<div dir="ltr"><div class="gmail_quote"><div dir="ltr">On Fri, Dec 8, 2017 at 11:07 PM Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</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>This is taking 'ready, fire, aim' too far IMO.<br></div><div><br></div><div>Reverting this commit won't prevent hitting the bug.</div></div></blockquote><div><br></div><div>Er, in fact, it will in one important case: a particular version of ffmpeg is impossible to compile with top-of-tree at the moment. Reverting will allow it to compile.</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> As I wrote in the bug report after I posted here, it only takes a 1-liner IR test to show the DAG bug. <br>And as David Blakie wrote and provided another example: "Interestingly, a nearby test case seems to fail with or without the patch". And yes, the real root cause of that hang is the same as the first example.<br></div><div><br></div><div>If it's urgent to prevent the potential hang, then you probably want to revert this:<br><a href="https://reviews.llvm.org/rL311083" target="_blank">https://reviews.llvm.org/rL311083</a> (ping Simon)<br></div></div></blockquote><div><br></div><div>I don't have any strong opinion about which patch gets reverted, but I *do* think we should revert to green here one way or the other.</div><div><br></div><div>I don't think the fact that there exist other test cases that wouldn't be fixed by a particular revert is terribly relevant: we have a *lot* of crasher and inf-loop fuzzer-found test cases, and I'm not arguing we should necessarily prioritize all of them.</div><div><br></div><div>The *only* reason I think that this is worth some revert is because we have a real world example of code that compiled before and now dosen't compile. The test cases are important to make sure folks can debug things, but in and of themselves I think are not enough to justify reverting.</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>But I'm working on a DAGCombiner fix to prevent the conflict between the DAG fold and the x86 fold.<br></div></div></blockquote><div><br></div><div>Awesome, this is clearly good to do regardless.</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></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Dec 8, 2017 at 2:39 PM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@gmail.com" target="_blank">chandlerc@gmail.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">I do think its reasonable to revert... We hit this in real code in the wild, and so others may as well. I think its worth reverting until we get it fixed as it seems unlikely to be a huge delay.</div><div class="m_-5216899527763084458HOEnZb"><div class="m_-5216899527763084458h5"><br><div class="gmail_quote"><div dir="ltr">On Fri, Dec 8, 2017 at 3:34 PM Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</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><div>Filed here:<br><a href="https://bugs.llvm.org/show_bug.cgi?id=35579" target="_blank">https://bugs.llvm.org/show_bug.cgi?id=35579</a><br><br></div>Looks like it's stuck toggling between mul and shl nodes. :)<br><br></div>I'm still not sure what the policy is for this kind of situation. Clearly, the bug exists independent of this commit, but if you think this commit makes the bug much more likely to be hit, then we still want to revert?<br></div><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Dec 8, 2017 at 7:20 AM, Sanjay Patel <span dir="ltr"><<a href="mailto:spatel@rotateright.com" target="_blank">spatel@rotateright.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">Just started looking at the example - this hangs somewhere in the backend, not in instcombine (IR below). Does that affect the decision about reverting?<br><br>$ ./llc -o - inf.ll -mtriple=x86_64<br> .text<br> .file "inf.c"<br>^C<br><br>$ cat inf.ll<br>target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"<br>target triple = "x86_64--linux-gnu"<br><br>%struct.c = type { i32, [0 x i8] }<br><br>@d = common local_unnamed_addr global i32 0, align 4<br>@llvm.global_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @msan.module_ctor, i8* null }]<br>@__msan_retval_tls = external thread_local(initialexec) global [100 x i64]<br>@__msan_retval_origin_tls = external thread_local(initialexec) global i32<br>@__msan_param_tls = external thread_local(initialexec) global [100 x i64]<br>@__msan_param_origin_tls = external thread_local(initialexec) global [200 x i32]<br>@__msan_va_arg_tls = external thread_local(initialexec) global [100 x i64]<br>@__msan_va_arg_overflow_size_tls = external thread_local(initialexec) global i64<br>@__msan_origin_tls = external thread_local(initialexec) global i32<br><br>; Function Attrs: norecurse nounwind sanitize_memory<br>define i32 @e(%struct.c* %f) local_unnamed_addr #0 {<br>entry:<br> %0 = load i64, i64* getelementptr inbounds ([100 x i64], [100 x i64]* @__msan_param_tls, i64 0, i64 0), align 8<br> %1 = ptrtoint %struct.c* %f to i64<br> %_msprop = trunc i64 %0 to i32<br> %2 = trunc <a href="https://maps.google.com/?q=i64+%251&entry=gmail&source=g" target="_blank">i64 %1</a> to i32<br> store i32 0, i32* inttoptr (i64 xor (i64 ptrtoint (i32* @d to i64), i64 87960930222080) to i32*), align 4<br> store i32 0, i32* @d, align 4, !tbaa !2<br> %3 = icmp eq i32 %_msprop, 0<br> br i1 %3, label %5, label %4, !prof !6<br><br>; <label>:4: ; preds = %entry<br> call void @__msan_warning_noreturn() #1<br> call void asm sideeffect "", ""() #1<br> unreachable<br><br>; <label>:5: ; preds = %entry<br> %cmp9 = icmp sgt i32 %2, 0<br> br i1 %cmp9, label %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a>, label %for.end<br><br><a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a>: ; preds = %5<br> %_msprop23 = trunc i64 %0 to i8<br> %6 = trunc <a href="https://maps.google.com/?q=i64+%251&entry=gmail&source=g" target="_blank">i64 %1</a> to i8<br> %arrayidx.phi.trans.insert = getelementptr inbounds %struct.c, %struct.c* %f, i64 0, <a href="https://maps.google.com/?q=i32+1,+i64&entry=gmail&source=g" target="_blank">i32 1, i64</a> 0<br> %_mscmp54 = icmp eq i64 %0, 0<br> br i1 %_mscmp54, label %8, label %7, !prof !6<br><br>; <label>:7: ; preds = %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a><br> call void @__msan_warning_noreturn() #1<br> call void asm sideeffect "", ""() #1<br> unreachable<br><br>; <label>:8: ; preds = %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a><br> %.pre = load i8, i8* %arrayidx.phi.trans.insert, align 1, !tbaa !7<br> %9 = ptrtoint i8* %arrayidx.phi.trans.insert to i64<br> %10 = xor i64 %9, 87960930222080<br> %11 = inttoptr i64 %10 to i8*<br> %_msld = load i8, i8* %11, align 1<br> %wide.trip.count = and <a href="https://maps.google.com/?q=i64+%251&entry=gmail&source=g" target="_blank">i64 %1</a>, 4294967295<br> br i1 false, label %12, label %13, !prof !8<br><br>; <label>:12: ; preds = %8<br> unreachable<br><br>; <label>:13: ; preds = %8<br> %14 = icmp ult i64 %wide.trip.count, 32<br> br i1 %14, label %for.body.preheader, label %<a href="http://vector.ph" target="_blank">vector.ph</a><br><br><a href="http://vector.ph" target="_blank">vector.ph</a>: ; preds = %13<br> %n.mod.vf = and <a href="https://maps.google.com/?q=i64+%251,+31&entry=gmail&source=g" target="_blank">i64 %1, 31</a><br> %n.vec = sub nsw i64 %wide.trip.count, %n.mod.vf<br> %cast.crd = trunc i64 %n.vec to i8<br> %15 = mul i8 %cast.crd, %6<br> %_msprop33 = or i8 %_msld, %_msprop23<br> %ind.end = add i8 %.pre, %15<br> %_msprop34 = insertelement <16 x i8> undef, i8 %_msld, i32 0<br> %.splatinsert = insertelement <16 x i8> undef, i8 %.pre, i32 0<br> %_msprop35 = shufflevector <16 x i8> %_msprop34, <16 x i8> undef, <16 x i32> zeroinitializer<br> %.splat = shufflevector <16 x i8> %.splatinsert, <16 x i8> undef, <16 x i32> zeroinitializer<br> %_msprop36 = insertelement <16 x i8> undef, i8 %_msprop23, i32 0<br> %.splatinsert12 = insertelement <16 x i8> undef, i8 %6, i32 0<br> %_msprop37 = shufflevector <16 x i8> %_msprop36, <16 x i8> undef, <16 x i32> zeroinitializer<br> %.splat13 = shufflevector <16 x i8> %.splatinsert12, <16 x i8> undef, <16 x i32> zeroinitializer<br> %msprop_mul_cst = mul <16 x i8> %_msprop37, <i8 0, i8 1, i8 2, i8 1, i8 4, i8 1, i8 2, i8 1, i8 8, i8 1, i8 2, i8 1, i8 4, i8 1, i8 2, i8 1><br> %16 = mul <16 x i8> %.splat13, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15><br> %_msprop38 = or <16 x i8> %_msprop35, %msprop_mul_cst<br> %induction = add <16 x i8> %.splat, %16<br> %17 = shl i8 %_msprop23, 4<br> %18 = shl i8 %6, 4<br> %_msprop39 = insertelement <16 x i8> undef, i8 %17, i32 0<br> %.splatinsert14 = insertelement <16 x i8> undef, i8 %18, i32 0<br> %_msprop40 = shufflevector <16 x i8> %_msprop39, <16 x i8> undef, <16 x i32> zeroinitializer<br> %.splat15 = shufflevector <16 x i8> %.splatinsert14, <16 x i8> undef, <16 x i32> zeroinitializer<br> br label %vector.body<br><br>vector.body: ; preds = %37, %<a href="http://vector.ph" target="_blank">vector.ph</a><br> %index = phi i64 [ 0, %<a href="http://vector.ph" target="_blank">vector.ph</a> ], [ %index.next, %37 ]<br> %_msphi_s46 = phi <16 x i8> [ %_msprop38, %<a href="http://vector.ph" target="_blank">vector.ph</a> ], [ %_msprop47, %37 ]<br> %vec.ind = phi <16 x i8> [ %induction, %<a href="http://vector.ph" target="_blank">vector.ph</a> ], [ %vec.ind.next, %37 ]<br> %_msprop47 = or <16 x i8> %_msphi_s46, %_msprop40<br> %step.add = add <16 x i8> %vec.ind, %.splat15<br> %_msprop49 = or <16 x i8> %_msprop47, %_msprop37<br> %19 = add <16 x i8> %step.add, %.splat13<br> %20 = or i64 %index, 1<br> %21 = getelementptr inbounds %struct.c, %struct.c* %f, i64 0, <a href="https://maps.google.com/?q=i32+1,+i64&entry=gmail&source=g" target="_blank">i32 1, i64</a> %20<br> br i1 false, label %22, label %23, !prof !8<br><br>; <label>:22: ; preds = %vector.body<br> unreachable<br><br>; <label>:23: ; preds = %vector.body<br> %24 = ptrtoint i8* %21 to i64<br> %25 = xor i64 %24, 87960930222080<br> %26 = inttoptr i64 %25 to <16 x i8>*<br> %27 = bitcast i8* %21 to <16 x i8>*<br> %28 = add <16 x i8> %vec.ind, %.splat13<br> %_msprop48 = or <16 x i8> %_msphi_s46, %_msprop37<br> store <16 x i8> %_msprop48, <16 x i8>* %26, align 1<br> store <16 x i8> %28, <16 x i8>* %27, align 1, !tbaa !7<br> %29 = getelementptr i8, i8* %21, i64 16<br> br i1 false, label %30, label %31, !prof !8<br><br>; <label>:30: ; preds = %23<br> unreachable<br><br>; <label>:31: ; preds = %23<br> %32 = ptrtoint i8* %29 to i64<br> %33 = xor i64 %32, 87960930222080<br> %34 = inttoptr i64 %33 to <16 x i8>*<br> %35 = bitcast i8* %29 to <16 x i8>*<br> store <16 x i8> %_msprop49, <16 x i8>* %34, align 1<br> store <16 x i8> %19, <16 x i8>* %35, align 1, !tbaa !7<br> %index.next = add i64 %index, 32<br> br i1 false, label %36, label %37, !prof !8<br><br>; <label>:36: ; preds = %31<br> unreachable<br><br>; <label>:37: ; preds = %31<br> %38 = icmp eq i64 %index.next, %n.vec<br> %vec.ind.next = add <16 x i8> %step.add, %.splat15<br> br i1 %38, label %middle.block, label %vector.body, !llvm.loop !9<br><br>middle.block: ; preds = %37<br> br i1 false, label %39, label %40, !prof !8<br><br>; <label>:39: ; preds = %middle.block<br> unreachable<br><br>; <label>:40: ; preds = %middle.block<br> %41 = icmp eq i64 %n.mod.vf, 0<br> br i1 %41, label %for.cond.for.end_crit_edge, label %for.body.preheader<br><br>for.body.preheader: ; preds = %40, %13<br> %_msphi_s = phi i8 [ %_msld, %13 ], [ %_msprop33, %40 ]<br> %.ph = phi i8 [ %.pre, %13 ], [ %ind.end, %40 ]<br> %<a href="http://indvars.iv.ph" target="_blank">indvars.iv.ph</a> = phi i64 [ 0, %13 ], [ %n.vec, %40 ]<br> br label %for.body<br><br>for.body: ; preds = %for.body.preheader, %49<br> %_msphi_s25 = phi i8 [ %_msprop27, %49 ], [ %_msphi_s, %for.body.preheader ]<br> %42 = phi i8 [ %conv1, %49 ], [ %.ph, %for.body.preheader ]<br> %indvars.iv = phi i64 [ %indvars.iv.next, %49 ], [ %<a href="http://indvars.iv.ph" target="_blank">indvars.iv.ph</a>, %for.body.preheader ]<br> %_msprop27 = or i8 %_msphi_s25, %_msprop23<br> %conv1 = add i8 %42, %6<br> %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br> %arrayidx5 = getelementptr inbounds %struct.c, %struct.c* %f, i64 0, <a href="https://maps.google.com/?q=i32+1,+i64&entry=gmail&source=g" target="_blank">i32 1, i64</a> %indvars.iv.next<br> br i1 false, label %43, label %44, !prof !8<br><br>; <label>:43: ; preds = %for.body<br> unreachable<br><br>; <label>:44: ; preds = %for.body<br> %45 = ptrtoint i8* %arrayidx5 to i64<br> %46 = xor i64 %45, 87960930222080<br> %47 = inttoptr i64 %46 to i8*<br> store i8 %_msprop27, i8* %47, align 1<br> store i8 %conv1, i8* %arrayidx5, align 1, !tbaa !7<br> br i1 false, label %48, label %49, !prof !8<br><br>; <label>:48: ; preds = %44<br> unreachable<br><br>; <label>:49: ; preds = %44<br> %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count<br> br i1 %exitcond, label %for.cond.for.end_crit_edge.loopexit, label %for.body, !llvm.loop !11<br><br>for.cond.for.end_crit_edge.loopexit: ; preds = %49<br> br label %for.cond.for.end_crit_edge<br><br>for.cond.for.end_crit_edge: ; preds = %for.cond.for.end_crit_edge.loopexit, %40<br> store i32 0, i32* inttoptr (i64 xor (i64 ptrtoint (i32* @d to i64), i64 87960930222080) to i32*), align 4<br> store i32 %2, i32* @d, align 4, !tbaa !2<br> br label %for.end<br><br>for.end: ; preds = %for.cond.for.end_crit_edge, %5<br> store i32 -1, i32* bitcast ([100 x i64]* @__msan_retval_tls to i32*), align 8<br> ret i32 undef<br>}<br><br>declare void @__msan_init() local_unnamed_addr<br><br>define internal void @msan.module_ctor() {<br> tail call void @__msan_init()<br> ret void<br>}<br><br>declare void @__msan_warning_noreturn()<br><br>declare void @__msan_maybe_warning_1(i8, i32)<br><br>declare void @__msan_maybe_store_origin_1(i8, i8*, i32)<br><br>declare void @__msan_maybe_warning_2(i16, i32)<br><br>declare void @__msan_maybe_store_origin_2(i16, i8*, i32)<br><br>declare void @__msan_maybe_warning_4(i32, i32)<br><br>declare void @__msan_maybe_store_origin_4(i32, i8*, i32)<br><br>declare void @__msan_maybe_warning_8(i64, i32)<br><br>declare void @__msan_maybe_store_origin_8(i64, i8*, i32)<br><br>declare void @__msan_set_alloca_origin4(i8*, i64, i8*, i64)<br><br>declare void @__msan_poison_stack(i8*, i64)<br><br>declare i32 @__msan_chain_origin(i32)<br><br>declare i8* @__msan_memmove(i8*, i8*, i64)<br><br>declare i8* @__msan_memcpy(i8*, i8*, i64)<br><br>declare i8* @__msan_memset(i8*, i32, i64)<br><br>attributes #0 = { norecurse nounwind sanitize_memory "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-features"="+mmx,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }<br>attributes #1 = { nounwind }<br><br>!llvm.module.flags = !{!0}<br>!llvm.ident = !{!1}<br><br>!0 = !{i32 1, !"wchar_size", i32 4}<br>!1 = !{!"clang version 6.0.0 (trunk 320064) (llvm/trunk 320063)"}<br>!2 = !{!3, !<a href="https://maps.google.com/?q=3,+i64&entry=gmail&source=g" target="_blank">3, i64</a> 0}<br>!3 = !{!"int", !<a href="https://maps.google.com/?q=4,+i64&entry=gmail&source=g" target="_blank">4, i64</a> 0}<br>!4 = !{!"omnipotent char", !<a href="https://maps.google.com/?q=5,+i64&entry=gmail&source=g" target="_blank">5, i64</a> 0}<br>!5 = !{!"Simple C/C++ TBAA"}<br>!6 = !{!"branch_weights", i32 1000, i32 1}<br>!7 = !{!4, !<a href="https://maps.google.com/?q=4,+i64&entry=gmail&source=g" target="_blank">4, i64</a> 0}<br>!8 = !{!"branch_weights", i32 1, i32 1000}<br>!9 = distinct !{!9, !10}<br>!10 = !{!"llvm.loop.isvectorized", i32 1}<br>!11 = distinct !{!11, !12, !10}<br>!12 = !{!"llvm.loop.unroll.runtime.disable"}<br></div><div class="m_-5216899527763084458m_-456079807275432973m_-8298998622042008141HOEnZb"><div class="m_-5216899527763084458m_-456079807275432973m_-8298998622042008141h5"><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Dec 8, 2017 at 3:24 AM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@gmail.com" target="_blank">chandlerc@gmail.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">I think this is still enough of a test case to revert while we sort it out.</div><div class="m_-5216899527763084458m_-456079807275432973m_-8298998622042008141m_-3376464414934424306HOEnZb"><div class="m_-5216899527763084458m_-456079807275432973m_-8298998622042008141m_-3376464414934424306h5"><br><div class="gmail_quote"><div dir="ltr">On Fri, Dec 8, 2017 at 4:38 AM David Blaikie <<a href="mailto:dblaikie@gmail.com" target="_blank">dblaikie@gmail.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">Haven't looked too far yet, but here's my current repro:<br><br><div>typedef struct {</div><div> int a;</div><div> char b[]</div><div>} c;</div><div>d;</div><div>e(c *f) {</div><div> int g, h = g = f;</div><div> d = 0;</div><div> for (; d < h; d++)</div><div> f->b[d + 1] = f->b[d] + g;</div><div>}<br><br>clang -cc1 -emit-obj -triple x86_64-linux-gnu -target-feature +sse4.2 -O2 -w -fsanitize=memory -vectorize-loops -o /dev/null foo.ii -x c<br><br>Appears to run forever/a long time.<br><br>Interestingly, a nearby test case seems to fail with or without the patch:<br><br><span style="font-family:monospace;font-size:13px">typedef struct {</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px"> int a;</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px"> char b[]</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px">} c;</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px">d;</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px">e(c *f) {</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px"> int g = f;</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px"> for (; d; d++)</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px"> f->b[d + 1] = f->b[d] + g;</span><br style="font-family:monospace;font-size:13px"><span style="font-family:monospace;font-size:13px">}<br></span><br>So I guess this might be one of those unfortunate cases of an optimization causing further exposure to an existing bug, but still.. </div></div><div dir="ltr"><br><div class="gmail_quote"><div dir="ltr">On Thu, Dec 7, 2017 at 1:56 PM Sanjay Patel <<a href="mailto:spatel@rotateright.com" target="_blank">spatel@rotateright.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>Thanks for letting me know.<br><br></div>This might be a good time to check out (though I haven't gotten to it yet):<br><a href="https://github.com/rutgers-apl/alive-loops" target="_blank">https://github.com/rutgers-apl/alive-loops</a><br><div class="gmail_extra"><br></div><div class="gmail_extra">This was mentioned on llvm-dev:</div><div class="gmail_extra"><a href="http://lists.llvm.org/pipermail/llvm-dev/2017-September/117466.html" target="_blank">http://lists.llvm.org/pipermail/llvm-dev/2017-September/117466.html</a><br></div></div><div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Dec 7, 2017 at 12:48 PM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@gmail.com" target="_blank">chandlerc@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">FYI, we've root caused an compile timeout to this revision. It seems quite likely this is fighting another instcombine.<div><br></div><div>We're still working on a test case, but wanted to go ahead and give a heads-up in case you can spot the place where we reverse this transform.<div><div class="m_-5216899527763084458m_-456079807275432973m_-8298998622042008141m_-3376464414934424306m_2622050965969556890m_5372016222329087132m_-8438521976343846489m_2461758821938462947gmail-m_2453132479159311356h5"><br><br><div class="gmail_quote"><div dir="ltr">On Wed, Nov 15, 2017 at 8:12 PM Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Author: spatel<br>
Date: Wed Nov 15 11:12:01 2017<br>
New Revision: 318323<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=318323&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project?rev=318323&view=rev</a><br>
Log:<br>
[InstCombine] trunc (binop X, C) --> binop (trunc X, C')<br>
<br>
Note that one-use and shouldChangeType() are checked ahead of the switch.<br>
<br>
Without the narrowing folds, we can produce inferior vector code as shown in PR35299:<br>
<a href="https://bugs.llvm.org/show_bug.cgi?id=35299" rel="noreferrer" target="_blank">https://bugs.llvm.org/show_bug.cgi?id=35299</a><br>
<br>
<br>
Modified:<br>
llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp<br>
llvm/trunk/test/Transforms/InstCombine/pr33765.ll<br>
llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll<br>
<br>
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp?rev=318323&r1=318322&r2=318323&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp?rev=318323&r1=318322&r2=318323&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp (original)<br>
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp Wed Nov 15 11:12:01 2017<br>
@@ -545,6 +545,8 @@ Instruction *InstCombiner::narrowBinOp(T<br>
if (!match(Trunc.getOperand(0), m_OneUse(m_BinOp(BinOp))))<br>
return nullptr;<br>
<br>
+ Value *BinOp0 = BinOp->getOperand(0);<br>
+ Value *BinOp1 = BinOp->getOperand(1);<br>
switch (BinOp->getOpcode()) {<br>
case Instruction::And:<br>
case Instruction::Or:<br>
@@ -552,20 +554,31 @@ Instruction *InstCombiner::narrowBinOp(T<br>
case Instruction::Add:<br>
case Instruction::Mul: {<br>
Constant *C;<br>
- if (match(BinOp->getOperand(1), m_Constant(C))) {<br>
+ if (match(BinOp1, m_Constant(C))) {<br>
// trunc (binop X, C) --> binop (trunc X, C')<br>
Constant *NarrowC = ConstantExpr::getTrunc(C, DestTy);<br>
- Value *TruncX = Builder.CreateTrunc(BinOp->getOperand(0), DestTy);<br>
+ Value *TruncX = Builder.CreateTrunc(BinOp0, DestTy);<br>
return BinaryOperator::Create(BinOp->getOpcode(), TruncX, NarrowC);<br>
}<br>
+ Value *X;<br>
+ if (match(BinOp0, m_ZExtOrSExt(m_Value(X))) && X->getType() == DestTy) {<br>
+ // trunc (binop (ext X), Y) --> binop X, (trunc Y)<br>
+ Value *NarrowOp1 = Builder.CreateTrunc(BinOp1, DestTy);<br>
+ return BinaryOperator::Create(BinOp->getOpcode(), X, NarrowOp1);<br>
+ }<br>
+ if (match(BinOp1, m_ZExtOrSExt(m_Value(X))) && X->getType() == DestTy) {<br>
+ // trunc (binop Y, (ext X)) --> binop (trunc Y), X<br>
+ Value *NarrowOp0 = Builder.CreateTrunc(BinOp0, DestTy);<br>
+ return BinaryOperator::Create(BinOp->getOpcode(), NarrowOp0, X);<br>
+ }<br>
break;<br>
}<br>
case Instruction::Sub: {<br>
Constant *C;<br>
- if (match(BinOp->getOperand(0), m_Constant(C))) {<br>
+ if (match(BinOp0, m_Constant(C))) {<br>
// trunc (binop C, X) --> binop (trunc C', X)<br>
Constant *NarrowC = ConstantExpr::getTrunc(C, DestTy);<br>
- Value *TruncX = Builder.CreateTrunc(BinOp->getOperand(1), DestTy);<br>
+ Value *TruncX = Builder.CreateTrunc(BinOp1, DestTy);<br>
return BinaryOperator::Create(BinOp->getOpcode(), NarrowC, TruncX);<br>
}<br>
break;<br>
<br>
Modified: llvm/trunk/test/Transforms/InstCombine/pr33765.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/pr33765.ll?rev=318323&r1=318322&r2=318323&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/pr33765.ll?rev=318323&r1=318322&r2=318323&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/Transforms/InstCombine/pr33765.ll (original)<br>
+++ llvm/trunk/test/Transforms/InstCombine/pr33765.ll Wed Nov 15 11:12:01 2017<br>
@@ -10,9 +10,8 @@ define void @patatino(i8 %beth) {<br>
; CHECK: if.then9:<br>
; CHECK-NEXT: [[MUL:%.*]] = mul nuw nsw i32 [[CONV]], [[CONV]]<br>
; CHECK-NEXT: [[TINKY:%.*]] = load i16, i16* @glob, align 2<br>
-; CHECK-NEXT: [[CONV131:%.*]] = zext i16 [[TINKY]] to i32<br>
-; CHECK-NEXT: [[AND:%.*]] = and i32 [[MUL]], [[CONV131]]<br>
-; CHECK-NEXT: [[CONV14:%.*]] = trunc i32 [[AND]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[MUL]] to i16<br>
+; CHECK-NEXT: [[CONV14:%.*]] = and i16 [[TINKY]], [[TMP1]]<br>
; CHECK-NEXT: store i16 [[CONV14]], i16* @glob, align 2<br>
; CHECK-NEXT: ret void<br>
;<br>
<br>
Modified: llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll?rev=318323&r1=318322&r2=318323&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll?rev=318323&r1=318322&r2=318323&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll (original)<br>
+++ llvm/trunk/test/Transforms/InstCombine/trunc-binop-ext.ll Wed Nov 15 11:12:01 2017<br>
@@ -2,9 +2,8 @@<br>
<br>
define i16 @narrow_sext_and(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_sext_and(<br>
-; CHECK-NEXT: [[X321:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = and i32 [[X321]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = and i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = sext i16 %x16 to i32<br>
@@ -15,9 +14,8 @@ define i16 @narrow_sext_and(i16 %x16, i3<br>
<br>
define i16 @narrow_zext_and(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_zext_and(<br>
-; CHECK-NEXT: [[X32:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = and i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = and i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = zext i16 %x16 to i32<br>
@@ -28,9 +26,8 @@ define i16 @narrow_zext_and(i16 %x16, i3<br>
<br>
define i16 @narrow_sext_or(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_sext_or(<br>
-; CHECK-NEXT: [[X321:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = or i32 [[X321]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = or i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = sext i16 %x16 to i32<br>
@@ -41,9 +38,8 @@ define i16 @narrow_sext_or(i16 %x16, i32<br>
<br>
define i16 @narrow_zext_or(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_zext_or(<br>
-; CHECK-NEXT: [[X32:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = or i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = or i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = zext i16 %x16 to i32<br>
@@ -54,9 +50,8 @@ define i16 @narrow_zext_or(i16 %x16, i32<br>
<br>
define i16 @narrow_sext_xor(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_sext_xor(<br>
-; CHECK-NEXT: [[X321:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = xor i32 [[X321]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = xor i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = sext i16 %x16 to i32<br>
@@ -67,9 +62,8 @@ define i16 @narrow_sext_xor(i16 %x16, i3<br>
<br>
define i16 @narrow_zext_xor(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_zext_xor(<br>
-; CHECK-NEXT: [[X32:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = xor i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = xor i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = zext i16 %x16 to i32<br>
@@ -80,9 +74,8 @@ define i16 @narrow_zext_xor(i16 %x16, i3<br>
<br>
define i16 @narrow_sext_add(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_sext_add(<br>
-; CHECK-NEXT: [[X321:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = add i32 [[X321]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = add i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = sext i16 %x16 to i32<br>
@@ -93,9 +86,8 @@ define i16 @narrow_sext_add(i16 %x16, i3<br>
<br>
define i16 @narrow_zext_add(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_zext_add(<br>
-; CHECK-NEXT: [[X32:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = add i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = add i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = zext i16 %x16 to i32<br>
@@ -106,9 +98,8 @@ define i16 @narrow_zext_add(i16 %x16, i3<br>
<br>
define i16 @narrow_sext_mul(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_sext_mul(<br>
-; CHECK-NEXT: [[X32:%.*]] = sext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = mul i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = mul i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = sext i16 %x16 to i32<br>
@@ -119,9 +110,8 @@ define i16 @narrow_sext_mul(i16 %x16, i3<br>
<br>
define i16 @narrow_zext_mul(i16 %x16, i32 %y32) {<br>
; CHECK-LABEL: @narrow_zext_mul(<br>
-; CHECK-NEXT: [[X32:%.*]] = zext i16 %x16 to i32<br>
-; CHECK-NEXT: [[B:%.*]] = mul i32 [[X32]], %y32<br>
-; CHECK-NEXT: [[R:%.*]] = trunc i32 [[B]] to i16<br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 %y32 to i16<br>
+; CHECK-NEXT: [[R:%.*]] = mul i16 [[TMP1]], %x16<br>
; CHECK-NEXT: ret i16 [[R]]<br>
;<br>
%x32 = zext i16 %x16 to i32<br>
@@ -130,15 +120,14 @@ define i16 @narrow_zext_mul(i16 %x16, i3<br>
ret i16 %r<br>
}<br>
<br>
-; Verify that the commuted patterns work. The div is to ensure that complexity-based<br>
+; Verify that the commuted patterns work. The div is to ensure that complexity-based<br>
; canonicalization doesn't swap the binop operands. Use vector types to show those work too.<br>
<br>
define <2 x i16> @narrow_sext_and_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_sext_and_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X321:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = and <2 x i32> [[Y32OP0]], [[X321]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = and <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -151,9 +140,8 @@ define <2 x i16> @narrow_sext_and_commut<br>
define <2 x i16> @narrow_zext_and_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_zext_and_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = and <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = and <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -166,9 +154,8 @@ define <2 x i16> @narrow_zext_and_commut<br>
define <2 x i16> @narrow_sext_or_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_sext_or_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X321:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = or <2 x i32> [[Y32OP0]], [[X321]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = or <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -181,9 +168,8 @@ define <2 x i16> @narrow_sext_or_commute<br>
define <2 x i16> @narrow_zext_or_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_zext_or_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = or <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = or <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -196,9 +182,8 @@ define <2 x i16> @narrow_zext_or_commute<br>
define <2 x i16> @narrow_sext_xor_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_sext_xor_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X321:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = xor <2 x i32> [[Y32OP0]], [[X321]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = xor <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -211,9 +196,8 @@ define <2 x i16> @narrow_sext_xor_commut<br>
define <2 x i16> @narrow_zext_xor_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_zext_xor_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = xor <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = xor <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -226,9 +210,8 @@ define <2 x i16> @narrow_zext_xor_commut<br>
define <2 x i16> @narrow_sext_add_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_sext_add_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X321:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = add <2 x i32> [[Y32OP0]], [[X321]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = add <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -241,9 +224,8 @@ define <2 x i16> @narrow_sext_add_commut<br>
define <2 x i16> @narrow_zext_add_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_zext_add_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = add <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = add <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -256,9 +238,8 @@ define <2 x i16> @narrow_zext_add_commut<br>
define <2 x i16> @narrow_sext_mul_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_sext_mul_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = sext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = mul <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
@@ -271,9 +252,8 @@ define <2 x i16> @narrow_sext_mul_commut<br>
define <2 x i16> @narrow_zext_mul_commute(<2 x i16> %x16, <2 x i32> %y32) {<br>
; CHECK-LABEL: @narrow_zext_mul_commute(<br>
; CHECK-NEXT: [[Y32OP0:%.*]] = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
-; CHECK-NEXT: [[X32:%.*]] = zext <2 x i16> %x16 to <2 x i32><br>
-; CHECK-NEXT: [[B:%.*]] = mul <2 x i32> [[Y32OP0]], [[X32]]<br>
-; CHECK-NEXT: [[R:%.*]] = trunc <2 x i32> [[B]] to <2 x i16><br>
+; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i32> [[Y32OP0]] to <2 x i16><br>
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i16> [[TMP1]], %x16<br>
; CHECK-NEXT: ret <2 x i16> [[R]]<br>
;<br>
%y32op0 = sdiv <2 x i32> %y32, <i32 7, i32 -17><br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div></div></div></div></div>
</blockquote></div><br></div></div></blockquote></div></div></blockquote></div>
</div></div></blockquote></div><br></div>
</div></div></blockquote></div><br></div>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div>
</div></div></blockquote></div><br></div>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div></div>