<div dir="ltr"><div class="gmail_quote"><div dir="ltr">On Mon, Dec 18, 2017 at 7:10 PM Xinliang David Li <<a href="mailto:davidxl@google.com">davidxl@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Mon, Dec 18, 2017 at 6:03 PM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@google.com" target="_blank">chandlerc@google.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><span><div dir="ltr">On Mon, Dec 18, 2017 at 5:46 PM Xinliang David Li <<a href="mailto:davidxl@google.com" target="_blank">davidxl@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">The introduction of cleanup.cond block in b.ll without loop-rotation already makes the layout worse than a.ll.<div><br></div><div><br></div><div>Without introducing cleanup.cond block, the layout out is</div><div><br></div><div>entry->while.cond -> while.body->ret </div><div><br></div><div>All the arrows are hot fall through edges which is good.</div><div><br></div><div>With cleanup.cond introduced in b.ll, the layout without tailDup nor loop rotation looks like:</div><div><br></div><div><br></div><div>entry->while.cond ->while.body->cleanup.cond,  ret</div><div><br></div><div>Note that now there is no fall through edge to 'ret' block and  while.body now needs to explicitly branch to 'ret'.   If loop rotation happens, we will have</div><div><br></div><div>entry, cleanup.cond -> while.cond -> while.body->ret</div><div><br></div><div>Not that there will be a hot branch from entry to while.cond.</div><div><br></div><div>LLVM actually does both tail dup and loop rotation. And the layout looks like:</div><div><br></div><div>entry --> while.cond.dup, cleanup.cond->while.cond -> while.body->ret</div><div><br></div><div>this is better than the previous one, but there is still a hot branch from while.cond.dup to while.body introduced.</div></div></blockquote><div> </div></span><div>(just relaying a comment I made in person)</div><div><br></div><div>This does seem like a less good layout, but as a consequence of it we do avoid one addition and comparison along the hot path. It's not obvious to me which is better: the better layout (with added instruction) or the minimal set of instructions with less good layout.</div><div><div class="m_2061202238979199105h5"><div> </div></div></div></div></div></blockquote><div><br></div></div></div></div><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>If the loop has high number of iterations, i.e., the while.body->ret path is less likely taken, then sinking the increment into a split block brings very little performance gain.  For short trip-counted loops, the sinking may seem beneficial, but it is at the cost of a new taken branch for every execution of the loop exit path -- so the overall performance gain for splitting a new block is also questionable -- and in this case, actually hurting performance.</div></div></div></div></blockquote><div><br></div><div>Do you see a good way to model this in SimplifyCFG? (Which is where I think this is occuring...) </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><br></div><div><br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div><div class="m_2061202238979199105h5"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div><br></div><div>David</div></div><div class="gmail_extra"><div class="gmail_quote">On Mon, Dec 18, 2017 at 4:14 PM, Wei Mi <span dir="ltr"><<a href="mailto:wmi@google.com" target="_blank">wmi@google.com</a>></span> wrote:<br></div></div><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">Hi,<br><br>Recently 10% performance regression on an important benchmark showed up after we integrated <a href="https://reviews.llvm.org/rL318299" target="_blank">https://reviews.llvm.org/rL318299</a>. The analysis showed that rL318299 triggered loop rotation on an multi exits loop, and the loop rotation introduced code layout issue. The performance regression is a side-effect of rL318299. I got two testcases a.ll and b.ll attached to illustrate the problem. a.ll was generated by rL318298 and b.ll was generated by rL318299.<div><br><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">-------------------------- a.ll ----------------------------</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">declare void @_Z1fv() local_unnamed_addr #2</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">@i = global i8 0, align 1</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">define i8* @_Z1gPcS_S_(i8* nocapture readonly %d, i8* %h, i8* readnone returned %p3) local_unnamed_addr #3 {</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">entry:</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  br label %while.cond</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">while.cond:                                       ; preds = %while.body, %entry</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %h.addr.0 = phi i8* [ %h, %entry ], [ %add.ptr4, %while.body ]</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %d.addr.0 = phi i8* [ %d, %entry ], [ %add.ptr3, %while.body ]</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %cmp = icmp ugt i8* %h.addr.0, @i</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  br i1 %cmp, label %while.end, label %while.body</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">while.body:                                       ; preds = %while.cond</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %0 = bitcast i8* %d.addr.0 to i64*</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %1 = load i64, i64* %0, align 1</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %2 = bitcast i8* %h.addr.0 to i64*</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  store i64 %1, i64* %2, align 1</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %add.ptr = getelementptr inbounds i8, i8* %d.addr.0, i64 8</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %3 = bitcast i8* %add.ptr to i64*</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %4 = load i64, i64* %3, align 1</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  store i64 %4, i64* %2, align 1</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %add.ptr3 = getelementptr inbounds i8, i8* %d.addr.0, i64 6</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %add.ptr4 = getelementptr inbounds i8, i8* %h.addr.0, i64 6</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  %cmp5 = icmp ult i8* %add.ptr4, %p3</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  br i1 %cmp5, label %while.cond, label %return</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">while.end:                                        ; preds = %while.cond</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  tail call void @_Z1fv()</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  unreachable</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">return:                                           ; preds = %while.body</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">  ret i8* %p3</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">}</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br><br>-------------------------- b.ll ----------------------------<br>declare void @_Z1fv() local_unnamed_addr #2<br>@i = global i8 0, align 1<br><br>define i8* @_Z1gPcS_S_(i8* nocapture readonly %d, i8* %h, i8* readnone returned %p3) local_unnamed_addr #3 {<br>entry:<br>  br label %while.cond<br><br>while.cond:                                       ; preds = %cleanup.cont, %entry<br>  %h.addr.0 = phi i8* [ %h, %entry ], [ %add.ptr4, %cleanup.cont ]<br>  %d.addr.0 = phi i8* [ %d, %entry ], [ %add.ptr3, %cleanup.cont ]<br>  %cmp = icmp ugt i8* %h.addr.0, @i<br>  br i1 %cmp, label %while.end, label %while.body<br><br>while.body:                                       ; preds = %while.cond<br>  %0 = bitcast i8* %d.addr.0 to i64*<br>  %1 = load i64, i64* %0, align 1<br>  %2 = bitcast i8* %h.addr.0 to i64*<br>  store i64 %1, i64* %2, align 1<br>  %add.ptr = getelementptr inbounds i8, i8* %d.addr.0, i64 8<br>  %3 = bitcast i8* %add.ptr to i64*<br>  %4 = load i64, i64* %3, align 1<br>  store i64 %4, i64* %2, align 1<br>  %add.ptr4 = getelementptr inbounds i8, i8* %h.addr.0, i64 6<br>  %cmp5 = icmp ult i8* %add.ptr4, %p3<br>  br i1 %cmp5, label %cleanup.cont, label %return<br><br>cleanup.cont:                                     ; preds = %while.body<br>  %add.ptr3 = getelementptr inbounds i8, i8* %d.addr.0, i64 6<br>  br label %while.cond<br><br>while.end:                                        ; preds = %while.cond<br>  tail call void @_Z1fv()<br>  unreachable<br><br>return:                                           ; preds = %while.body<br>  ret i8* %p3<br>}<br><br>The only difference between a.ll and b.ll is the basicblock cleanup.cont. <br><br>-------------------------- a.ll after loop rotate, same as a.ll before loop rotate ----------------------------<br>~/workarea/llvm-r318298/dbuild/bin/opt -loop-rotate -S < a.ll<br><br>; ModuleID = '<stdin>'<br>source_filename = "<stdin>"<br><br>@i = global i8 0, align 1<br><br>declare void @_Z1fv() local_unnamed_addr<br><br>define i8* @_Z1gPcS_S_(i8* nocapture readonly %d, i8* %h, i8* readnone returned %p3) local_unnamed_addr {<br>entry:<br>  br label %while.cond<br><br>while.cond:                                       ; preds = %while.body, %entry<br>  %h.addr.0 = phi i8* [ %h, %entry ], [ %add.ptr4, %while.body ]<br>  %d.addr.0 = phi i8* [ %d, %entry ], [ %add.ptr3, %while.body ]<br>  %cmp = icmp ugt i8* %h.addr.0, @i<br>  br i1 %cmp, label %while.end, label %while.body<br><br>while.body:                                       ; preds = %while.cond<br>  %0 = bitcast i8* %d.addr.0 to i64*<br>  %1 = load i64, i64* %0, align 1<br>  %2 = bitcast i8* %h.addr.0 to i64*<br>  store i64 %1, i64* %2, align 1<br>  %add.ptr = getelementptr inbounds i8, i8* %d.addr.0, i64 8<br>  %3 = bitcast i8* %add.ptr to i64*<br>  %4 = load i64, i64* %3, align 1<br>  store i64 %4, i64* %2, align 1<br>  %add.ptr3 = getelementptr inbounds i8, i8* %d.addr.0, i64 6<br>  %add.ptr4 = getelementptr inbounds i8, i8* %h.addr.0, i64 6<br>  %cmp5 = icmp ult i8* %add.ptr4, %p3<br>  br i1 %cmp5, label %while.cond, label %return<br><br>while.end:                                        ; preds = %while.cond<br>  tail call void @_Z1fv()<br>  unreachable<br><br>return:                                           ; preds = %while.body<br>  ret i8* %p3<br>}<br><br>-------------------------- b.ll after loop rotate ----------------------------</div><div><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">~/workarea/llvm-r318298/dbuild/bin/opt -loop-rotate -S < b.ll</span></div><div><br>@i = global i8 0, align 1<br>declare void @_Z1fv() local_unnamed_addr<br><br>define i8* @_Z1gPcS_S_(i8* nocapture readonly %d, i8* %h, i8* readnone returned %p3) local_unnamed_addr {<br>entry:<br>  %cmp1 = icmp ugt i8* %h, @i<br>  br i1 %cmp1, label %while.end, label %<a href="http://while.body.lr.ph" target="_blank">while.body.lr.ph</a><br><br><a href="http://while.body.lr.ph" target="_blank">while.body.lr.ph</a>:                                 ; preds = %entry<br>  br label %while.body<br><br>while.cond:                                       ; preds = %while.body<br>  %h.addr.0 = phi i8* [ %add.ptr4, %while.body ]<br>  %d.addr.0 = phi i8* [ %add.ptr3, %while.body ]<br>  %cmp = icmp ugt i8* %h.addr.0, @i<br>  br i1 %cmp, label %while.cond.while.end_crit_edge, label %while.body<br><br>while.body:                                       ; preds = %<a href="http://while.body.lr.ph" target="_blank">while.body.lr.ph</a>, %while.cond<br>  %d.addr.03 = phi i8* [ %d, %<a href="http://while.body.lr.ph" target="_blank">while.body.lr.ph</a> ], [ %d.addr.0, %while.cond ]<br>  %h.addr.02 = phi i8* [ %h, %<a href="http://while.body.lr.ph" target="_blank">while.body.lr.ph</a> ], [ %h.addr.0, %while.cond ]<br>  %0 = bitcast i8* %d.addr.03 to i64*<br>  %1 = load i64, i64* %0, align 1<br>  %2 = bitcast i8* %h.addr.02 to i64*<br>  store i64 %1, i64* %2, align 1<br>  %add.ptr = getelementptr inbounds i8, i8* %d.addr.03, i64 8<br>  %3 = bitcast i8* %add.ptr to i64*<br>  %4 = load i64, i64* %3, align 1<br>  store i64 %4, i64* %2, align 1<br>  %add.ptr4 = getelementptr inbounds i8, i8* %h.addr.02, i64 6<br>  %cmp5 = icmp ult i8* %add.ptr4, %p3<br>  %add.ptr3 = getelementptr inbounds i8, i8* %d.addr.03, i64 6<br>  br i1 %cmp5, label %while.cond, label %return<br><br>while.cond.while.end_crit_edge:                   ; preds = %while.cond<br>  br label %while.end<br><br>while.end:                                        ; preds = %while.cond.while.end_crit_edge, %entry<br>  tail call void @_Z1fv()<br>  unreachable<br><br>return:                                           ; preds = %while.body<br>  ret i8* %p3<br>}<br><br>a.ll and b.ll have different results after loop rotation because of <a href="https://www.google.com/url?q=http://llvm.org/viewvc/llvm-project?view%3Drevision%26revision%3D181230&sa=D&usg=AFQjCNHIQDnlfGByPF-pvV991MH72_ExNg" rel="noreferrer nofollow" style="margin:0px;border:0px;outline:0px;font-weight:400;font-style:normal;font-size:13px;font-family:Arial,Helvetica,sans-serif;vertical-align:baseline;text-decoration:none;color:rgb(221,75,57);font-variant-ligatures:normal;font-variant-caps:normal;letter-spacing:normal;text-align:left;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255)" target="_blank">http://llvm.org/viewvc/llvm-project?view=revision&revision=181230</a><div><br><div><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">-------------------------- a.s generated from a.ll ----------------------------</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">~/workarea/llvm-r318298/dbuild/bin/opt -loop-rotate -S < a.ll |~/workarea/llvm-r318298/dbuild/bin/llc</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">       </span>.cfi_startproc</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"># BB#0:                                 # %entry</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">     </span>pushq<span style="white-space:pre-wrap">   </span>%rax</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">      </span>.cfi_def_cfa_offset 16</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">    </span>movl<span style="white-space:pre-wrap">    </span>$i, %eax</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">  </span>.p2align<span style="white-space:pre-wrap">        </span>4, 0x90</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">.LBB0_1:                                # %while.cond</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">                                        # =>This Inner Loop Header: Depth=1</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">        </span>cmpq<span style="white-space:pre-wrap">    </span>%rax, %rsi</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">        </span>ja<span style="white-space:pre-wrap">      </span>.LBB0_4</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"># BB#2:                                 # %while.body</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">                                        #   in Loop: Header=BB0_1 Depth=1</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">  </span>movq<span style="white-space:pre-wrap">    </span>(%rdi), %rcx</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">      </span>movq<span style="white-space:pre-wrap">    </span>%rcx, (%rsi)</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">      </span>movq<span style="white-space:pre-wrap">    </span>8(%rdi), %rcx</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">     </span>movq<span style="white-space:pre-wrap">    </span>%rcx, (%rsi)</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">      </span>addq<span style="white-space:pre-wrap">    </span>$6, %rdi</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">  </span>addq<span style="white-space:pre-wrap">    </span>$6, %rsi</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">  </span>cmpq<span style="white-space:pre-wrap">    </span>%rdx, %rsi</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">        </span>jb<span style="white-space:pre-wrap">      </span>.LBB0_1</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"># BB#3:                                 # %return</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">   </span>movq<span style="white-space:pre-wrap">    </span>%rdx, %rax</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">        </span>popq<span style="white-space:pre-wrap">    </span>%rcx</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">      </span>retq</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">.LBB0_4:                                # %while.end</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">    </span>callq<span style="white-space:pre-wrap">   </span>_Z1fv</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">.Lfunc_end0:</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">   </span>.size<span style="white-space:pre-wrap">   </span>_Z1gPcS_S_, .Lfunc_end0-_Z1gPcS_S_</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="white-space:pre-wrap">        </span>.cfi_endproc</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">call <span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">_Z1fv is unreachable. </span>Suppose loop LBB0_1 has few iterations, a.s will contain mostly fall through branches.</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><br></div><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">-------------------------- b.s generated from b.ll ----------------------------</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">~/workarea/llvm-r318298/dbuild/bin/opt -loop-rotate -S < b.ll |~/workarea/llvm-r318298/dbuild/bin/llc</span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><div><span style="white-space:pre-wrap">     </span>.cfi_startproc</div><div># BB#0:                                 # %entry</div><div><span style="white-space:pre-wrap">   </span>pushq<span style="white-space:pre-wrap">   </span>%rax</div><div><span style="white-space:pre-wrap">     </span>.cfi_def_cfa_offset 16</div><div><span style="white-space:pre-wrap">   </span>movl<span style="white-space:pre-wrap">    </span>$i, %eax</div><div><span style="white-space:pre-wrap"> </span>cmpq<span style="white-space:pre-wrap">    </span>%rax, %rsi</div><div><span style="white-space:pre-wrap">       </span>ja<span style="white-space:pre-wrap">      </span>.LBB0_5</div><div># BB#1:</div><div><span style="white-space:pre-wrap">    </span>movl<span style="white-space:pre-wrap">    </span>$i, %eax</div><div><span style="white-space:pre-wrap"> </span>.p2align<span style="white-space:pre-wrap">        </span>4, 0x90</div><div>.LBB0_3:                                # %while.body</div><div>                                        # =>This Inner Loop Header: Depth=1</div><div><span style="white-space:pre-wrap">     </span>movq<span style="white-space:pre-wrap">    </span>(%rdi), %rcx</div><div><span style="white-space:pre-wrap">     </span>movq<span style="white-space:pre-wrap">    </span>%rcx, (%rsi)</div><div><span style="white-space:pre-wrap">     </span>movq<span style="white-space:pre-wrap">    </span>8(%rdi), %rcx</div><div><span style="white-space:pre-wrap">    </span>movq<span style="white-space:pre-wrap">    </span>%rcx, (%rsi)</div><div><span style="white-space:pre-wrap">     </span>addq<span style="white-space:pre-wrap">    </span>$6, %rsi</div><div><span style="white-space:pre-wrap"> </span>cmpq<span style="white-space:pre-wrap">    </span>%rdx, %rsi</div><div><span style="white-space:pre-wrap">       </span>jae<span style="white-space:pre-wrap">     </span>.LBB0_4</div><div># BB#2:                                 # %while.cond</div><div>                                        #   in Loop: Header=BB0_3 Depth=1</div><div><span style="white-space:pre-wrap">       </span>addq<span style="white-space:pre-wrap">    </span>$6, %rdi</div><div><span style="white-space:pre-wrap"> </span>cmpq<span style="white-space:pre-wrap">    </span>%rax, %rsi</div><div><span style="white-space:pre-wrap">       </span>jbe<span style="white-space:pre-wrap">     </span>.LBB0_3</div><div>.LBB0_5:                                # %while.end</div><div><span style="white-space:pre-wrap">       </span>callq<span style="white-space:pre-wrap">   </span>_Z1fv</div><div>.LBB0_4:                                # %return</div><div><span style="white-space:pre-wrap">    </span>movq<span style="white-space:pre-wrap">    </span>%rdx, %rax</div><div><span style="white-space:pre-wrap">       </span>popq<span style="white-space:pre-wrap">    </span>%rcx</div><div><span style="white-space:pre-wrap">     </span>retq</div><div>.Lfunc_end0:</div><div><span style="white-space:pre-wrap">  </span>.size<span style="white-space:pre-wrap">   </span>_Z1gPcS_S_, .Lfunc_end0-_Z1gPcS_S_</div><div><span style="white-space:pre-wrap">       </span>.cfi_endproc</div></div><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">call </span><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;background-color:rgb(255,255,255);float:none;display:inline">_Z1fv is unreachable. </span>Here we have "<span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">jae .</span><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">LBB0_4" which will not fall through when exiting the loop. The non-fall-through branch increases branch misses significantly and regresses the benchmark by 10%. </span></span><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial"><br style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial">Now a possible way to fix it is to duplicate basicblock .LBB0_3, just like what tail duplication does, but basicblock .LBB0_3 contains 7 instructions and it will introduce some cost of code size increase.</div><div><br></div><div>Any suggestion to fix the issue are welcomed! </div></div></div></div>
</blockquote></div></div></blockquote></div></div></div></div>
</blockquote></div></div></div></blockquote></div></div>