<div dir="ltr"><div>Thank you for review, Adam! I will add a comment.</div><div><br></div>Yes, actually we can convert VSELECT to AND on CPP, not in TD. <div>There is a case inside function PerformSELECTCombine, where VSELECT w/selector produced by SETCC considered for convertion to AND or OR nodes. I can teach this case to allow transform after our intrinsic lowering (Actually, I tried change primitively and it worked). I will prepare separate patch for this soon. </div></div><div class="gmail_extra"><br><div class="gmail_quote">2014-09-30 11:46 GMT+04:00 Demikhovsky, Elena <span dir="ltr"><<a href="mailto:elena.demikhovsky@intel.com" target="_blank">elena.demikhovsky@intel.com</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div lang="EN-US" link="blue" vlink="purple">
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">I agree with Adam. Can we convert VSELECT to AND in td?<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<div>
<p class="MsoNormal" style="margin-left:36.0pt">
<u></u><span style="font-family:"Calibri","sans-serif";color:#31849b"><span>-<span style="font:7.0pt "Times New Roman"">
</span></span></span><u></u><span dir="LTR"></span><b><i><span style="color:#31849b"> Elena<u></u><u></u></span></i></b></p>
</div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<div>
<div style="border:none;border-top:solid #b5c4df 1.0pt;padding:3.0pt 0cm 0cm 0cm">
<p class="MsoNormal"><b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif"">From:</span></b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif""> Adam Nemet [mailto:<a href="mailto:anemet@apple.com" target="_blank">anemet@apple.com</a>]
<br>
<b>Sent:</b> Monday, September 29, 2014 20:27<br>
<b>To:</b> Robert Khasanov<br>
<b>Cc:</b> Demikhovsky, Elena; <a href="mailto:llvm-commits@cs.uiuc.edu" target="_blank">llvm-commits@cs.uiuc.edu</a><br>
<b>Subject:</b> Re: [PATCH][AVX512] Enable intrinsics for vpcmpeq<u></u><u></u></span></p>
</div>
</div><div><div class="h5">
<p class="MsoNormal"><u></u> <u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div>
<p class="MsoNormal">On Sep 29, 2014, at 7:22 AM, Robert Khasanov <<a href="mailto:rob.khasanov@gmail.com" target="_blank">rob.khasanov@gmail.com</a>> wrote:<u></u><u></u></p>
</div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<p class="MsoNormal">Hi Adam,<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal">2014-09-28 11:14 GMT+04:00 Adam Nemet <<a href="mailto:anemet@apple.com" target="_blank">anemet@apple.com</a>>:<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div>
<p class="MsoNormal">On Sep 26, 2014, at 8:04 AM, Robert Khasanov <<a href="mailto:rob.khasanov@gmail.com" target="_blank">rob.khasanov@gmail.com</a>> wrote:<u></u><u></u></p>
</div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<p class="MsoNormal">Hi Adam,<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Thanks for review.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">I agree with you comment about reusing <span style="font-size:10.0pt;font-family:"Arial","sans-serif"">
getVectorMaskingNode. See new patches attached.</span><u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Is 003 the correct version now?<u></u><u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">yes. <u></u><u></u></p>
</div>
<blockquote style="border:none;border-left:solid #cccccc 1.0pt;padding:0cm 0cm 0cm 6.0pt;margin-left:4.8pt;margin-right:0cm">
<div>
<blockquote style="margin-top:5.0pt;margin-bottom:5.0pt">
<div>
<p class="MsoNormal">However, I don't understand your comment about ISD::INSERT_SUBVECTOR. Could you please clarify your suggestion?<u></u><u></u></p>
</div>
</blockquote>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">You’re making this op legal for v8i1, so we will now generate things like:<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal"> (v8i1 (insert_subvector (v8i1 foo), (v4i1 bar), (iPTR 0)))<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Do we have instructions to match these in the TD file?<u></u><u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</blockquote>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">We generate things like (v8i1 (insert_subvector undef, (v4i1 src), (iPTR 0))).<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">For this we have following rules in TD (see 003 patch):<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<div>
<p class="MsoNormal">+ def : Pat<(v8i1 (insert_subvector undef, (v4i1 VK4:$src), (iPTR 0))),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ (v8i1 (COPY_TO_REGCLASS VK4:$src, VK8))>;<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ def : Pat<(v8i1 (insert_subvector undef, (v2i1 VK2:$src), (iPTR 0))),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ (v8i1 (COPY_TO_REGCLASS VK2:$src, VK8))>;<u></u><u></u></p>
</div>
</div>
</div>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">OK, I missed this one.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">You still owe me :) the sample DAG in the comment to: <u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<blockquote style="margin-top:5.0pt;margin-bottom:5.0pt">
<div>
<p class="MsoNormal">+ SDValue Mask = Op.getOperand(3);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ Mask.getValueType().getSizeInBits());<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"> SDValue Cmp = DAG.getNode(IntrData->Opc0, dl, MaskVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"> Op.getOperand(1), Op.getOperand(2));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">- SDValue Res = getVectorMaskingNode(Cmp, Op.getOperand(3),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">- DAG.getTargetConstant(0, MaskVT), DAG);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ SDValue CmpMask = getVectorMaskingNode(Cmp, Op.getOperand(3),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getTargetConstant(0, MaskVT), DAG);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ SDValue Res = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, BitcastVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getUNDEF(BitcastVT), CmpMask,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getIntPtrConstant(0));<u></u><u></u></p>
</div>
</blockquote>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Other than that LGTM, thanks for splitting it up like this.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">One more question more of as a follow-on discussion. Do we actually have to generate AND for masking compares? I understand that that is what they get canonicalized into but perhaps we can still generate vselect uniformly.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Adam<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<div>
<div>
<div>
<p class="MsoNormal"> <u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Thanks,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Robert <u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<blockquote style="border:none;border-left:solid #cccccc 1.0pt;padding:0cm 0cm 0cm 6.0pt;margin-left:4.8pt;margin-right:0cm">
<div>
<div>
<p class="MsoNormal"><span style="color:#888888">Adam<u></u><u></u></span></p>
</div>
<div>
<div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<div>
<p class="MsoNormal">Thanks, <u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Robert <u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal">2014-09-26 4:19 GMT+04:00 Adam Nemet <<a href="mailto:anemet@apple.com" target="_blank">anemet@apple.com</a>>:<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div>
<div>
<p class="MsoNormal">On Sep 25, 2014, at 12:43 AM, Robert Khasanov <<a href="mailto:rob.khasanov@gmail.com" target="_blank">rob.khasanov@gmail.com</a>> wrote:<u></u><u></u></p>
</div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<div>
<p class="MsoNormal">The last patch is incorrect. See this version instead of previous one.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Sorry for that.<u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal">2014-09-25 11:40 GMT+04:00 Robert Khasanov <<a href="mailto:rob.khasanov@gmail.com" target="_blank">rob.khasanov@gmail.com</a>>:<u></u><u></u></p>
<div>
<p class="MsoNormal">Hi Elena, Adam,<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">In these patches I enable intrinsics for vpcmpeq{bwdq} instructions.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Since result of the instructions is mask, I enable new intrinsics type CMP_MASK and assume to enable CMP_MASK_CC in future for vpcmp{bwdq} instructions (due to additional CC argument).<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Also I extended argument types when intrinsics generated through TableGen (IIT_V64) to support 64 packed data.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">In last patch I enabled INSERT_SUBVECTOR for v8i1 to legalizer to support converting v2i1 and v4i1 to v8i1 and then bitcasting to i8.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Arial","sans-serif"">Please let me know if it looks good.</span><u></u><u></u></p>
</div>
</div>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
<div>
<p class="MsoNormal">Hi Robert,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<div>
<p class="MsoNormal">+ case CMP_MASK: {<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ EVT VT = Op.getOperand(1).getValueType();<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ VT.getVectorNumElements());<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ SDValue Cmp = DAG.getNode(IntrData->Opc0, dl, MaskVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ Op.getOperand(1), Op.getOperand(2));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ SDValue Res;<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ if (isAllOnes(Op.getOperand(3))) {<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ Res = Cmp;<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ } else {<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ Res = DAG.getNode(ISD::AND, dl, MaskVT, Cmp,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getNode(ISD::BITCAST, dl, MaskVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ Op.getOperand(3)));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ }<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ }<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">It would be good to reuse (and extend if needed) the helper getVectorMaskingNode for this. That should capture what it takes from a expression to become a masked expression in one place (just like AVX512_masking in td). Obviously the
BITCAST at the end does not belong there there but the information that masked compare is canonicalized to an AND rather than a VSELECT is useful.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">As you see I also don’t check for the mask value to be AllOnes in getVectorMaskingNode. We can add that for now if you need but I think that should be an orthogonal DAGCombiner transformation to cover more cases.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<div>
<p class="MsoNormal">+ setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v8i1, Legal);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Can you just enable it like that? I understand that you need it to lower the intrinsics but you may need to add some more patterns to TD to recognize this in the general case.<u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<div>
<p class="MsoNormal">- Res = DAG.getNode(ISD::AND, dl, MaskVT, Cmp,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">- DAG.getNode(ISD::BITCAST, dl, MaskVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">- Op.getOperand(3)));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ RMask = DAG.getNode(ISD::AND, dl, MaskVT, Cmp,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getNode(ISD::BITCAST, dl, BitcastVT, Mask),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getIntPtrConstant(0)));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"> }<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">- return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(),<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getNode(ISD::INSERT_SUBVECTOR, dl, BitcastVT,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getUNDEF(BitcastVT), RMask,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">+ DAG.getIntPtrConstant(0)));<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">This could use a comment with an example DAG that we’re trying to create.<u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Thanks,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">Adam<u></u><u></u></p>
</div>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<div>
<div>
<blockquote style="border:none;border-left:solid #cccccc 1.0pt;padding:0cm 0cm 0cm 6.0pt;margin-left:4.8pt;margin-right:0cm">
<div>
<div>
<p class="MsoNormal"><span style="color:#888888"><u></u> <u></u></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Arial","sans-serif";color:#888888">Robert</span><span style="color:#888888"><u></u><u></u></span></p>
</div>
</div>
</blockquote>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
</div>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</blockquote>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div></div></div>
<p>---------------------------------------------------------------------<br>
Intel Israel (74) Limited</p>
<p>This e-mail and any attachments may contain confidential material for<br>
the sole use of the intended recipient(s). Any review or distribution<br>
by others is strictly prohibited. If you are not the intended<br>
recipient, please contact the sender and delete all copies.</p></div>
</blockquote></div><br></div>