[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc
Subject:    Re: Asm volatile causing performance regressions on ARM
From:       Eric Botcazou <ebotcazou () adacore ! com>
Date:       2014-02-28 12:35:16
Message-ID: 1427919.0sXGq7JqJd () polaris
[Download RAW message or body]

> But here too the point is that we don't assume the same thing at the
> tree level or during register allocation.  It seems a bit silly for
> the scheduler to assume that all hard registers are clobbered when the
> register allocator itself doesn't assume that.  And most rtl passes
> assume that changes to pseudo registers are explicitly modelled via SETs
> or CLOBBERs.

OK, let's declare volatile asms dead as full optimization barriers then, which 
means that they will need to be taken out of volatile_insn_p eventually and 
the comments in the RTL middle-end adjusted.

In the short term, i.e. for 4.8.x/4.9.x, we cannot simply revert the patch 
because it has been written to catch UNSPEC_VOLATILE.  So we could devise 
another, temporary predicate in rtlanal.c that would be volatile_insn_p minus 
volatile asms and invoke it from the places in cse.c, cselib.c and dse.c which 
have been changed by the patch.

-- 
Eric Botcazou
[prev in list] [next in list] [prev in thread] [next in thread]