[llvm-dev] CTPOP and zeroes

ORiordan, Martin via llvm-dev llvm-dev at lists.llvm.org
Fri Jan 19 00:48:04 PST 2018


Thanks very much Eli, that is a good approach.  I assumed that this is the canonicalized form, but if not, I am sure I can add a simple pass to canonicalize other probable patterns to ensure a match.

All the best,

    MartinO

-----Original Message-----
From: Friedman, Eli [mailto:efriedma at codeaurora.org] 
Sent: Thursday, January 18, 2018 7:02 PM
To: ORiordan, Martin <martin.oriordan at intel.com>; llvm-dev <llvm-dev at lists.llvm.org>
Subject: Re: [llvm-dev] CTPOP and zeroes

On 1/18/2018 12:57 AM, ORiordan, Martin via llvm-dev wrote:
> Quick question.  The 'ISD::CTPOP' node allows a target to lower the counting of ones in a word to a single instruction.  Our target also has an instruction for counting the zeroes in a word.  Does CTPOP support counting of zeroes as well as ones instead of doing either "CTPOP(INVERT(operand))" or "N-bits - CTPOP(operand)"?

There is no target-independent ISD opcode for the operation you're describing (as far as I know, no in-tree target has an instruction like that).  I would suggest handling this using pattern-matching in your backend, e.g. "def : Pat<(i32 (ctpop (not i32:$a))), (INVCTPOP R:$a)>;".

-Eli

--
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

--------------------------------------------------------------
Intel Research and Development Ireland Limited
Registered in Ireland
Registered Office: Collinstown Industrial Park, Leixlip, County Kildare
Registered Number: 308263


This e-mail and any attachments may contain confidential material for the sole
use of the intended recipient(s). Any review or distribution by others is
strictly prohibited. If you are not the intended recipient, please contact the
sender and delete all copies.


More information about the llvm-dev mailing list