Re: [dataflow] Provide a blockage insn for targets that don't have it

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

Ian Lance Taylor wrote:
> Paolo Bonzini <paolo.bonzini@lu.unisi.ch> writes:
>
> > This patch provides a gen_blockage routine for targets that don't have
> > a blockage insn.  This is implemented as a dummy `asm ("")'
> > instruction. Unlike previous code that used ASM_INPUT in this way,
> > this instruction is made volatile so that it is not deleted by the DCE
> > pass.
>
> I don't have any particular objections to this patch, but does anybody
> know why we have machine specific blockage insns?  Isn't blockage a
> relatively machine independent concept?

Yes, in fact these blockage insns are all basically an unspec_volatile. 
 But then, asm ("") is a poorer replacement for that than an `asm 
volatile (""::)'.

Let me try making a mainline patch that removes the blockage insns 
completely then.

Paolo