<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: arial,helvetica,sans-serif; font-size: 10pt; color: #000000'><br><hr id="zwchr"><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><b>From: </b>"Dangeti Tharun kumar via llvm-dev" <llvm-dev@lists.llvm.org><br><b>To: </b>llvm-dev@lists.llvm.org<br><b>Cc: </b>"Santanu Das" <cs15mtech11018@iith.ac.in><br><b>Sent: </b>Sunday, October 9, 2016 12:09:01 PM<br><b>Subject: </b>[llvm-dev] On Loop Distribution pass<br><br><div dir="ltr">Dear community,<div><br></div><div>Our team at IITH have been experimenting with loop-distribution pass in LLVM. We see the following results on few benchmarks.</div><div><br></div><div><span id="gmail-docs-internal-guid-e7e6be9a-aa54-3006-2059-effa9157d485"><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; text-indent: 36pt;"><span style="font-size: 13.3333px; font-family: "courier new"; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">clang -O3 -mllvm -enable-loop-distribute -Rpass=loop-distribute file.c</span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; text-indent: 36pt;"><span style="font-size: 13.3333px; font-family: "courier new"; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><span style="font-size: 13.3333px; line-height: 18.4px;">clang -O3 -mllvm -enable-loop-distribute -Rpass-analysis=loop-distribute file.c</span><br></span></p><br><br><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; text-indent: 36pt;"><font face="monospace, monospace"><a href="http://crd.lbl.gov/departments/computer-science/PAR/research/previous-projects/torch-testbed/" style="text-decoration: none;" target="_blank"><span style="font-size: 16px; font-weight: 700; text-decoration: underline; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">TORCH</span></a><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">:</span></font></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font face="monospace, monospace">There are nearly 488 loops in this benchmark. LLVM was not able to distribute any loop.</font></span></p><font face="monospace, monospace"><br></font><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; text-indent: 36pt;"><font face="monospace, monospace"><a href="https://github.com/shantanuatiith/TSVC_" style="text-decoration: none;" target="_blank"><span style="font-size: 16px; font-weight: 700; text-decoration: underline; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">TSVC</span></a><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">:</span></font></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><font face="monospace, monospace"><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">There are 151 loops coded in plain ‘C’, </span><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">none</span><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"> of them got distributed. TSVC has particularly candidates valid for distribution like the one below. The inner loop in this example can be distributed.</span></font></p><font face="monospace, monospace"><br></font><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 36pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">for (int nl = 0; nl < ntimes/2; nl++) {</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">for (int i = 1; i < LEN; i++) {</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 108pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">a[i] += c[i] * d[i];</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 108pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">b[i] = b[i - 1] + a[i] + d[i];</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">}</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace">dummy(a, b, c, d, e, aa, bb, cc, 0.);</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 36pt; text-indent: 36pt;"><span style="font-size: 13.3333px; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font color="#444444" face="monospace, monospace"> }</font></span></p><font face="monospace, monospace"><br></font><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt;"><font face="monospace, monospace"><span style="font-size: 16px; vertical-align: baseline; background-color: transparent;"> <a href="http://vhosts.eecs.umich.edu/mibench//" style="font-weight: 700; white-space: pre-wrap; text-decoration: none;" target="_blank">MiBench</a></span><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">:</span></font></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap; background-color: transparent;"><font face="monospace, monospace">There are 6539 loops in MiBench. <b>None</b> of the loops were distributed by the loop-distribution pass of LLVM.</font></span></p><font face="monospace, monospace"><br></font><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt;"><font face="monospace, monospace"><span style="font-size: 16px; vertical-align: baseline; background-color: transparent;"> <a href="https://github.com/exmatex/CoMD" style="font-weight: 700; white-space: pre-wrap; text-decoration: none;" target="_blank">CoMD</a></span><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap; background-color: transparent;">:</span></font></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap;"><font face="monospace, monospace">CoMD is a reference implementation of typical classical molecular dynamics algorithms and workloads.</font></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><font face="monospace, monospace"><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap;">Out of 112 loops </span><span style="font-size: 16px; color: rgb(0, 0, 0); font-weight: 700; vertical-align: baseline; white-space: pre-wrap;">none</span><span style="font-size: 16px; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap;"> of them are distributed.</span></font></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><span style="font-size: 16px; font-family: arial; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap;"><br></span></p><p dir="ltr" style="line-height: 1.38; margin-top: 0pt; margin-bottom: 0pt; margin-left: 72pt;"><span style="font-size: 16px; font-family: arial; color: rgb(0, 0, 0); vertical-align: baseline; white-space: pre-wrap;"><br></span></p>For the specific loop I have shown above which is straight forward a good distribution candidate, the remark is "<b>loop not distributed: memory operations are safe for vectorization [-Rpass-analysis=loop-distribute]</b>". </span></div><div><span><br></span></div><div><span>Can someone reason these results and this remark?</span></div><div><span id="DWT390"><br></span></div></div></blockquote>The current loop distribution implementation is focused on enabling vectorization. The comment at the top of the file says:<br><br> // This file implements the Loop Distribution Pass. Its main focus is to<br> // distribute loops that cannot be vectorized due to dependence cycles. It<br> // tries to isolate the offending dependences into a new loop allowing<br> // vectorization of the remaining parts.<br><br>Hopefully, this helps explain the diagnostic. Regarding why the loop you highlight is not distributed to enable the vectorization of the a[i] update, I don't know. Adam?<br><br> -Hal<br><span id="gmail-docs-internal-guid-e7e6be9a-aa54-3006-2059-effa9157d485"></span><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div><span><br>I humbly request the community to correct me, if am missing something in my analysis.</span><div><br></div>-- <br><div class="gmail_signature"><div dir="ltr"><div><div dir="ltr">Thank you<div>D Tharun kumar</div><div>CS15MTECH11002</div><div>9948373970</div><div>CSE-IITH</div></div></div></div></div>
</div></div>
<br>_______________________________________________<br>LLVM Developers mailing list<br>llvm-dev@lists.llvm.org<br>http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev<br></blockquote><br><br><br>-- <br><div><span name="x"></span>Hal Finkel<br>Lead, Compiler Technology and Programming Languages<br>Leadership Computing Facility<br>Argonne National Laboratory<span name="x"></span><br></div></div></body></html>