On Fri, Jan 13, 2012 at 10:00 AM, Eric Botcazou <ebotca...@adacore.com> wrote: >> Yes, it is a complete solution. Tracking register liveness is >> different issue, and a register is indeed dead after instruction, if >> it has been clobbered by insn, or when marked unused. > > My concern was liveness within the new sequence of instructions: suppose you > have > > (set (reg:M x) (...)) REG_UNUSED (reg:M x) > > and we "peephole" the instruction. Is the live range of (reg:M x) properly > extended within the new sequence of instructions?
The patch doesn't change this functionality, but: (match_scratch:M 1 "r") (set (reg:M x) (...)) REG_UNUSED (reg:M x) (match_dup 1) Previously, register x would be allocated as a scracth register, but now it isn't. Regarding the liveness analysis of a new sequence - this functionality is not changed at all. As said, the new code only prevents x to be allocated as a scratch that must live up to the (match_dup 1). >> The problem my patch solves is the answer to the question "Is the >> choosen non-live temporary register untouched over the insn >> sequence?". The answer: "Yes, if it was not set or clobbered by any >> insn in the sequence". > > Are you sure that you don't need to do this in addition to the existing test, > instead of in lieu of the existing test? Yes, because new test ALWAYS includes the registers that were wrongly marked as dead by previous test due to REG_UNUSED and noclobber processing. Uros.
