<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Wed, May 18, 2016 at 5:56 AM, Martin J. O'Riordan <span dir="ltr"><<a href="mailto:martin.oriordan@movidius.com" target="_blank">martin.oriordan@movidius.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div lang="EN-IE" link="blue" vlink="purple"><div><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">Hi Rail,<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">We used a very simple pattern expansion (actually, not a pattern fragment). For example, for AND, ADD (horizontal sum), OR and XOR of 4 elements we use something like the following TableGen structure:<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">class HORIZ_Op4<SDNode opc, RegisterClass regVT, ValueType rt, ValueType vt, string asmstr> :<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">   SHAVE_Instr<(outs regVT:$dst), (ins VRF128:$src),<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                   !strconcat(asmstr, " $dst $src"),<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                   [(set regVT:$dst,<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                     (opc (rt (vector_extract(vt VRF128:$src), 0 ) ),<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                       (opc (rt (vector_extract(vt VRF128:$src), 1 ) ),<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                         (opc (rt (vector_extract(vt VRF128:$src), 2 ) ),<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                           (rt (vector_extract(vt VRF128:$src), 3 ) )<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                         )<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                       )<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                     )<u></u><u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">                   )]>;<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">This is okay for 4 element vectors, and it will get selected if the programmer writes something like:<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">vec[0] & vec[1] & vec[2] & vec[3]<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">but not with a simple variant like:<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal" style="margin-left:36.0pt"><span style="font-family:"Courier New";color:black">vec[0] & vec[2] & vec[1] & vec[3]<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">If this was properly represented by an ISD node, the other permutations could be more easily handled through normalisation. We “could†write patterns for each of the permutations, but it is verbose, and in practice most people only write it one way anyway.<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">The 8-lane equivalent has TableGen left thinking for quite a long time, and the 16-lane equivalent seems to hang TableGen.<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> <u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d">           MartinO<u></u><u></u></span></p><p class="MsoNormal"><span style="font-family:"Book Antiqua","serif";color:#1f497d"><u></u> </span></p></div></div></blockquote><div> </div><div>Just out of curiosity how do I specify multiple patterns for the same instruction? I tried putting the pattern in the pattern list but it LLVM complained that I can't set the same register multiple times.</div><div><br></div></div><div><br></div>-- <br><div class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div>Rail Shafigulin<br></div>Software Engineer <br>Esencia Technologies<br></div></div></div></div>
</div></div>