This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/82255] Vectorizer cost model overcounts cost of some vectorized loads
- From: "rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Wed, 20 Sep 2017 08:00:40 +0000
- Subject: [Bug tree-optimization/82255] Vectorizer cost model overcounts cost of some vectorized loads
- Auto-submitted: auto-generated
- References: <bug-82255-4@http.gcc.gnu.org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82255
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yeah, this is a known "missing feature" in the cost model. It's pessimistic
about the fancy way we can process both strided SLP loads and stores.
I didn't want to tackle this before trying to sort out the ugliness in SLP
vs. non-SLP costing (that we need two copies of all the cost compute in two
different places). I wanted to add sth like
STMT_VINFO_{PROLOGUE,BODY}_COST_VEC
and SLP_NODE_{PROLOGUE,BODY}_COST_VEC that can be filled from vectorizable_*
and just collected from the cost processing code.