 4

[Julian Seward [:jseward]]

Attachment: bug1741395-mir2lir-printing.diff-11

Updated to current (2022 Sept 09) m-c.