On 12/11/25 20:11, Richard Henderson wrote:
On 12/10/25 07:16, Paolo Bonzini wrote:
This is more efficient both when generating code and when testing
flags.
I guess sbb x,x appears quite frequently in x86 setcc computation, and
the testing of the flags is much less important than the straight line
code generation?
Yes. And to be honest, in the most common idioms generated for a modern
processor the whole computation ends up being dead, so it doesn't really
matter to have this vs. CC_OP_SBB or CC_OP_SUB. For example memcmp uses
it for "(x < y) ? -1 : 1":
subq %rcx, %rax
sbbl %eax, %eax
orl $1, %eax
and this is also common, for "(x < y) ? VALUE : 0"
subq %rcx, %rax
sbbq %rax, %rax ; could be sbbl :)
andl $0x1234, %eax
In old hand-written assembly it is used more creatively, and having
simpler generated code can matter if there are memory operations after
the sbb. I did this just because it's silly to compute both negsetcond
and setcond...
Paolo
