The old code resulted in one memory fetch for the test plus another one
for the return value (at the assembler level). The new code reuses the
already-loaded value:
- movl _ZZN11QMetaTypeIdI12QDBusMessageE14qt_metatype_idEvE11metatype_id(%rip), %edx
- testl %edx, %edx
- je .L158
movl _ZZN11QMetaTypeIdI12QDBusMessageE14qt_metatype_idEvE11metatype_id(%rip), %eax
- ret
+ testl %eax, %eax
+ je .L160
+ rep; ret
It also saves one load in the not-yet-initialised case:
-.L158:
+.L160:
leaq .LC7(%rip), %rdi
subq , %rsp
.cfi_def_cfa_offset 16
movq himBH1, %rsi
call _Z17qRegisterMetaTypeI12QDBusMessageEiPKcPT_
movl %eax, _ZZN11QMetaTypeIdI12QDBusMessageE14qt_metatype_idEvE11metatype_id(%rip)
- movl _ZZN11QMetaTypeIdI12QDBusMessageE14qt_metatype_idEvE11metatype_id(%rip), %eax
addq , %rsp
.cfi_def_cfa_offset 8
ret
Change-Id: I769950449822f2b1587680e05518be0a4f3120a2
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
56603a8d83