Refine fast jit backend translation of IR lookupswitch #1264
Conversation
| bh_list_insert(jmp_info_list, node); | ||
|
|
||
| imm.setValue(INT32_MAX); | ||
| a.je(imm); |
There was a problem hiding this comment.
in another PR, after je to INT32_MAX, there is an error checking. Do we need it here?
There was a problem hiding this comment.
It is because that I don't want to write offset to the invalid place when asmjit error occurs. Normally we check the asm error in jit_codegen_gen_native, after generating machine code for each INSN, in other place, we can skip the check to reduce the footprint.
There was a problem hiding this comment.
but if the error was raised,
will be triggered?There was a problem hiding this comment.
Yes, it will be triggered, in the beginning of jit_codegen_gen_native, we set the error handler of code buffer:
wasm-micro-runtime/core/iwasm/fast-jit/cg/x86-64/jit_codegen_x86_64.cpp
Lines 6049 to 6051 in 7b72af3
And the JitErrorHandler is defined here:
wasm-micro-runtime/core/iwasm/fast-jit/cg/x86-64/jit_codegen_x86_64.cpp
Lines 221 to 234 in 7b72af3
Merge bytecodealliance:dev/fast_jit into wenyongh:opt_fast_jit_br_table_cg
Refine fast jit backend translation of IR lookupswitch (bytecodealliance#1264)
If the branches of lookupswitch is relatively big, lookup the dst
label address to jump from the address table but not compare
the value of branch one by one.