This is the mail archive of the
mailing list for the GCC project.
Re: [tree-ssa] Supefluous phi removal
- From: law at redhat dot com
- To: Richard Henderson <rth at redhat dot com>
- Cc: Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>, gcc-patches at gcc dot gnu dot org
- Date: Tue, 06 Jan 2004 12:41:11 -0700
- Subject: Re: [tree-ssa] Supefluous phi removal
- Reply-to: law at redhat dot com
In message <20040106192642.GB17687@redhat.com>, Richard Henderson writes:
>On Mon, Jan 05, 2004 at 06:39:19PM -0700, email@example.com wrote:
>> On a high level, don't you have to verify that you don't have dependencies
>> between PHI nodes before copy propagating values like this?
>We have code to solve this problem in the unssa pass already. It
>just needs to get reused here.
Actually, I don't think that's necessary. The serialization issues I
think are a result of turning PHIs into copies.
Zdenek's code never does that. If we can not fully propagate the RHS of
the PHI into all uses of the LHS of the PHI, then we simply leave the PHI