<div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr">Thanks for taking a look -- I don't see "Can't unroll" when I run it on the real test case I have, but I do see:</div><div dir="ltr"><div dir="ltr">Loop Unroll: F[<method name>] Loop %for.body</div><div dir="ltr"> Loop Size = 16</div><div dir="ltr"> partially unrolling with count: 1</div><div>And it looks like a count of 1 gets treated similarly to disabling loop unrolling. The bitcode is identical in this case as well. (Attaching a better creduced test case that demonstrates this).</div><div><br></div><div>I guess I have a cleanup to do to remove a bunch of unnecessary pragmas! Is there a simple way to (temporarily) disable this warning? My naive attempt of -Wno-pass-failed=transform-warning doesn't seem to be accepted.</div></div></div></div></div></div><br><div class="gmail_quote"><div dir="ltr">On Thu, Dec 13, 2018 at 3:35 PM Michael Kruse <<a href="mailto:llvm@meinersbur.de">llvm@meinersbur.de</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">A)<br>
I just discovered that some code in the clang frontend disables<br>
unrolling in -O1:<br>
<br>
Opts.UnrollLoops =<br>
Args.hasFlag(OPT_funroll_loops, OPT_fno_unroll_loops,<br>
(Opts.OptimizationLevel > 1));<br>
<br>
B)<br>
Even when emitting the IR to a file and processing with opt (which<br>
skips the clang driver), then processing using opt -O1<br>
-debug-only=loop-unroll<br>
we get:<br>
<br>
Loop Unroll: F[??$g@V<lambda_1>@?0???Rj@@QEAAXH@Z@@f@@QEAAXHV<lambda_1>@?0???Rj@@QEAAXH@Z@@Z]<br>
Loop %for.cond<br>
Loop Size = 5<br>
partially unrolling with count: 8<br>
Can't unroll; loop not terminated by a conditional branch.<br>
Loop Unroll: F[??$g@V<lambda_1>@?0???Rj@@QEAAXH@Z@@f@@QEAAXHV<lambda_1>@?0???Rj@@QEAAXH@Z@@Z]<br>
Loop %for.cond<br>
Loop Size = 5<br>
partially unrolling with count: 8<br>
Can't unroll; loop not terminated by a conditional branch.<br>
<br>
<br>
In either case, the warning did its job to notify you that a #pragma<br>
unroll had no effect. It never had, even before this commit.<br>
This commit does not change the IR output.<br>
<br>
Michael<br>
<br>
Am Do., 13. Dez. 2018 um 16:01 Uhr schrieb Jordan Rupprecht<br>
<<a href="mailto:rupprecht@google.com" target="_blank">rupprecht@google.com</a>>:<br>
><br>
> The code in question is actually using #pragma unroll, not unroll_and_jam.<br>
> Here's a reduced test case (creduce may have gone too far, but this might be enough to figure out what's going on...):<br>
> $ cat reduced.cc<br>
> char c;<br>
> class f {<br>
> public:<br>
> template <typename b><br>
> void g(int, b i) {<br>
> #pragma unroll<br>
> for (int e = 0;;) i(e, &c);<br>
> }<br>
> };<br>
> class j {<br>
> public:<br>
> j(bool);<br>
> bool k() {<br>
> (*this)({});<br>
> return true;<br>
> }<br>
> void operator()(int) {<br>
> auto d = [](int, char *) {};<br>
> h.g(0, d);<br>
> }<br>
> f h;<br>
> };<br>
> void l() {<br>
> j a(false);<br>
> a.k();<br>
> }<br>
> $ /path/to/clang++ -Werror -O1 -c reduced.cc<br>
> reduced.cc:5:8: error: loop not unrolled: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
> [-Werror,-Wpass-failed=transform-warning]<br>
> void g(int, b i) {<br>
><br>
><br>
> On Thu, Dec 13, 2018 at 8:21 AM Michael Kruse <<a href="mailto:llvm@meinersbur.de" target="_blank">llvm@meinersbur.de</a>> wrote:<br>
>><br>
>> Yes, additional warnings are expected. For instance, #pragma<br>
>> unroll_and_jam should trigger such a warning since the<br>
>> LoopUnrollAndJam pass is not in the default pipeline. However,<br>
>> LoopUnroll should always be present and may indeed not have unrolled<br>
>> the loop (although atm I cannot think of a reason why it should). A<br>
>> reduced test case would be useful.<br>
>><br>
>> Michael<br>
>><br>
>> Am Do., 13. Dez. 2018 um 09:58 Uhr schrieb Jordan Rupprecht<br>
>> <<a href="mailto:rupprecht@google.com" target="_blank">rupprecht@google.com</a>>:<br>
>> ><br>
>> > I'm seeing a new error in some internal code (loop not unrolled: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering [-Werror,-Wpass-failed=transform-warning]). Is it expected that this patch can trigger warnings in more cases? I couldn't figure that out from the patch description. I think we're using the new pass manager, fwiw.<br>
>> ><br>
>> > I'll see if I can reduce the internal test case to something that I can report back on this list in a bit.<br>
>> ><br>
>> > On Wed, Dec 12, 2018 at 9:36 AM Michael Kruse via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>> wrote:<br>
>> >><br>
>> >> Author: meinersbur<br>
>> >> Date: Wed Dec 12 09:32:52 2018<br>
>> >> New Revision: 348944<br>
>> >><br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=348944&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project?rev=348944&view=rev</a><br>
>> >> Log:<br>
>> >> [Unroll/UnrollAndJam/Vectorizer/Distribute] Add followup loop attributes.<br>
>> >><br>
>> >> When multiple loop transformation are defined in a loop's metadata, their order of execution is defined by the order of their respective passes in the pass pipeline. For instance, e.g.<br>
>> >><br>
>> >> #pragma clang loop unroll_and_jam(enable)<br>
>> >> #pragma clang loop distribute(enable)<br>
>> >><br>
>> >> is the same as<br>
>> >><br>
>> >> #pragma clang loop distribute(enable)<br>
>> >> #pragma clang loop unroll_and_jam(enable)<br>
>> >><br>
>> >> and will try to loop-distribute before Unroll-And-Jam because the LoopDistribute pass is scheduled after UnrollAndJam pass. UnrollAndJamPass only supports one inner loop, i.e. it will necessarily fail after loop distribution. It is not possible to specify another execution order. Also,t the order of passes in the pipeline is subject to change between versions of LLVM, optimization options and which pass manager is used.<br>
>> >><br>
>> >> This patch adds 'followup' attributes to various loop transformation passes. These attributes define which attributes the resulting loop of a transformation should have. For instance,<br>
>> >><br>
>> >> !0 = !{!0, !1, !2}<br>
>> >> !1 = !{!"llvm.loop.unroll_and_jam.enable"}<br>
>> >> !2 = !{!"llvm.loop.unroll_and_jam.followup_inner", !3}<br>
>> >> !3 = !{!"llvm.loop.distribute.enable"}<br>
>> >><br>
>> >> defines a loop ID (!0) to be unrolled-and-jammed (!1) and then the attribute !3 to be added to the jammed inner loop, which contains the instruction to distribute the inner loop.<br>
>> >><br>
>> >> Currently, in both pass managers, pass execution is in a fixed order and UnrollAndJamPass will not execute again after LoopDistribute. We hope to fix this in the future by allowing pass managers to run passes until a fixpoint is reached, use Polly to perform these transformations, or add a loop transformation pass which takes the order issue into account.<br>
>> >><br>
>> >> For mandatory/forced transformations (e.g. by having been declared by #pragma omp simd), the user must be notified when a transformation could not be performed. It is not possible that the responsible pass emits such a warning because the transformation might be 'hidden' in a followup attribute when it is executed, or it is not present in the pipeline at all. For this reason, this patche introduces a WarnMissedTransformations pass, to warn about orphaned transformations.<br>
>> >><br>
>> >> Since this changes the user-visible diagnostic message when a transformation is applied, two test cases in the clang repository need to be updated.<br>
>> >><br>
>> >> To ensure that no other transformation is executed before the intended one, the attribute `llvm.loop.disable_nonforced` can be added which should disable transformation heuristics before the intended transformation is applied. E.g. it would be surprising if a loop is distributed before a #pragma unroll_and_jam is applied.<br>
>> >><br>
>> >> With more supported code transformations (loop fusion, interchange, stripmining, offloading, etc.), transformations can be used as building blocks for more complex transformations (e.g. stripmining+stripmining+interchange -> tiling).<br>
>> >><br>
>> >> Reviewed By: hfinkel, dmgreen<br>
>> >><br>
>> >> Differential Revision: <a href="https://reviews.llvm.org/D49281" rel="noreferrer" target="_blank">https://reviews.llvm.org/D49281</a><br>
>> >> Differential Revision: <a href="https://reviews.llvm.org/D55288" rel="noreferrer" target="_blank">https://reviews.llvm.org/D55288</a><br>
>> >><br>
>> >> Added:<br>
>> >> llvm/trunk/docs/TransformMetadata.rst<br>
>> >> llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h<br>
>> >> llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp<br>
>> >> llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll<br>
>> >> llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll<br>
>> >> llvm/trunk/test/Transforms/LoopDistribute/followup.ll<br>
>> >> llvm/trunk/test/Transforms/LoopTransformWarning/<br>
>> >> llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll<br>
>> >> llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll<br>
>> >> llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll<br>
>> >> llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnroll/followup.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/followup.ll<br>
>> >> Modified:<br>
>> >> llvm/trunk/docs/LangRef.rst<br>
>> >> llvm/trunk/docs/Passes.rst<br>
>> >> llvm/trunk/docs/index.rst<br>
>> >> llvm/trunk/include/llvm/InitializePasses.h<br>
>> >> llvm/trunk/include/llvm/LinkAllPasses.h<br>
>> >> llvm/trunk/include/llvm/Transforms/Scalar.h<br>
>> >> llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h<br>
>> >> llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h<br>
>> >> llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h<br>
>> >> llvm/trunk/lib/Analysis/LoopInfo.cpp<br>
>> >> llvm/trunk/lib/Passes/PassBuilder.cpp<br>
>> >> llvm/trunk/lib/Passes/PassRegistry.def<br>
>> >> llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp<br>
>> >> llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt<br>
>> >> llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp<br>
>> >> llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp<br>
>> >> llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp<br>
>> >> llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp<br>
>> >> llvm/trunk/lib/Transforms/Scalar/Scalar.cpp<br>
>> >> llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp<br>
>> >> llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp<br>
>> >> llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp<br>
>> >> llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp<br>
>> >> llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp<br>
>> >> llvm/trunk/test/Other/new-pm-defaults.ll<br>
>> >> llvm/trunk/test/Other/new-pm-thinlto-defaults.ll<br>
>> >> llvm/trunk/test/Other/opt-O2-pipeline.ll<br>
>> >> llvm/trunk/test/Other/opt-O3-pipeline.ll<br>
>> >> llvm/trunk/test/Other/opt-Os-pipeline.ll<br>
>> >> llvm/trunk/test/Other/opt-hot-cold-split.ll<br>
>> >> llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll<br>
>> >> llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll<br>
>> >><br>
>> >> Modified: llvm/trunk/docs/LangRef.rst<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/LangRef.rst?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/LangRef.rst?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/docs/LangRef.rst (original)<br>
>> >> +++ llvm/trunk/docs/LangRef.rst Wed Dec 12 09:32:52 2018<br>
>> >> @@ -5076,6 +5076,8 @@ optimizations related to compare and bra<br>
>> >> is treated as a boolean value; if it exists, it signals that the branch<br>
>> >> or switch that it is attached to is completely unpredictable.<br>
>> >><br>
>> >> +.. _llvm.loop:<br>
>> >> +<br>
>> >> '``llvm.loop``'<br>
>> >> ^^^^^^^^^^^^^^^<br>
>> >><br>
>> >> @@ -5109,6 +5111,26 @@ suggests an unroll factor to the loop un<br>
>> >> !0 = !{!0, !1}<br>
>> >> !1 = !{!"llvm.loop.unroll.count", i32 4}<br>
>> >><br>
>> >> +'``llvm.loop.disable_nonforced``'<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata disables all optional loop transformations unless<br>
>> >> +explicitly instructed using other transformation metdata such as<br>
>> >> +``llvm.loop.unroll.enable''. That is, no heuristic will try to determine<br>
>> >> +whether a transformation is profitable. The purpose is to avoid that the<br>
>> >> +loop is transformed to a different loop before an explicitly requested<br>
>> >> +(forced) transformation is applied. For instance, loop fusion can make<br>
>> >> +other transformations impossible. Mandatory loop canonicalizations such<br>
>> >> +as loop rotation are still applied.<br>
>> >> +<br>
>> >> +It is recommended to use this metadata in addition to any llvm.loop.*<br>
>> >> +transformation directive. Also, any loop should have at most one<br>
>> >> +directive applied to it (and a sequence of transformations built using<br>
>> >> +followup-attributes). Otherwise, which transformation will be applied<br>
>> >> +depends on implementation details such as the pass pipeline order.<br>
>> >> +<br>
>> >> +See :ref:`transformation-metadata` for details.<br>
>> >> +<br>
>> >> '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'<br>
>> >> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >><br>
>> >> @@ -5167,6 +5189,29 @@ vectorization of the loop. If ``llvm.loo<br>
>> >> 0 or if the loop does not have this metadata the width will be<br>
>> >> determined automatically.<br>
>> >><br>
>> >> +'``llvm.loop.vectorize.followup_vectorized``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the vectorized loop will<br>
>> >> +have. See :ref:`transformation-metadata` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.vectorize.followup_epilogue``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the epilogue will have. The<br>
>> >> +epilogue is not vectorized and is executed when either the vectorized<br>
>> >> +loop is not known to preserve semantics (because e.g., it processes two<br>
>> >> +arrays that are found to alias by a runtime check) or for the last<br>
>> >> +iterations that do not fill a complete set of vector lanes. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.vectorize.followup_all``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +Attributes in the metadata will be added to both the vectorized and<br>
>> >> +epilogue loop.<br>
>> >> +See :ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> '``llvm.loop.unroll``'<br>
>> >> ^^^^^^^^^^^^^^^^^^^^^^<br>
>> >><br>
>> >> @@ -5235,6 +5280,19 @@ For example:<br>
>> >><br>
>> >> !0 = !{!"llvm.loop.unroll.full"}<br>
>> >><br>
>> >> +'``llvm.loop.unroll.followup``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the unrolled loop will have.<br>
>> >> +See :ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.unroll.followup_remainder``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the remainder loop after<br>
>> >> +partial/runtime unrolling will have. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> '``llvm.loop.unroll_and_jam``'<br>
>> >> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >><br>
>> >> @@ -5288,6 +5346,43 @@ string ``llvm.loop.unroll_and_jam.enable<br>
>> >><br>
>> >> !0 = !{!"llvm.loop.unroll_and_jam.enable"}<br>
>> >><br>
>> >> +'``llvm.loop.unroll_and_jam.followup_outer``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the outer unrolled loop will<br>
>> >> +have. See :ref:`Transformation Metadata <transformation-metadata>` for<br>
>> >> +details.<br>
>> >> +<br>
>> >> +'``llvm.loop.unroll_and_jam.followup_inner``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which loop attributes the inner jammed loop will<br>
>> >> +have. See :ref:`Transformation Metadata <transformation-metadata>` for<br>
>> >> +details.<br>
>> >> +<br>
>> >> +'``llvm.loop.unroll_and_jam.followup_remainder_outer``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which attributes the epilogue of the outer loop<br>
>> >> +will have. This loop is usually unrolled, meaning there is no such<br>
>> >> +loop. This attribute will be ignored in this case. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.unroll_and_jam.followup_remainder_inner``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which attributes the inner loop of the epilogue<br>
>> >> +will have. The outer epilogue will usually be unrolled, meaning there<br>
>> >> +can be multiple inner remainder loops. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.unroll_and_jam.followup_all``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +Attributes specified in the metadata is added to all<br>
>> >> +``llvm.loop.unroll_and_jam.*`` loops. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> '``llvm.loop.licm_versioning.disable``' Metadata<br>
>> >> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >><br>
>> >> @@ -5320,6 +5415,34 @@ enabled. A value of 0 disables distribut<br>
>> >> This metadata should be used in conjunction with ``llvm.loop`` loop<br>
>> >> identification metadata.<br>
>> >><br>
>> >> +'``llvm.loop.distribute.followup_coincident``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which attributes extracted loops with no cyclic<br>
>> >> +dependencies will have (i.e. can be vectorized). See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.distribute.followup_sequential``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +This metadata defines which attributes the isolated loops with unsafe<br>
>> >> +memory dependencies will have. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.distribute.followup_fallback``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +If loop versioning is necessary, this metadata defined the attributes<br>
>> >> +the non-distributed fallback version will have. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> +'``llvm.loop.distribute.followup_all``' Metadata<br>
>> >> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br>
>> >> +<br>
>> >> +Thes attributes in this metdata is added to all followup loops of the<br>
>> >> +loop distribution pass. See<br>
>> >> +:ref:`Transformation Metadata <transformation-metadata>` for details.<br>
>> >> +<br>
>> >> '``llvm.mem``'<br>
>> >> ^^^^^^^^^^^^^^^<br>
>> >><br>
>> >><br>
>> >> Modified: llvm/trunk/docs/Passes.rst<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/Passes.rst?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/Passes.rst?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/docs/Passes.rst (original)<br>
>> >> +++ llvm/trunk/docs/Passes.rst Wed Dec 12 09:32:52 2018<br>
>> >> @@ -1224,3 +1224,8 @@ Displays the post dominator tree using t<br>
>> >> Displays the post dominator tree using the GraphViz tool, but omitting function<br>
>> >> bodies.<br>
>> >><br>
>> >> +``-transform-warning``: Report missed forced transformations<br>
>> >> +------------------------------------------------------------<br>
>> >> +<br>
>> >> +Emits warnings about not yet applied forced transformations (e.g. from<br>
>> >> +``#pragma omp simd``).<br>
>> >><br>
>> >> Added: llvm/trunk/docs/TransformMetadata.rst<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/TransformMetadata.rst?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/TransformMetadata.rst?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/docs/TransformMetadata.rst (added)<br>
>> >> +++ llvm/trunk/docs/TransformMetadata.rst Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,441 @@<br>
>> >> +.. _transformation-metadata:<br>
>> >> +<br>
>> >> +============================<br>
>> >> +Code Transformation Metadata<br>
>> >> +============================<br>
>> >> +<br>
>> >> +.. contents::<br>
>> >> + :local:<br>
>> >> +<br>
>> >> +Overview<br>
>> >> +========<br>
>> >> +<br>
>> >> +LLVM transformation passes can be controlled by attaching metadata to<br>
>> >> +the code to transform. By default, transformation passes use heuristics<br>
>> >> +to determine whether or not to perform transformations, and when doing<br>
>> >> +so, other details of how the transformations are applied (e.g., which<br>
>> >> +vectorization factor to select).<br>
>> >> +Unless the optimizer is otherwise directed, transformations are applied<br>
>> >> +conservatively. This conservatism generally allows the optimizer to<br>
>> >> +avoid unprofitable transformations, but in practice, this results in the<br>
>> >> +optimizer not applying transformations that would be highly profitable.<br>
>> >> +<br>
>> >> +Frontends can give additional hints to LLVM passes on which<br>
>> >> +transformations they should apply. This can be additional knowledge that<br>
>> >> +cannot be derived from the emitted IR, or directives passed from the<br>
>> >> +user/programmer. OpenMP pragmas are an example of the latter.<br>
>> >> +<br>
>> >> +If any such metadata is dropped from the program, the code's semantics<br>
>> >> +must not change.<br>
>> >> +<br>
>> >> +Metadata on Loops<br>
>> >> +=================<br>
>> >> +<br>
>> >> +Attributes can be attached to loops as described in :ref:`llvm.loop`.<br>
>> >> +Attributes can describe properties of the loop, disable transformations,<br>
>> >> +force specific transformations and set transformation options.<br>
>> >> +<br>
>> >> +Because metadata nodes are immutable (with the exception of<br>
>> >> +``MDNode::replaceOperandWith`` which is dangerous to use on uniqued<br>
>> >> +metadata), in order to add or remove a loop attributes, a new ``MDNode``<br>
>> >> +must be created and assigned as the new ``llvm.loop`` metadata. Any<br>
>> >> +connection between the old ``MDNode`` and the loop is lost. The<br>
>> >> +``llvm.loop`` node is also used as LoopID (``Loop::getLoopID()``), i.e.<br>
>> >> +the loop effectively gets a new identifier. For instance,<br>
>> >> +``llvm.mem.parallel_loop_access`` references the LoopID. Therefore, if<br>
>> >> +the parallel access property is to be preserved after adding/removing<br>
>> >> +loop attributes, any ``llvm.mem.parallel_loop_access`` reference must be<br>
>> >> +updated to the new LoopID.<br>
>> >> +<br>
>> >> +Transformation Metadata Structure<br>
>> >> +=================================<br>
>> >> +<br>
>> >> +Some attributes describe code transformations (unrolling, vectorizing,<br>
>> >> +loop distribution, etc.). They can either be a hint to the optimizer<br>
>> >> +that a transformation might be beneficial, instruction to use a specific<br>
>> >> +option, , or convey a specific request from the user (such as<br>
>> >> +``#pragma clang loop`` or ``#pragma omp simd``).<br>
>> >> +<br>
>> >> +If a transformation is forced but cannot be carried-out for any reason,<br>
>> >> +an optimization-missed warning must be emitted. Semantic information<br>
>> >> +such as a transformation being safe (e.g.<br>
>> >> +``llvm.mem.parallel_loop_access``) can be unused by the optimizer<br>
>> >> +without generating a warning.<br>
>> >> +<br>
>> >> +Unless explicitly disabled, any optimization pass may heuristically<br>
>> >> +determine whether a transformation is beneficial and apply it. If<br>
>> >> +metadata for another transformation was specified, applying a different<br>
>> >> +transformation before it might be inadvertent due to being applied on a<br>
>> >> +different loop or the loop not existing anymore. To avoid having to<br>
>> >> +explicitly disable an unknown number of passes, the attribute<br>
>> >> +``llvm.loop.disable_nonforced`` disables all optional, high-level,<br>
>> >> +restructuring transformations.<br>
>> >> +<br>
>> >> +The following example avoids the loop being altered before being<br>
>> >> +vectorized, for instance being unrolled.<br>
>> >> +<br>
>> >> +.. code-block:: llvm<br>
>> >> +<br>
>> >> + br i1 %exitcond, label %for.exit, label %for.header, !llvm.loop !0<br>
>> >> + ...<br>
>> >> + !0 = distinct !{!0, !1, !2}<br>
>> >> + !1 = !{!"llvm.loop.vectorize.enable", i1 true}<br>
>> >> + !2 = !{!"llvm.loop.disable_nonforced"}<br>
>> >> +<br>
>> >> +After a transformation is applied, follow-up attributes are set on the<br>
>> >> +transformed and/or new loop(s). This allows additional attributes<br>
>> >> +including followup-transformations to be specified. Specifying multiple<br>
>> >> +transformations in the same metadata node is possible for compatibility<br>
>> >> +reasons, but their execution order is undefined. For instance, when<br>
>> >> +``llvm.loop.vectorize.enable`` and ``llvm.loop.unroll.enable`` are<br>
>> >> +specified at the same time, unrolling may occur either before or after<br>
>> >> +vectorization.<br>
>> >> +<br>
>> >> +As an example, the following instructs a loop to be vectorized and only<br>
>> >> +then unrolled.<br>
>> >> +<br>
>> >> +.. code-block:: llvm<br>
>> >> +<br>
>> >> + !0 = distinct !{!0, !1, !2, !3}<br>
>> >> + !1 = !{!"llvm.loop.vectorize.enable", i1 true}<br>
>> >> + !2 = !{!"llvm.loop.disable_nonforced"}<br>
>> >> + !3 = !{!"llvm.loop.vectorize.followup_vectorized", !{"llvm.loop.unroll.enable"}}<br>
>> >> +<br>
>> >> +If, and only if, no followup is specified, the pass may add attributes itself.<br>
>> >> +For instance, the vectorizer adds a ``llvm.loop.isvectorized`` attribute and<br>
>> >> +all attributes from the original loop excluding its loop vectorizer<br>
>> >> +attributes. To avoid this, an empty followup attribute can be used, e.g.<br>
>> >> +<br>
>> >> +.. code-block:: llvm<br>
>> >> +<br>
>> >> + !3 = !{!"llvm.loop.vectorize.followup_vectorized"}<br>
>> >> +<br>
>> >> +The followup attributes of a transformation that cannot be applied will<br>
>> >> +never be added to a loop and are therefore effectively ignored. This means<br>
>> >> +that any followup-transformation in such attributes requires that its<br>
>> >> +prior transformations are applied before the followup-transformation.<br>
>> >> +The user should receive a warning about the first transformation in the<br>
>> >> +transformation chain that could not be applied if it a forced<br>
>> >> +transformation. All following transformations are skipped.<br>
>> >> +<br>
>> >> +Pass-Specific Transformation Metadata<br>
>> >> +=====================================<br>
>> >> +<br>
>> >> +Transformation options are specific to each transformation. In the<br>
>> >> +following, we present the model for each LLVM loop optimization pass and<br>
>> >> +the metadata to influence them.<br>
>> >> +<br>
>> >> +Loop Vectorization and Interleaving<br>
>> >> +-----------------------------------<br>
>> >> +<br>
>> >> +Loop vectorization and interleaving is interpreted as a single<br>
>> >> +transformation. It is interpreted as forced if<br>
>> >> +``!{"llvm.loop.vectorize.enable", i1 true}`` is set.<br>
>> >> +<br>
>> >> +Assuming the pre-vectorization loop is<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + for (int i = 0; i < n; i+=1) // original loop<br>
>> >> + Stmt(i);<br>
>> >> +<br>
>> >> +then the code after vectorization will be approximately (assuming an<br>
>> >> +SIMD width of 4):<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + int i = 0;<br>
>> >> + if (rtc) {<br>
>> >> + for (; i + 3 < n; i+=4) // vectorized/interleaved loop<br>
>> >> + Stmt(i:i+3);<br>
>> >> + }<br>
>> >> + for (; i < n; i+=1) // epilogue loop<br>
>> >> + Stmt(i);<br>
>> >> +<br>
>> >> +where ``rtc`` is a generated runtime check.<br>
>> >> +<br>
>> >> +``llvm.loop.vectorize.followup_vectorized`` will set the attributes for<br>
>> >> +the vectorized loop. If not specified, ``llvm.loop.isvectorized`` is<br>
>> >> +combined with the original loop's attributes to avoid it being<br>
>> >> +vectorized multiple times.<br>
>> >> +<br>
>> >> +``llvm.loop.vectorize.followup_epilogue`` will set the attributes for<br>
>> >> +the remainder loop. If not specified, it will have the original loop's<br>
>> >> +attributes combined with ``llvm.loop.isvectorized`` and<br>
>> >> +``llvm.loop.unroll.runtime.disable`` (unless the original loop already<br>
>> >> +has unroll metadata).<br>
>> >> +<br>
>> >> +The attributes specified by ``llvm.loop.vectorize.followup_all`` are<br>
>> >> +added to both loops.<br>
>> >> +<br>
>> >> +When using a follow-up attribute, it replaces any automatically deduced<br>
>> >> +attributes for the generated loop in question. Therefore it is<br>
>> >> +recommended to add ``llvm.loop.isvectorized`` to<br>
>> >> +``llvm.loop.vectorize.followup_all`` which avoids that the loop<br>
>> >> +vectorizer tries to optimize the loops again.<br>
>> >> +<br>
>> >> +Loop Unrolling<br>
>> >> +--------------<br>
>> >> +<br>
>> >> +Unrolling is interpreted as forced any ``!{!"llvm.loop.unroll.enable"}``<br>
>> >> +metadata or option (``llvm.loop.unroll.count``, ``llvm.loop.unroll.full``)<br>
>> >> +is present. Unrolling can be full unrolling, partial unrolling of a loop<br>
>> >> +with constant trip count or runtime unrolling of a loop with a trip<br>
>> >> +count unknown at compile-time.<br>
>> >> +<br>
>> >> +If the loop has been unrolled fully, there is no followup-loop. For<br>
>> >> +partial/runtime unrolling, the original loop of<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + for (int i = 0; i < n; i+=1) // original loop<br>
>> >> + Stmt(i);<br>
>> >> +<br>
>> >> +is transformed into (using an unroll factor of 4):<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + int i = 0;<br>
>> >> + for (; i + 3 < n; i+=4) // unrolled loop<br>
>> >> + Stmt(i);<br>
>> >> + Stmt(i+1);<br>
>> >> + Stmt(i+2);<br>
>> >> + Stmt(i+3);<br>
>> >> + }<br>
>> >> + for (; i < n; i+=1) // remainder loop<br>
>> >> + Stmt(i);<br>
>> >> +<br>
>> >> +``llvm.loop.unroll.followup_unrolled`` will set the loop attributes of<br>
>> >> +the unrolled loop. If not specified, the attributes of the original loop<br>
>> >> +without the ``llvm.loop.unroll.*`` attributes are copied and<br>
>> >> +``llvm.loop.unroll.disable`` added to it.<br>
>> >> +<br>
>> >> +``llvm.loop.unroll.followup_remainder`` defines the attributes of the<br>
>> >> +remainder loop. If not specified the remainder loop will have no<br>
>> >> +attributes. The remainder loop might not be present due to being fully<br>
>> >> +unrolled in which case this attribute has no effect.<br>
>> >> +<br>
>> >> +Attributes defined in ``llvm.loop.unroll.followup_all`` are added to the<br>
>> >> +unrolled and remainder loops.<br>
>> >> +<br>
>> >> +To avoid that the partially unrolled loop is unrolled again, it is<br>
>> >> +recommended to add ``llvm.loop.unroll.disable`` to<br>
>> >> +``llvm.loop.unroll.followup_all``. If no follow-up attribute specified<br>
>> >> +for a generated loop, it is added automatically.<br>
>> >> +<br>
>> >> +Unroll-And-Jam<br>
>> >> +--------------<br>
>> >> +<br>
>> >> +Unroll-and-jam uses the following transformation model (here with an<br>
>> >> +unroll factor if 2). Currently, it does not support a fallback version<br>
>> >> +when the transformation is unsafe.<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + for (int i = 0; i < n; i+=1) { // original outer loop<br>
>> >> + Fore(i);<br>
>> >> + for (int j = 0; j < m; j+=1) // original inner loop<br>
>> >> + SubLoop(i, j);<br>
>> >> + Aft(i);<br>
>> >> + }<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + int i = 0;<br>
>> >> + for (; i + 1 < n; i+=2) { // unrolled outer loop<br>
>> >> + Fore(i);<br>
>> >> + Fore(i+1);<br>
>> >> + for (int j = 0; j < m; j+=1) { // unrolled inner loop<br>
>> >> + SubLoop(i, j);<br>
>> >> + SubLoop(i+1, j);<br>
>> >> + }<br>
>> >> + Aft(i);<br>
>> >> + Aft(i+1);<br>
>> >> + }<br>
>> >> + for (; i < n; i+=1) { // remainder outer loop<br>
>> >> + Fore(i);<br>
>> >> + for (int j = 0; j < m; j+=1) // remainder inner loop<br>
>> >> + SubLoop(i, j);<br>
>> >> + Aft(i);<br>
>> >> + }<br>
>> >> +<br>
>> >> +``llvm.loop.unroll_and_jam.followup_outer`` will set the loop attributes<br>
>> >> +of the unrolled outer loop. If not specified, the attributes of the<br>
>> >> +original outer loop without the ``llvm.loop.unroll.*`` attributes are<br>
>> >> +copied and ``llvm.loop.unroll.disable`` added to it.<br>
>> >> +<br>
>> >> +``llvm.loop.unroll_and_jam.followup_inner`` will set the loop attributes<br>
>> >> +of the unrolled inner loop. If not specified, the attributes of the<br>
>> >> +original inner loop are used unchanged.<br>
>> >> +<br>
>> >> +``llvm.loop.unroll_and_jam.followup_remainder_outer`` sets the loop<br>
>> >> +attributes of the outer remainder loop. If not specified it will not<br>
>> >> +have any attributes. The remainder loop might not be present due to<br>
>> >> +being fully unrolled.<br>
>> >> +<br>
>> >> +``llvm.loop.unroll_and_jam.followup_remainder_inner`` sets the loop<br>
>> >> +attributes of the inner remainder loop. If not specified it will have<br>
>> >> +the attributes of the original inner loop. It the outer remainder loop<br>
>> >> +is unrolled, the inner remainder loop might be present multiple times.<br>
>> >> +<br>
>> >> +Attributes defined in ``llvm.loop.unroll_and_jam.followup_all`` are<br>
>> >> +added to all of the aforementioned output loops.<br>
>> >> +<br>
>> >> +To avoid that the unrolled loop is unrolled again, it is<br>
>> >> +recommended to add ``llvm.loop.unroll.disable`` to<br>
>> >> +``llvm.loop.unroll_and_jam.followup_all``. It suppresses unroll-and-jam<br>
>> >> +as well as an additional inner loop unrolling. If no follow-up<br>
>> >> +attribute specified for a generated loop, it is added automatically.<br>
>> >> +<br>
>> >> +Loop Distribution<br>
>> >> +-----------------<br>
>> >> +<br>
>> >> +The LoopDistribution pass tries to separate vectorizable parts of a loop<br>
>> >> +from the non-vectorizable part (which otherwise would make the entire<br>
>> >> +loop non-vectorizable). Conceptually, it transforms a loop such as<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + for (int i = 1; i < n; i+=1) { // original loop<br>
>> >> + A[i] = i;<br>
>> >> + B[i] = 2 + B[i];<br>
>> >> + C[i] = 3 + C[i - 1];<br>
>> >> + }<br>
>> >> +<br>
>> >> +into the following code:<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + if (rtc) {<br>
>> >> + for (int i = 1; i < n; i+=1) // coincident loop<br>
>> >> + A[i] = i;<br>
>> >> + for (int i = 1; i < n; i+=1) // coincident loop<br>
>> >> + B[i] = 2 + B[i];<br>
>> >> + for (int i = 1; i < n; i+=1) // sequential loop<br>
>> >> + C[i] = 3 + C[i - 1];<br>
>> >> + } else {<br>
>> >> + for (int i = 1; i < n; i+=1) { // fallback loop<br>
>> >> + A[i] = i;<br>
>> >> + B[i] = 2 + B[i];<br>
>> >> + C[i] = 3 + C[i - 1];<br>
>> >> + }<br>
>> >> + }<br>
>> >> +<br>
>> >> +where ``rtc`` is a generated runtime check.<br>
>> >> +<br>
>> >> +``llvm.loop.distribute.followup_coincident`` sets the loop attributes of<br>
>> >> +all loops without loop-carried dependencies (i.e. vectorizable loops).<br>
>> >> +There might be more than one such loops. If not defined, the loops will<br>
>> >> +inherit the original loop's attributes.<br>
>> >> +<br>
>> >> +``llvm.loop.distribute.followup_sequential`` sets the loop attributes of the<br>
>> >> +loop with potentially unsafe dependencies. There should be at most one<br>
>> >> +such loop. If not defined, the loop will inherit the original loop's<br>
>> >> +attributes.<br>
>> >> +<br>
>> >> +``llvm.loop.distribute.followup_fallback`` defines the loop attributes<br>
>> >> +for the fallback loop, which is a copy of the original loop for when<br>
>> >> +loop versioning is required. If undefined, the fallback loop inherits<br>
>> >> +all attributes from the original loop.<br>
>> >> +<br>
>> >> +Attributes defined in ``llvm.loop.distribute.followup_all`` are added to<br>
>> >> +all of the aforementioned output loops.<br>
>> >> +<br>
>> >> +It is recommended to add ``llvm.loop.disable_nonforced`` to<br>
>> >> +``llvm.loop.distribute.followup_fallback``. This avoids that the<br>
>> >> +fallback version (which is likely never executed) is further optimzed<br>
>> >> +which would increase the code size.<br>
>> >> +<br>
>> >> +Versioning LICM<br>
>> >> +---------------<br>
>> >> +<br>
>> >> +The pass hoists code out of loops that are only loop-invariant when<br>
>> >> +dynamic conditions apply. For instance, it transforms the loop<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + for (int i = 0; i < n; i+=1) // original loop<br>
>> >> + A[i] = B[0];<br>
>> >> +<br>
>> >> +into:<br>
>> >> +<br>
>> >> +.. code-block:: c<br>
>> >> +<br>
>> >> + if (rtc) {<br>
>> >> + auto b = B[0];<br>
>> >> + for (int i = 0; i < n; i+=1) // versioned loop<br>
>> >> + A[i] = b;<br>
>> >> + } else {<br>
>> >> + for (int i = 0; i < n; i+=1) // unversioned loop<br>
>> >> + A[i] = B[0];<br>
>> >> + }<br>
>> >> +<br>
>> >> +The runtime condition (``rtc``) checks that the array ``A`` and the<br>
>> >> +element `B[0]` do not alias.<br>
>> >> +<br>
>> >> +Currently, this transformation does not support followup-attributes.<br>
>> >> +<br>
>> >> +Loop Interchange<br>
>> >> +----------------<br>
>> >> +<br>
>> >> +Currently, the ``LoopInterchange`` pass does not use any metadata.<br>
>> >> +<br>
>> >> +Ambiguous Transformation Order<br>
>> >> +==============================<br>
>> >> +<br>
>> >> +If there multiple transformations defined, the order in which they are<br>
>> >> +executed depends on the order in LLVM's pass pipeline, which is subject<br>
>> >> +to change. The default optimization pipeline (anything higher than<br>
>> >> +``-O0``) has the following order.<br>
>> >> +<br>
>> >> +When using the legacy pass manager:<br>
>> >> +<br>
>> >> + - LoopInterchange (if enabled)<br>
>> >> + - SimpleLoopUnroll/LoopFullUnroll (only performs full unrolling)<br>
>> >> + - VersioningLICM (if enabled)<br>
>> >> + - LoopDistribute<br>
>> >> + - LoopVectorizer<br>
>> >> + - LoopUnrollAndJam (if enabled)<br>
>> >> + - LoopUnroll (partial and runtime unrolling)<br>
>> >> +<br>
>> >> +When using the legacy pass manager with LTO:<br>
>> >> +<br>
>> >> + - LoopInterchange (if enabled)<br>
>> >> + - SimpleLoopUnroll/LoopFullUnroll (only performs full unrolling)<br>
>> >> + - LoopVectorizer<br>
>> >> + - LoopUnroll (partial and runtime unrolling)<br>
>> >> +<br>
>> >> +When using the new pass manager:<br>
>> >> +<br>
>> >> + - SimpleLoopUnroll/LoopFullUnroll (only performs full unrolling)<br>
>> >> + - LoopDistribute<br>
>> >> + - LoopVectorizer<br>
>> >> + - LoopUnrollAndJam (if enabled)<br>
>> >> + - LoopUnroll (partial and runtime unrolling)<br>
>> >> +<br>
>> >> +Leftover Transformations<br>
>> >> +========================<br>
>> >> +<br>
>> >> +Forced transformations that have not been applied after the last<br>
>> >> +transformation pass should be reported to the user. The transformation<br>
>> >> +passes themselves cannot be responsible for this reporting because they<br>
>> >> +might not be in the pipeline, there might be multiple passes able to<br>
>> >> +apply a transformation (e.g. ``LoopInterchange`` and Polly) or a<br>
>> >> +transformation attribute may be 'hidden' inside another passes' followup<br>
>> >> +attribute.<br>
>> >> +<br>
>> >> +The pass ``-transform-warning`` (``WarnMissedTransformationsPass``)<br>
>> >> +emits such warnings. It should be placed after the last transformation<br>
>> >> +pass.<br>
>> >> +<br>
>> >> +The current pass pipeline has a fixed order in which transformations<br>
>> >> +passes are executed. A transformation can be in the followup of a pass<br>
>> >> +that is executed later and thus leftover. For instance, a loop nest<br>
>> >> +cannot be distributed and then interchanged with the current pass<br>
>> >> +pipeline. The loop distribution will execute, but there is no loop<br>
>> >> +interchange pass following such that any loop interchange metadata will<br>
>> >> +be ignored. The ``-transform-warning`` should emit a warning in this<br>
>> >> +case.<br>
>> >> +<br>
>> >> +Future versions of LLVM may fix this by executing transformations using<br>
>> >> +a dynamic ordering.<br>
>> >><br>
>> >> Modified: llvm/trunk/docs/index.rst<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/index.rst?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/index.rst?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/docs/index.rst (original)<br>
>> >> +++ llvm/trunk/docs/index.rst Wed Dec 12 09:32:52 2018<br>
>> >> @@ -292,6 +292,7 @@ For API clients and LLVM developers.<br>
>> >> Statepoints<br>
>> >> MergeFunctions<br>
>> >> TypeMetadata<br>
>> >> + TransformMetadata<br>
>> >> FaultMaps<br>
>> >> MIRLangRef<br>
>> >> Coroutines<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/InitializePasses.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InitializePasses.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InitializePasses.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/InitializePasses.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/InitializePasses.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -400,6 +400,7 @@ void initializeUnreachableMachineBlockEl<br>
>> >> void initializeVerifierLegacyPassPass(PassRegistry&);<br>
>> >> void initializeVirtRegMapPass(PassRegistry&);<br>
>> >> void initializeVirtRegRewriterPass(PassRegistry&);<br>
>> >> +void initializeWarnMissedTransformationsLegacyPass(PassRegistry &);<br>
>> >> void initializeWasmEHPreparePass(PassRegistry&);<br>
>> >> void initializeWholeProgramDevirtPass(PassRegistry&);<br>
>> >> void initializeWinEHPreparePass(PassRegistry&);<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/LinkAllPasses.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/LinkAllPasses.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/LinkAllPasses.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/LinkAllPasses.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/LinkAllPasses.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -220,6 +220,7 @@ namespace {<br>
>> >> (void) llvm::createFloat2IntPass();<br>
>> >> (void) llvm::createEliminateAvailableExternallyPass();<br>
>> >> (void) llvm::createScalarizeMaskedMemIntrinPass();<br>
>> >> + (void) llvm::createWarnMissedTransformationsPass();<br>
>> >><br>
>> >> (void)new llvm::IntervalPartition();<br>
>> >> (void)new llvm::ScalarEvolutionWrapperPass();<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/Transforms/Scalar.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Scalar.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Scalar.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/Transforms/Scalar.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/Transforms/Scalar.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -484,6 +484,13 @@ FunctionPass *createLibCallsShrinkWrapPa<br>
>> >> // primarily to help other loop passes.<br>
>> >> //<br>
>> >> Pass *createLoopSimplifyCFGPass();<br>
>> >> +<br>
>> >> +//===----------------------------------------------------------------------===//<br>
>> >> +//<br>
>> >> +// WarnMissedTransformations - This pass emits warnings for leftover forced<br>
>> >> +// transformations.<br>
>> >> +//<br>
>> >> +Pass *createWarnMissedTransformationsPass();<br>
>> >> } // End llvm namespace<br>
>> >><br>
>> >> #endif<br>
>> >><br>
>> >> Added: llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h (added)<br>
>> >> +++ llvm/trunk/include/llvm/Transforms/Scalar/WarnMissedTransforms.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,38 @@<br>
>> >> +//===- WarnMissedTransforms.h -----------------------------------*- C++ -*-===//<br>
>> >> +//<br>
>> >> +// The LLVM Compiler Infrastructure<br>
>> >> +//<br>
>> >> +// This file is distributed under the University of Illinois Open Source<br>
>> >> +// License. See LICENSE.TXT for details.<br>
>> >> +//<br>
>> >> +//===----------------------------------------------------------------------===//<br>
>> >> +//<br>
>> >> +// Emit warnings if forced code transformations have not been performed.<br>
>> >> +//<br>
>> >> +//===----------------------------------------------------------------------===//<br>
>> >> +<br>
>> >> +#ifndef LLVM_TRANSFORMS_SCALAR_WARNMISSEDTRANSFORMS_H<br>
>> >> +#define LLVM_TRANSFORMS_SCALAR_WARNMISSEDTRANSFORMS_H<br>
>> >> +<br>
>> >> +#include "llvm/IR/PassManager.h"<br>
>> >> +<br>
>> >> +namespace llvm {<br>
>> >> +class Function;<br>
>> >> +class Loop;<br>
>> >> +class LPMUpdater;<br>
>> >> +<br>
>> >> +// New pass manager boilerplate.<br>
>> >> +class WarnMissedTransformationsPass<br>
>> >> + : public PassInfoMixin<WarnMissedTransformationsPass> {<br>
>> >> +public:<br>
>> >> + explicit WarnMissedTransformationsPass() {}<br>
>> >> +<br>
>> >> + PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);<br>
>> >> +};<br>
>> >> +<br>
>> >> +// Legacy pass manager boilerplate.<br>
>> >> +Pass *createWarnMissedTransformationsPass();<br>
>> >> +void initializeWarnMissedTransformationsLegacyPass(PassRegistry &);<br>
>> >> +} // end namespace llvm<br>
>> >> +<br>
>> >> +#endif // LLVM_TRANSFORMS_SCALAR_WARNMISSEDTRANSFORMS_H<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/Transforms/Utils/LoopUtils.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -171,6 +171,77 @@ SmallVector<Instruction *, 8> findDefsUs<br>
>> >> Optional<const MDOperand *> findStringMetadataForLoop(Loop *TheLoop,<br>
>> >> StringRef Name);<br>
>> >><br>
>> >> +/// Find named metadata for a loop with an integer value.<br>
>> >> +llvm::Optional<int> getOptionalIntLoopAttribute(Loop *TheLoop, StringRef Name);<br>
>> >> +<br>
>> >> +/// Create a new loop identifier for a loop created from a loop transformation.<br>
>> >> +///<br>
>> >> +/// @param OrigLoopID The loop ID of the loop before the transformation.<br>
>> >> +/// @param FollowupAttrs List of attribute names that contain attributes to be<br>
>> >> +/// added to the new loop ID.<br>
>> >> +/// @param InheritAttrsExceptPrefix Selects which attributes should be inherited<br>
>> >> +/// from the original loop. The following values<br>
>> >> +/// are considered:<br>
>> >> +/// nullptr : Inherit all attributes from @p OrigLoopID.<br>
>> >> +/// "" : Do not inherit any attribute from @p OrigLoopID; only use<br>
>> >> +/// those specified by a followup attribute.<br>
>> >> +/// "<prefix>": Inherit all attributes except those which start with<br>
>> >> +/// <prefix>; commonly used to remove metadata for the<br>
>> >> +/// applied transformation.<br>
>> >> +/// @param AlwaysNew If true, do not try to reuse OrigLoopID and never return<br>
>> >> +/// None.<br>
>> >> +///<br>
>> >> +/// @return The loop ID for the after-transformation loop. The following values<br>
>> >> +/// can be returned:<br>
>> >> +/// None : No followup attribute was found; it is up to the<br>
>> >> +/// transformation to choose attributes that make sense.<br>
>> >> +/// @p OrigLoopID: The original identifier can be reused.<br>
>> >> +/// nullptr : The new loop has no attributes.<br>
>> >> +/// MDNode* : A new unique loop identifier.<br>
>> >> +Optional<MDNode *><br>
>> >> +makeFollowupLoopID(MDNode *OrigLoopID, ArrayRef<StringRef> FollowupAttrs,<br>
>> >> + const char *InheritOptionsAttrsPrefix = "",<br>
>> >> + bool AlwaysNew = false);<br>
>> >> +<br>
>> >> +/// Look for the loop attribute that disables all transformation heuristic.<br>
>> >> +bool hasDisableAllTransformsHint(const Loop *L);<br>
>> >> +<br>
>> >> +/// The mode sets how eager a transformation should be applied.<br>
>> >> +enum TransformationMode {<br>
>> >> + /// The pass can use heuristics to determine whether a transformation should<br>
>> >> + /// be applied.<br>
>> >> + TM_Unspecified,<br>
>> >> +<br>
>> >> + /// The transformation should be applied without considering a cost model.<br>
>> >> + TM_Enable,<br>
>> >> +<br>
>> >> + /// The transformation should not be applied.<br>
>> >> + TM_Disable,<br>
>> >> +<br>
>> >> + /// Force is a flag and should not be used alone.<br>
>> >> + TM_Force = 0x04,<br>
>> >> +<br>
>> >> + /// The transformation was directed by the user, e.g. by a #pragma in<br>
>> >> + /// the source code. If the transformation could not be applied, a<br>
>> >> + /// warning should be emitted.<br>
>> >> + TM_ForcedByUser = TM_Enable | TM_Force,<br>
>> >> +<br>
>> >> + /// The transformation must not be applied. For instance, `#pragma clang loop<br>
>> >> + /// unroll(disable)` explicitly forbids any unrolling to take place. Unlike<br>
>> >> + /// general loop metadata, it must not be dropped. Most passes should not<br>
>> >> + /// behave differently under TM_Disable and TM_SuppressedByUser.<br>
>> >> + TM_SuppressedByUser = TM_Disable | TM_Force<br>
>> >> +};<br>
>> >> +<br>
>> >> +/// @{<br>
>> >> +/// Get the mode for LLVM's supported loop transformations.<br>
>> >> +TransformationMode hasUnrollTransformation(Loop *L);<br>
>> >> +TransformationMode hasUnrollAndJamTransformation(Loop *L);<br>
>> >> +TransformationMode hasVectorizeTransformation(Loop *L);<br>
>> >> +TransformationMode hasDistributeTransformation(Loop *L);<br>
>> >> +TransformationMode hasLICMVersioningTransformation(Loop *L);<br>
>> >> +/// @}<br>
>> >> +<br>
>> >> /// Set input string into loop metadata by keeping other values intact.<br>
>> >> void addStringMetadataToLoop(Loop *TheLoop, const char *MDString,<br>
>> >> unsigned V = 0);<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/Transforms/Utils/UnrollLoop.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -35,6 +35,15 @@ class ScalarEvolution;<br>
>> >><br>
>> >> using NewLoopsMap = SmallDenseMap<const Loop *, Loop *, 4>;<br>
>> >><br>
>> >> +/// @{<br>
>> >> +/// Metadata attribute names<br>
>> >> +const char *const LLVMLoopUnrollFollowupAll = "llvm.loop.unroll.followup_all";<br>
>> >> +const char *const LLVMLoopUnrollFollowupUnrolled =<br>
>> >> + "llvm.loop.unroll.followup_unrolled";<br>
>> >> +const char *const LLVMLoopUnrollFollowupRemainder =<br>
>> >> + "llvm.loop.unroll.followup_remainder";<br>
>> >> +/// @}<br>
>> >> +<br>
>> >> const Loop* addClonedBlockToLoopInfo(BasicBlock *OriginalBB,<br>
>> >> BasicBlock *ClonedBB, LoopInfo *LI,<br>
>> >> NewLoopsMap &NewLoops);<br>
>> >> @@ -61,15 +70,16 @@ LoopUnrollResult UnrollLoop(Loop *L, uns<br>
>> >> unsigned PeelCount, bool UnrollRemainder,<br>
>> >> LoopInfo *LI, ScalarEvolution *SE,<br>
>> >> DominatorTree *DT, AssumptionCache *AC,<br>
>> >> - OptimizationRemarkEmitter *ORE, bool PreserveLCSSA);<br>
>> >> + OptimizationRemarkEmitter *ORE, bool PreserveLCSSA,<br>
>> >> + Loop **RemainderLoop = nullptr);<br>
>> >><br>
>> >> bool UnrollRuntimeLoopRemainder(Loop *L, unsigned Count,<br>
>> >> bool AllowExpensiveTripCount,<br>
>> >> bool UseEpilogRemainder, bool UnrollRemainder,<br>
>> >> - LoopInfo *LI,<br>
>> >> - ScalarEvolution *SE, DominatorTree *DT,<br>
>> >> - AssumptionCache *AC,<br>
>> >> - bool PreserveLCSSA);<br>
>> >> + LoopInfo *LI, ScalarEvolution *SE,<br>
>> >> + DominatorTree *DT, AssumptionCache *AC,<br>
>> >> + bool PreserveLCSSA,<br>
>> >> + Loop **ResultLoop = nullptr);<br>
>> >><br>
>> >> void computePeelCount(Loop *L, unsigned LoopSize,<br>
>> >> TargetTransformInfo::UnrollingPreferences &UP,<br>
>> >> @@ -84,7 +94,8 @@ LoopUnrollResult UnrollAndJamLoop(Loop *<br>
>> >> unsigned TripMultiple, bool UnrollRemainder,<br>
>> >> LoopInfo *LI, ScalarEvolution *SE,<br>
>> >> DominatorTree *DT, AssumptionCache *AC,<br>
>> >> - OptimizationRemarkEmitter *ORE);<br>
>> >> + OptimizationRemarkEmitter *ORE,<br>
>> >> + Loop **EpilogueLoop = nullptr);<br>
>> >><br>
>> >> bool isSafeToUnrollAndJam(Loop *L, ScalarEvolution &SE, DominatorTree &DT,<br>
>> >> DependenceInfo &DI);<br>
>> >><br>
>> >> Modified: llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h (original)<br>
>> >> +++ llvm/trunk/include/llvm/Transforms/Vectorize/LoopVectorizationLegality.h Wed Dec 12 09:32:52 2018<br>
>> >> @@ -113,7 +113,11 @@ public:<br>
>> >> unsigned getWidth() const { return Width.Value; }<br>
>> >> unsigned getInterleave() const { return Interleave.Value; }<br>
>> >> unsigned getIsVectorized() const { return IsVectorized.Value; }<br>
>> >> - enum ForceKind getForce() const { return (ForceKind)Force.Value; }<br>
>> >> + enum ForceKind getForce() const {<br>
>> >> + if (Force.Value == FK_Undefined && hasDisableAllTransformsHint(TheLoop))<br>
>> >> + return FK_Disabled;<br>
>> >> + return (ForceKind)Force.Value;<br>
>> >> + }<br>
>> >><br>
>> >> /// If hints are provided that force vectorization, use the AlwaysPrint<br>
>> >> /// pass name to force the frontend to print the diagnostic.<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Analysis/LoopInfo.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/LoopInfo.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/LoopInfo.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Analysis/LoopInfo.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Analysis/LoopInfo.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -237,23 +237,19 @@ MDNode *Loop::getLoopID() const {<br>
>> >> }<br>
>> >><br>
>> >> void Loop::setLoopID(MDNode *LoopID) const {<br>
>> >> - assert(LoopID && "Loop ID should not be null");<br>
>> >> - assert(LoopID->getNumOperands() > 0 && "Loop ID needs at least one operand");<br>
>> >> - assert(LoopID->getOperand(0) == LoopID && "Loop ID should refer to itself");<br>
>> >> + assert((!LoopID || LoopID->getNumOperands() > 0) &&<br>
>> >> + "Loop ID needs at least one operand");<br>
>> >> + assert((!LoopID || LoopID->getOperand(0) == LoopID) &&<br>
>> >> + "Loop ID should refer to itself");<br>
>> >><br>
>> >> - if (BasicBlock *Latch = getLoopLatch()) {<br>
>> >> - Latch->getTerminator()->setMetadata(LLVMContext::MD_loop, LoopID);<br>
>> >> - return;<br>
>> >> - }<br>
>> >> -<br>
>> >> - assert(!getLoopLatch() &&<br>
>> >> - "The loop should have no single latch at this point");<br>
>> >> BasicBlock *H = getHeader();<br>
>> >> for (BasicBlock *BB : this->blocks()) {<br>
>> >> Instruction *TI = BB->getTerminator();<br>
>> >> for (BasicBlock *Successor : successors(TI)) {<br>
>> >> - if (Successor == H)<br>
>> >> + if (Successor == H) {<br>
>> >> TI->setMetadata(LLVMContext::MD_loop, LoopID);<br>
>> >> + break;<br>
>> >> + }<br>
>> >> }<br>
>> >> }<br>
>> >> }<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Passes/PassBuilder.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Passes/PassBuilder.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Passes/PassBuilder.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Passes/PassBuilder.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Passes/PassBuilder.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -148,6 +148,7 @@<br>
>> >> #include "llvm/Transforms/Scalar/SpeculateAroundPHIs.h"<br>
>> >> #include "llvm/Transforms/Scalar/SpeculativeExecution.h"<br>
>> >> #include "llvm/Transforms/Scalar/TailRecursionElimination.h"<br>
>> >> +#include "llvm/Transforms/Scalar/WarnMissedTransforms.h"<br>
>> >> #include "llvm/Transforms/Utils/AddDiscriminators.h"<br>
>> >> #include "llvm/Transforms/Utils/BreakCriticalEdges.h"<br>
>> >> #include "llvm/Transforms/Utils/EntryExitInstrumenter.h"<br>
>> >> @@ -835,6 +836,7 @@ PassBuilder::buildModuleOptimizationPipe<br>
>> >> createFunctionToLoopPassAdaptor(LoopUnrollAndJamPass(Level)));<br>
>> >> }<br>
>> >> OptimizePM.addPass(LoopUnrollPass(LoopUnrollOptions(Level)));<br>
>> >> + OptimizePM.addPass(WarnMissedTransformationsPass());<br>
>> >> OptimizePM.addPass(InstCombinePass());<br>
>> >> OptimizePM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>());<br>
>> >> OptimizePM.addPass(createFunctionToLoopPassAdaptor(LICMPass(), DebugLogging));<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Passes/PassRegistry.def<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Passes/PassRegistry.def?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Passes/PassRegistry.def?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Passes/PassRegistry.def (original)<br>
>> >> +++ llvm/trunk/lib/Passes/PassRegistry.def Wed Dec 12 09:32:52 2018<br>
>> >> @@ -230,6 +230,7 @@ FUNCTION_PASS("verify<memoryssa>", Memor<br>
>> >> FUNCTION_PASS("verify<regions>", RegionInfoVerifierPass())<br>
>> >> FUNCTION_PASS("view-cfg", CFGViewerPass())<br>
>> >> FUNCTION_PASS("view-cfg-only", CFGOnlyViewerPass())<br>
>> >> +FUNCTION_PASS("transform-warning", WarnMissedTransformationsPass())<br>
>> >> #undef FUNCTION_PASS<br>
>> >><br>
>> >> #ifndef LOOP_ANALYSIS<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -702,6 +702,8 @@ void PassManagerBuilder::populateModuleP<br>
>> >> MPM.add(createLICMPass());<br>
>> >> }<br>
>> >><br>
>> >> + MPM.add(createWarnMissedTransformationsPass());<br>
>> >> +<br>
>> >> // After vectorization and unrolling, assume intrinsics may tell us more<br>
>> >> // about pointer alignments.<br>
>> >> MPM.add(createAlignmentFromAssumptionsPass());<br>
>> >> @@ -877,6 +879,8 @@ void PassManagerBuilder::addLTOOptimizat<br>
>> >> if (!DisableUnrollLoops)<br>
>> >> PM.add(createLoopUnrollPass(OptLevel));<br>
>> >><br>
>> >> + PM.add(createWarnMissedTransformationsPass());<br>
>> >> +<br>
>> >> // Now that we've optimized loops (in particular loop induction variables),<br>
>> >> // we may have exposed more scalar opportunities. Run parts of the scalar<br>
>> >> // optimizer again at this point.<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/CMakeLists.txt Wed Dec 12 09:32:52 2018<br>
>> >> @@ -69,6 +69,7 @@ add_llvm_library(LLVMScalarOpts<br>
>> >> StraightLineStrengthReduce.cpp<br>
>> >> StructurizeCFG.cpp<br>
>> >> TailRecursionElimination.cpp<br>
>> >> + WarnMissedTransforms.cpp<br>
>> >><br>
>> >> ADDITIONAL_HEADER_DIRS<br>
>> >> ${LLVM_MAIN_INCLUDE_DIR}/llvm/Transforms<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/LoopDistribute.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -78,6 +78,18 @@ using namespace llvm;<br>
>> >> #define LDIST_NAME "loop-distribute"<br>
>> >> #define DEBUG_TYPE LDIST_NAME<br>
>> >><br>
>> >> +/// @{<br>
>> >> +/// Metadata attribute names<br>
>> >> +static const char *const LLVMLoopDistributeFollowupAll =<br>
>> >> + "llvm.loop.distribute.followup_all";<br>
>> >> +static const char *const LLVMLoopDistributeFollowupCoincident =<br>
>> >> + "llvm.loop.distribute.followup_coincident";<br>
>> >> +static const char *const LLVMLoopDistributeFollowupSequential =<br>
>> >> + "llvm.loop.distribute.followup_sequential";<br>
>> >> +static const char *const LLVMLoopDistributeFollowupFallback =<br>
>> >> + "llvm.loop.distribute.followup_fallback";<br>
>> >> +/// @}<br>
>> >> +<br>
>> >> static cl::opt<bool><br>
>> >> LDistVerify("loop-distribute-verify", cl::Hidden,<br>
>> >> cl::desc("Turn on DominatorTree and LoopInfo verification "<br>
>> >> @@ -186,7 +198,7 @@ public:<br>
>> >> /// Returns the loop where this partition ends up after distribution.<br>
>> >> /// If this partition is mapped to the original loop then use the block from<br>
>> >> /// the loop.<br>
>> >> - const Loop *getDistributedLoop() const {<br>
>> >> + Loop *getDistributedLoop() const {<br>
>> >> return ClonedLoop ? ClonedLoop : OrigLoop;<br>
>> >> }<br>
>> >><br>
>> >> @@ -443,6 +455,9 @@ public:<br>
>> >> assert(&*OrigPH->begin() == OrigPH->getTerminator() &&<br>
>> >> "preheader not empty");<br>
>> >><br>
>> >> + // Preserve the original loop ID for use after the transformation.<br>
>> >> + MDNode *OrigLoopID = L->getLoopID();<br>
>> >> +<br>
>> >> // Create a loop for each partition except the last. Clone the original<br>
>> >> // loop before PH along with adding a preheader for the cloned loop. Then<br>
>> >> // update PH to point to the newly added preheader.<br>
>> >> @@ -457,9 +472,13 @@ public:<br>
>> >><br>
>> >> Part->getVMap()[ExitBlock] = TopPH;<br>
>> >> Part->remapInstructions();<br>
>> >> + setNewLoopID(OrigLoopID, Part);<br>
>> >> }<br>
>> >> Pred->getTerminator()->replaceUsesOfWith(OrigPH, TopPH);<br>
>> >><br>
>> >> + // Also set a new loop ID for the last loop.<br>
>> >> + setNewLoopID(OrigLoopID, &PartitionContainer.back());<br>
>> >> +<br>
>> >> // Now go in forward order and update the immediate dominator for the<br>
>> >> // preheaders with the exiting block of the previous loop. Dominance<br>
>> >> // within the loop is updated in cloneLoopWithPreheader.<br>
>> >> @@ -575,6 +594,19 @@ private:<br>
>> >> }<br>
>> >> }<br>
>> >> }<br>
>> >> +<br>
>> >> + /// Assign new LoopIDs for the partition's cloned loop.<br>
>> >> + void setNewLoopID(MDNode *OrigLoopID, InstPartition *Part) {<br>
>> >> + Optional<MDNode *> PartitionID = makeFollowupLoopID(<br>
>> >> + OrigLoopID,<br>
>> >> + {LLVMLoopDistributeFollowupAll,<br>
>> >> + Part->hasDepCycle() ? LLVMLoopDistributeFollowupSequential<br>
>> >> + : LLVMLoopDistributeFollowupCoincident});<br>
>> >> + if (PartitionID.hasValue()) {<br>
>> >> + Loop *NewLoop = Part->getDistributedLoop();<br>
>> >> + NewLoop->setLoopID(PartitionID.getValue());<br>
>> >> + }<br>
>> >> + }<br>
>> >> };<br>
>> >><br>
>> >> /// For each memory instruction, this class maintains difference of the<br>
>> >> @@ -743,6 +775,9 @@ public:<br>
>> >> return fail("TooManySCEVRuntimeChecks",<br>
>> >> "too many SCEV run-time checks needed.\n");<br>
>> >><br>
>> >> + if (!IsForced.getValueOr(false) && hasDisableAllTransformsHint(L))<br>
>> >> + return fail("HeuristicDisabled", "distribution heuristic disabled");<br>
>> >> +<br>
>> >> LLVM_DEBUG(dbgs() << "\nDistributing loop: " << *L << "\n");<br>
>> >> // We're done forming the partitions set up the reverse mapping from<br>
>> >> // instructions to partitions.<br>
>> >> @@ -762,6 +797,8 @@ public:<br>
>> >> RtPtrChecking);<br>
>> >><br>
>> >> if (!Pred.isAlwaysTrue() || !Checks.empty()) {<br>
>> >> + MDNode *OrigLoopID = L->getLoopID();<br>
>> >> +<br>
>> >> LLVM_DEBUG(dbgs() << "\nPointers:\n");<br>
>> >> LLVM_DEBUG(LAI->getRuntimePointerChecking()->printChecks(dbgs(), Checks));<br>
>> >> LoopVersioning LVer(*LAI, L, LI, DT, SE, false);<br>
>> >> @@ -769,6 +806,17 @@ public:<br>
>> >> LVer.setSCEVChecks(LAI->getPSE().getUnionPredicate());<br>
>> >> LVer.versionLoop(DefsUsedOutside);<br>
>> >> LVer.annotateLoopWithNoAlias();<br>
>> >> +<br>
>> >> + // The unversioned loop will not be changed, so we inherit all attributes<br>
>> >> + // from the original loop, but remove the loop distribution metadata to<br>
>> >> + // avoid to distribute it again.<br>
>> >> + MDNode *UnversionedLoopID =<br>
>> >> + makeFollowupLoopID(OrigLoopID,<br>
>> >> + {LLVMLoopDistributeFollowupAll,<br>
>> >> + LLVMLoopDistributeFollowupFallback},<br>
>> >> + "llvm.loop.distribute.", true)<br>
>> >> + .getValue();<br>
>> >> + LVer.getNonVersionedLoop()->setLoopID(UnversionedLoopID);<br>
>> >> }<br>
>> >><br>
>> >> // Create identical copies of the original loop for each partition and hook<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/LoopUnrollAndJamPass.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -56,6 +56,20 @@ using namespace llvm;<br>
>> >><br>
>> >> #define DEBUG_TYPE "loop-unroll-and-jam"<br>
>> >><br>
>> >> +/// @{<br>
>> >> +/// Metadata attribute names<br>
>> >> +static const char *const LLVMLoopUnrollAndJamFollowupAll =<br>
>> >> + "llvm.loop.unroll_and_jam.followup_all";<br>
>> >> +static const char *const LLVMLoopUnrollAndJamFollowupInner =<br>
>> >> + "llvm.loop.unroll_and_jam.followup_inner";<br>
>> >> +static const char *const LLVMLoopUnrollAndJamFollowupOuter =<br>
>> >> + "llvm.loop.unroll_and_jam.followup_outer";<br>
>> >> +static const char *const LLVMLoopUnrollAndJamFollowupRemainderInner =<br>
>> >> + "llvm.loop.unroll_and_jam.followup_remainder_inner";<br>
>> >> +static const char *const LLVMLoopUnrollAndJamFollowupRemainderOuter =<br>
>> >> + "llvm.loop.unroll_and_jam.followup_remainder_outer";<br>
>> >> +/// @}<br>
>> >> +<br>
>> >> static cl::opt<bool><br>
>> >> AllowUnrollAndJam("allow-unroll-and-jam", cl::Hidden,<br>
>> >> cl::desc("Allows loops to be unroll-and-jammed."));<br>
>> >> @@ -112,11 +126,6 @@ static bool HasUnrollAndJamEnablePragma(<br>
>> >> return GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.enable");<br>
>> >> }<br>
>> >><br>
>> >> -// Returns true if the loop has an unroll_and_jam(disable) pragma.<br>
>> >> -static bool HasUnrollAndJamDisablePragma(const Loop *L) {<br>
>> >> - return GetUnrollMetadataForLoop(L, "llvm.loop.unroll_and_jam.disable");<br>
>> >> -}<br>
>> >> -<br>
>> >> // If loop has an unroll_and_jam_count pragma return the (necessarily<br>
>> >> // positive) value from the pragma. Otherwise return 0.<br>
>> >> static unsigned UnrollAndJamCountPragmaValue(const Loop *L) {<br>
>> >> @@ -299,13 +308,16 @@ tryToUnrollAndJamLoop(Loop *L, Dominator<br>
>> >> << L->getHeader()->getParent()->getName() << "] Loop %"<br>
>> >> << L->getHeader()->getName() << "\n");<br>
>> >><br>
>> >> + TransformationMode EnableMode = hasUnrollAndJamTransformation(L);<br>
>> >> + if (EnableMode & TM_Disable)<br>
>> >> + return LoopUnrollResult::Unmodified;<br>
>> >> +<br>
>> >> // A loop with any unroll pragma (enabling/disabling/count/etc) is left for<br>
>> >> // the unroller, so long as it does not explicitly have unroll_and_jam<br>
>> >> // metadata. This means #pragma nounroll will disable unroll and jam as well<br>
>> >> // as unrolling<br>
>> >> - if (HasUnrollAndJamDisablePragma(L) ||<br>
>> >> - (HasAnyUnrollPragma(L, "llvm.loop.unroll.") &&<br>
>> >> - !HasAnyUnrollPragma(L, "llvm.loop.unroll_and_jam."))) {<br>
>> >> + if (HasAnyUnrollPragma(L, "llvm.loop.unroll.") &&<br>
>> >> + !HasAnyUnrollPragma(L, "llvm.loop.unroll_and_jam.")) {<br>
>> >> LLVM_DEBUG(dbgs() << " Disabled due to pragma.\n");<br>
>> >> return LoopUnrollResult::Unmodified;<br>
>> >> }<br>
>> >> @@ -344,6 +356,19 @@ tryToUnrollAndJamLoop(Loop *L, Dominator<br>
>> >> return LoopUnrollResult::Unmodified;<br>
>> >> }<br>
>> >><br>
>> >> + // Save original loop IDs for after the transformation.<br>
>> >> + MDNode *OrigOuterLoopID = L->getLoopID();<br>
>> >> + MDNode *OrigSubLoopID = SubLoop->getLoopID();<br>
>> >> +<br>
>> >> + // To assign the loop id of the epilogue, assign it before unrolling it so it<br>
>> >> + // is applied to every inner loop of the epilogue. We later apply the loop ID<br>
>> >> + // for the jammed inner loop.<br>
>> >> + Optional<MDNode *> NewInnerEpilogueLoopID = makeFollowupLoopID(<br>
>> >> + OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,<br>
>> >> + LLVMLoopUnrollAndJamFollowupRemainderInner});<br>
>> >> + if (NewInnerEpilogueLoopID.hasValue())<br>
>> >> + SubLoop->setLoopID(NewInnerEpilogueLoopID.getValue());<br>
>> >> +<br>
>> >> // Find trip count and trip multiple<br>
>> >> unsigned OuterTripCount = SE.getSmallConstantTripCount(L, Latch);<br>
>> >> unsigned OuterTripMultiple = SE.getSmallConstantTripMultiple(L, Latch);<br>
>> >> @@ -359,9 +384,39 @@ tryToUnrollAndJamLoop(Loop *L, Dominator<br>
>> >> if (OuterTripCount && UP.Count > OuterTripCount)<br>
>> >> UP.Count = OuterTripCount;<br>
>> >><br>
>> >> - LoopUnrollResult UnrollResult =<br>
>> >> - UnrollAndJamLoop(L, UP.Count, OuterTripCount, OuterTripMultiple,<br>
>> >> - UP.UnrollRemainder, LI, &SE, &DT, &AC, &ORE);<br>
>> >> + Loop *EpilogueOuterLoop = nullptr;<br>
>> >> + LoopUnrollResult UnrollResult = UnrollAndJamLoop(<br>
>> >> + L, UP.Count, OuterTripCount, OuterTripMultiple, UP.UnrollRemainder, LI,<br>
>> >> + &SE, &DT, &AC, &ORE, &EpilogueOuterLoop);<br>
>> >> +<br>
>> >> + // Assign new loop attributes.<br>
>> >> + if (EpilogueOuterLoop) {<br>
>> >> + Optional<MDNode *> NewOuterEpilogueLoopID = makeFollowupLoopID(<br>
>> >> + OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,<br>
>> >> + LLVMLoopUnrollAndJamFollowupRemainderOuter});<br>
>> >> + if (NewOuterEpilogueLoopID.hasValue())<br>
>> >> + EpilogueOuterLoop->setLoopID(NewOuterEpilogueLoopID.getValue());<br>
>> >> + }<br>
>> >> +<br>
>> >> + Optional<MDNode *> NewInnerLoopID =<br>
>> >> + makeFollowupLoopID(OrigOuterLoopID, {LLVMLoopUnrollAndJamFollowupAll,<br>
>> >> + LLVMLoopUnrollAndJamFollowupInner});<br>
>> >> + if (NewInnerLoopID.hasValue())<br>
>> >> + SubLoop->setLoopID(NewInnerLoopID.getValue());<br>
>> >> + else<br>
>> >> + SubLoop->setLoopID(OrigSubLoopID);<br>
>> >> +<br>
>> >> + if (UnrollResult == LoopUnrollResult::PartiallyUnrolled) {<br>
>> >> + Optional<MDNode *> NewOuterLoopID = makeFollowupLoopID(<br>
>> >> + OrigOuterLoopID,<br>
>> >> + {LLVMLoopUnrollAndJamFollowupAll, LLVMLoopUnrollAndJamFollowupOuter});<br>
>> >> + if (NewOuterLoopID.hasValue()) {<br>
>> >> + L->setLoopID(NewOuterLoopID.getValue());<br>
>> >> +<br>
>> >> + // Do not setLoopAlreadyUnrolled if a followup was given.<br>
>> >> + return UnrollResult;<br>
>> >> + }<br>
>> >> + }<br>
>> >><br>
>> >> // If loop has an unroll count pragma or unrolled by explicitly set count<br>
>> >> // mark loop as unrolled to prevent unrolling beyond that requested.<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -661,11 +661,6 @@ static bool HasUnrollEnablePragma(const<br>
>> >> return GetUnrollMetadataForLoop(L, "llvm.loop.unroll.enable");<br>
>> >> }<br>
>> >><br>
>> >> -// Returns true if the loop has an unroll(disable) pragma.<br>
>> >> -static bool HasUnrollDisablePragma(const Loop *L) {<br>
>> >> - return GetUnrollMetadataForLoop(L, "llvm.loop.unroll.disable");<br>
>> >> -}<br>
>> >> -<br>
>> >> // Returns true if the loop has an runtime unroll(disable) pragma.<br>
>> >> static bool HasRuntimeUnrollDisablePragma(const Loop *L) {<br>
>> >> return GetUnrollMetadataForLoop(L, "llvm.loop.unroll.runtime.disable");<br>
>> >> @@ -713,12 +708,19 @@ static uint64_t getUnrolledLoopSize(<br>
>> >><br>
>> >> // Returns true if unroll count was set explicitly.<br>
>> >> // Calculates unroll count and writes it to UP.Count.<br>
>> >> +// Unless IgnoreUser is true, will also use metadata and command-line options<br>
>> >> +// that are specific to to the LoopUnroll pass (which, for instance, are<br>
>> >> +// irrelevant for the LoopUnrollAndJam pass).<br>
>> >> +// FIXME: This function is used by LoopUnroll and LoopUnrollAndJam, but consumes<br>
>> >> +// many LoopUnroll-specific options. The shared functionality should be<br>
>> >> +// refactored into it own function.<br>
>> >> bool llvm::computeUnrollCount(<br>
>> >> Loop *L, const TargetTransformInfo &TTI, DominatorTree &DT, LoopInfo *LI,<br>
>> >> ScalarEvolution &SE, const SmallPtrSetImpl<const Value *> &EphValues,<br>
>> >> OptimizationRemarkEmitter *ORE, unsigned &TripCount, unsigned MaxTripCount,<br>
>> >> unsigned &TripMultiple, unsigned LoopSize,<br>
>> >> TargetTransformInfo::UnrollingPreferences &UP, bool &UseUpperBound) {<br>
>> >> +<br>
>> >> // Check for explicit Count.<br>
>> >> // 1st priority is unroll count set by "unroll-count" option.<br>
>> >> bool UserUnrollCount = UnrollCount.getNumOccurrences() > 0;<br>
>> >> @@ -969,7 +971,7 @@ static LoopUnrollResult tryToUnrollLoop(<br>
>> >> LLVM_DEBUG(dbgs() << "Loop Unroll: F["<br>
>> >> << L->getHeader()->getParent()->getName() << "] Loop %"<br>
>> >> << L->getHeader()->getName() << "\n");<br>
>> >> - if (HasUnrollDisablePragma(L))<br>
>> >> + if (hasUnrollTransformation(L) & TM_Disable)<br>
>> >> return LoopUnrollResult::Unmodified;<br>
>> >> if (!L->isLoopSimplifyForm()) {<br>
>> >> LLVM_DEBUG(<br>
>> >> @@ -1066,14 +1068,39 @@ static LoopUnrollResult tryToUnrollLoop(<br>
>> >> if (TripCount && UP.Count > TripCount)<br>
>> >> UP.Count = TripCount;<br>
>> >><br>
>> >> + // Save loop properties before it is transformed.<br>
>> >> + MDNode *OrigLoopID = L->getLoopID();<br>
>> >> +<br>
>> >> // Unroll the loop.<br>
>> >> + Loop *RemainderLoop = nullptr;<br>
>> >> LoopUnrollResult UnrollResult = UnrollLoop(<br>
>> >> L, UP.Count, TripCount, UP.Force, UP.Runtime, UP.AllowExpensiveTripCount,<br>
>> >> UseUpperBound, MaxOrZero, TripMultiple, UP.PeelCount, UP.UnrollRemainder,<br>
>> >> - LI, &SE, &DT, &AC, &ORE, PreserveLCSSA);<br>
>> >> + LI, &SE, &DT, &AC, &ORE, PreserveLCSSA, &RemainderLoop);<br>
>> >> if (UnrollResult == LoopUnrollResult::Unmodified)<br>
>> >> return LoopUnrollResult::Unmodified;<br>
>> >><br>
>> >> + if (RemainderLoop) {<br>
>> >> + Optional<MDNode *> RemainderLoopID =<br>
>> >> + makeFollowupLoopID(OrigLoopID, {LLVMLoopUnrollFollowupAll,<br>
>> >> + LLVMLoopUnrollFollowupRemainder});<br>
>> >> + if (RemainderLoopID.hasValue())<br>
>> >> + RemainderLoop->setLoopID(RemainderLoopID.getValue());<br>
>> >> + }<br>
>> >> +<br>
>> >> + if (UnrollResult != LoopUnrollResult::FullyUnrolled) {<br>
>> >> + Optional<MDNode *> NewLoopID =<br>
>> >> + makeFollowupLoopID(OrigLoopID, {LLVMLoopUnrollFollowupAll,<br>
>> >> + LLVMLoopUnrollFollowupUnrolled});<br>
>> >> + if (NewLoopID.hasValue()) {<br>
>> >> + L->setLoopID(NewLoopID.getValue());<br>
>> >> +<br>
>> >> + // Do not setLoopAlreadyUnrolled if loop attributes have been specified<br>
>> >> + // explicitly.<br>
>> >> + return UnrollResult;<br>
>> >> + }<br>
>> >> + }<br>
>> >> +<br>
>> >> // If loop has an unroll count pragma or unrolled by explicitly set count<br>
>> >> // mark loop as unrolled to prevent unrolling beyond that requested.<br>
>> >> // If the loop was peeled, we already "used up" the profile information<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/LoopVersioningLICM.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -594,6 +594,11 @@ bool LoopVersioningLICM::runOnLoop(Loop<br>
>> >><br>
>> >> if (skipLoop(L))<br>
>> >> return false;<br>
>> >> +<br>
>> >> + // Do not do the transformation if disabled by metadata.<br>
>> >> + if (hasLICMVersioningTransformation(L) & TM_Disable)<br>
>> >> + return false;<br>
>> >> +<br>
>> >> // Get Analysis information.<br>
>> >> AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();<br>
>> >> SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Scalar/Scalar.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/Scalar.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/Scalar.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/Scalar.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/Scalar.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -75,6 +75,7 @@ void llvm::initializeScalarOpts(PassRegi<br>
>> >> initializeLoopUnrollPass(Registry);<br>
>> >> initializeLoopUnrollAndJamPass(Registry);<br>
>> >> initializeLoopUnswitchPass(Registry);<br>
>> >> + initializeWarnMissedTransformationsLegacyPass(Registry);<br>
>> >> initializeLoopVersioningLICMPass(Registry);<br>
>> >> initializeLoopIdiomRecognizeLegacyPassPass(Registry);<br>
>> >> initializeLowerAtomicLegacyPassPass(Registry);<br>
>> >><br>
>> >> Added: llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp (added)<br>
>> >> +++ llvm/trunk/lib/Transforms/Scalar/WarnMissedTransforms.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,144 @@<br>
>> >> +//===- LoopTransformWarning.cpp - ----------------------------------------===//<br>
>> >> +//<br>
>> >> +// The LLVM Compiler Infrastructure<br>
>> >> +//<br>
>> >> +// This file is distributed under the University of Illinois Open Source<br>
>> >> +// License. See LICENSE.TXT for details.<br>
>> >> +//<br>
>> >> +//===----------------------------------------------------------------------===//<br>
>> >> +//<br>
>> >> +// Emit warnings if forced code transformations have not been performed.<br>
>> >> +//<br>
>> >> +//===----------------------------------------------------------------------===//<br>
>> >> +<br>
>> >> +#include "llvm/Transforms/Scalar/WarnMissedTransforms.h"<br>
>> >> +#include "llvm/Analysis/OptimizationRemarkEmitter.h"<br>
>> >> +#include "llvm/Transforms/Utils/LoopUtils.h"<br>
>> >> +<br>
>> >> +using namespace llvm;<br>
>> >> +<br>
>> >> +#define DEBUG_TYPE "transform-warning"<br>
>> >> +<br>
>> >> +/// Emit warnings for forced (i.e. user-defined) loop transformations which have<br>
>> >> +/// still not been performed.<br>
>> >> +static void warnAboutLeftoverTransformations(Loop *L,<br>
>> >> + OptimizationRemarkEmitter *ORE) {<br>
>> >> + if (hasUnrollTransformation(L) == TM_ForcedByUser) {<br>
>> >> + LLVM_DEBUG(dbgs() << "Leftover unroll transformation\n");<br>
>> >> + ORE->emit(<br>
>> >> + DiagnosticInfoOptimizationFailure(DEBUG_TYPE,<br>
>> >> + "FailedRequestedUnrolling",<br>
>> >> + L->getStartLoc(), L->getHeader())<br>
>> >> + << "loop not unrolled: the optimizer was unable to perform the "<br>
>> >> + "requested transformation; the transformation might be disabled or "<br>
>> >> + "specified as part of an unsupported transformation ordering");<br>
>> >> + }<br>
>> >> +<br>
>> >> + if (hasUnrollAndJamTransformation(L) == TM_ForcedByUser) {<br>
>> >> + LLVM_DEBUG(dbgs() << "Leftover unroll-and-jam transformation\n");<br>
>> >> + ORE->emit(<br>
>> >> + DiagnosticInfoOptimizationFailure(DEBUG_TYPE,<br>
>> >> + "FailedRequestedUnrollAndJamming",<br>
>> >> + L->getStartLoc(), L->getHeader())<br>
>> >> + << "loop not unroll-and-jammed: the optimizer was unable to perform "<br>
>> >> + "the requested transformation; the transformation might be disabled "<br>
>> >> + "or specified as part of an unsupported transformation ordering");<br>
>> >> + }<br>
>> >> +<br>
>> >> + if (hasVectorizeTransformation(L) == TM_ForcedByUser) {<br>
>> >> + LLVM_DEBUG(dbgs() << "Leftover vectorization transformation\n");<br>
>> >> + Optional<int> VectorizeWidth =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.vectorize.width");<br>
>> >> + Optional<int> InterleaveCount =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.interleave.count");<br>
>> >> +<br>
>> >> + if (VectorizeWidth.getValueOr(0) != 1)<br>
>> >> + ORE->emit(<br>
>> >> + DiagnosticInfoOptimizationFailure(DEBUG_TYPE,<br>
>> >> + "FailedRequestedVectorization",<br>
>> >> + L->getStartLoc(), L->getHeader())<br>
>> >> + << "loop not vectorized: the optimizer was unable to perform the "<br>
>> >> + "requested transformation; the transformation might be disabled "<br>
>> >> + "or specified as part of an unsupported transformation ordering");<br>
>> >> + else if (InterleaveCount.getValueOr(0) != 1)<br>
>> >> + ORE->emit(<br>
>> >> + DiagnosticInfoOptimizationFailure(DEBUG_TYPE,<br>
>> >> + "FailedRequestedInterleaving",<br>
>> >> + L->getStartLoc(), L->getHeader())<br>
>> >> + << "loop not interleaved: the optimizer was unable to perform the "<br>
>> >> + "requested transformation; the transformation might be disabled "<br>
>> >> + "or specified as part of an unsupported transformation ordering");<br>
>> >> + }<br>
>> >> +<br>
>> >> + if (hasDistributeTransformation(L) == TM_ForcedByUser) {<br>
>> >> + LLVM_DEBUG(dbgs() << "Leftover distribute transformation\n");<br>
>> >> + ORE->emit(<br>
>> >> + DiagnosticInfoOptimizationFailure(DEBUG_TYPE,<br>
>> >> + "FailedRequestedDistribution",<br>
>> >> + L->getStartLoc(), L->getHeader())<br>
>> >> + << "loop not distributed: the optimizer was unable to perform the "<br>
>> >> + "requested transformation; the transformation might be disabled or "<br>
>> >> + "specified as part of an unsupported transformation ordering");<br>
>> >> + }<br>
>> >> +}<br>
>> >> +<br>
>> >> +static void warnAboutLeftoverTransformations(Function *F, LoopInfo *LI,<br>
>> >> + OptimizationRemarkEmitter *ORE) {<br>
>> >> + for (auto *L : LI->getLoopsInPreorder())<br>
>> >> + warnAboutLeftoverTransformations(L, ORE);<br>
>> >> +}<br>
>> >> +<br>
>> >> +// New pass manager boilerplate<br>
>> >> +PreservedAnalyses<br>
>> >> +WarnMissedTransformationsPass::run(Function &F, FunctionAnalysisManager &AM) {<br>
>> >> + auto &ORE = AM.getResult<OptimizationRemarkEmitterAnalysis>(F);<br>
>> >> + auto &LI = AM.getResult<LoopAnalysis>(F);<br>
>> >> +<br>
>> >> + warnAboutLeftoverTransformations(&F, &LI, &ORE);<br>
>> >> +<br>
>> >> + return PreservedAnalyses::all();<br>
>> >> +}<br>
>> >> +<br>
>> >> +// Legacy pass manager boilerplate<br>
>> >> +namespace {<br>
>> >> +class WarnMissedTransformationsLegacy : public FunctionPass {<br>
>> >> +public:<br>
>> >> + static char ID;<br>
>> >> +<br>
>> >> + explicit WarnMissedTransformationsLegacy() : FunctionPass(ID) {<br>
>> >> + initializeWarnMissedTransformationsLegacyPass(<br>
>> >> + *PassRegistry::getPassRegistry());<br>
>> >> + }<br>
>> >> +<br>
>> >> + bool runOnFunction(Function &F) override {<br>
>> >> + if (skipFunction(F))<br>
>> >> + return false;<br>
>> >> +<br>
>> >> + auto &ORE = getAnalysis<OptimizationRemarkEmitterWrapperPass>().getORE();<br>
>> >> + auto &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();<br>
>> >> +<br>
>> >> + warnAboutLeftoverTransformations(&F, &LI, &ORE);<br>
>> >> + return false;<br>
>> >> + }<br>
>> >> +<br>
>> >> + void getAnalysisUsage(AnalysisUsage &AU) const override {<br>
>> >> + AU.addRequired<OptimizationRemarkEmitterWrapperPass>();<br>
>> >> + AU.addRequired<LoopInfoWrapperPass>();<br>
>> >> +<br>
>> >> + AU.setPreservesAll();<br>
>> >> + }<br>
>> >> +};<br>
>> >> +} // end anonymous namespace<br>
>> >> +<br>
>> >> +char WarnMissedTransformationsLegacy::ID = 0;<br>
>> >> +<br>
>> >> +INITIALIZE_PASS_BEGIN(WarnMissedTransformationsLegacy, "transform-warning",<br>
>> >> + "Warn about non-applied transformations", false, false)<br>
>> >> +INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)<br>
>> >> +INITIALIZE_PASS_DEPENDENCY(OptimizationRemarkEmitterWrapperPass)<br>
>> >> +INITIALIZE_PASS_END(WarnMissedTransformationsLegacy, "transform-warning",<br>
>> >> + "Warn about non-applied transformations", false, false)<br>
>> >> +<br>
>> >> +Pass *llvm::createWarnMissedTransformationsPass() {<br>
>> >> + return new WarnMissedTransformationsLegacy();<br>
>> >> +}<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Utils/LoopUnroll.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -329,12 +329,15 @@ void llvm::simplifyLoopAfterUnroll(Loop<br>
>> >> ///<br>
>> >> /// This utility preserves LoopInfo. It will also preserve ScalarEvolution and<br>
>> >> /// DominatorTree if they are non-null.<br>
>> >> +///<br>
>> >> +/// If RemainderLoop is non-null, it will receive the remainder loop (if<br>
>> >> +/// required and not fully unrolled).<br>
>> >> LoopUnrollResult llvm::UnrollLoop(<br>
>> >> Loop *L, unsigned Count, unsigned TripCount, bool Force, bool AllowRuntime,<br>
>> >> bool AllowExpensiveTripCount, bool PreserveCondBr, bool PreserveOnlyFirst,<br>
>> >> unsigned TripMultiple, unsigned PeelCount, bool UnrollRemainder,<br>
>> >> LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT, AssumptionCache *AC,<br>
>> >> - OptimizationRemarkEmitter *ORE, bool PreserveLCSSA) {<br>
>> >> + OptimizationRemarkEmitter *ORE, bool PreserveLCSSA, Loop **RemainderLoop) {<br>
>> >><br>
>> >> BasicBlock *Preheader = L->getLoopPreheader();<br>
>> >> if (!Preheader) {<br>
>> >> @@ -468,7 +471,7 @@ LoopUnrollResult llvm::UnrollLoop(<br>
>> >> if (RuntimeTripCount && TripMultiple % Count != 0 &&<br>
>> >> !UnrollRuntimeLoopRemainder(L, Count, AllowExpensiveTripCount,<br>
>> >> EpilogProfitability, UnrollRemainder, LI, SE,<br>
>> >> - DT, AC, PreserveLCSSA)) {<br>
>> >> + DT, AC, PreserveLCSSA, RemainderLoop)) {<br>
>> >> if (Force)<br>
>> >> RuntimeTripCount = false;<br>
>> >> else {<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Utils/LoopUnrollAndJam.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -167,12 +167,14 @@ static void moveHeaderPhiOperandsToForeB<br>
>> >><br>
>> >> isSafeToUnrollAndJam should be used prior to calling this to make sure the<br>
>> >> unrolling will be valid. Checking profitablility is also advisable.<br>
>> >> +<br>
>> >> + If EpilogueLoop is non-null, it receives the epilogue loop (if it was<br>
>> >> + necessary to create one and not fully unrolled).<br>
>> >> */<br>
>> >> -LoopUnrollResult<br>
>> >> -llvm::UnrollAndJamLoop(Loop *L, unsigned Count, unsigned TripCount,<br>
>> >> - unsigned TripMultiple, bool UnrollRemainder,<br>
>> >> - LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT,<br>
>> >> - AssumptionCache *AC, OptimizationRemarkEmitter *ORE) {<br>
>> >> +LoopUnrollResult llvm::UnrollAndJamLoop(<br>
>> >> + Loop *L, unsigned Count, unsigned TripCount, unsigned TripMultiple,<br>
>> >> + bool UnrollRemainder, LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT,<br>
>> >> + AssumptionCache *AC, OptimizationRemarkEmitter *ORE, Loop **EpilogueLoop) {<br>
>> >><br>
>> >> // When we enter here we should have already checked that it is safe<br>
>> >> BasicBlock *Header = L->getHeader();<br>
>> >> @@ -196,7 +198,8 @@ llvm::UnrollAndJamLoop(Loop *L, unsigned<br>
>> >> if (TripMultiple == 1 || TripMultiple % Count != 0) {<br>
>> >> if (!UnrollRuntimeLoopRemainder(L, Count, /*AllowExpensiveTripCount*/ false,<br>
>> >> /*UseEpilogRemainder*/ true,<br>
>> >> - UnrollRemainder, LI, SE, DT, AC, true)) {<br>
>> >> + UnrollRemainder, LI, SE, DT, AC, true,<br>
>> >> + EpilogueLoop)) {<br>
>> >> LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; remainder loop could not be "<br>
>> >> "generated when assuming runtime trip count\n");<br>
>> >> return LoopUnrollResult::Unmodified;<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Utils/LoopUnrollRuntime.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -380,6 +380,7 @@ CloneLoopBlocks(Loop *L, Value *NewIter,<br>
>> >> }<br>
>> >> if (CreateRemainderLoop) {<br>
>> >> Loop *NewLoop = NewLoops[L];<br>
>> >> + MDNode *LoopID = NewLoop->getLoopID();<br>
>> >> assert(NewLoop && "L should have been cloned");<br>
>> >><br>
>> >> // Only add loop metadata if the loop is not going to be completely<br>
>> >> @@ -387,6 +388,16 @@ CloneLoopBlocks(Loop *L, Value *NewIter,<br>
>> >> if (UnrollRemainder)<br>
>> >> return NewLoop;<br>
>> >><br>
>> >> + Optional<MDNode *> NewLoopID = makeFollowupLoopID(<br>
>> >> + LoopID, {LLVMLoopUnrollFollowupAll, LLVMLoopUnrollFollowupRemainder});<br>
>> >> + if (NewLoopID.hasValue()) {<br>
>> >> + NewLoop->setLoopID(NewLoopID.getValue());<br>
>> >> +<br>
>> >> + // Do not setLoopAlreadyUnrolled if loop attributes have been defined<br>
>> >> + // explicitly.<br>
>> >> + return NewLoop;<br>
>> >> + }<br>
>> >> +<br>
>> >> // Add unroll disable metadata to disable future unrolling for this loop.<br>
>> >> NewLoop->setLoopAlreadyUnrolled();<br>
>> >> return NewLoop;<br>
>> >> @@ -525,10 +536,10 @@ static bool canProfitablyUnrollMultiExit<br>
>> >> bool llvm::UnrollRuntimeLoopRemainder(Loop *L, unsigned Count,<br>
>> >> bool AllowExpensiveTripCount,<br>
>> >> bool UseEpilogRemainder,<br>
>> >> - bool UnrollRemainder,<br>
>> >> - LoopInfo *LI, ScalarEvolution *SE,<br>
>> >> - DominatorTree *DT, AssumptionCache *AC,<br>
>> >> - bool PreserveLCSSA) {<br>
>> >> + bool UnrollRemainder, LoopInfo *LI,<br>
>> >> + ScalarEvolution *SE, DominatorTree *DT,<br>
>> >> + AssumptionCache *AC, bool PreserveLCSSA,<br>
>> >> + Loop **ResultLoop) {<br>
>> >> LLVM_DEBUG(dbgs() << "Trying runtime unrolling on Loop: \n");<br>
>> >> LLVM_DEBUG(L->dump());<br>
>> >> LLVM_DEBUG(UseEpilogRemainder ? dbgs() << "Using epilog remainder.\n"<br>
>> >> @@ -911,16 +922,20 @@ bool llvm::UnrollRuntimeLoopRemainder(Lo<br>
>> >> formDedicatedExitBlocks(remainderLoop, DT, LI, PreserveLCSSA);<br>
>> >> }<br>
>> >><br>
>> >> + auto UnrollResult = LoopUnrollResult::Unmodified;<br>
>> >> if (remainderLoop && UnrollRemainder) {<br>
>> >> LLVM_DEBUG(dbgs() << "Unrolling remainder loop\n");<br>
>> >> - UnrollLoop(remainderLoop, /*Count*/ Count - 1, /*TripCount*/ Count - 1,<br>
>> >> - /*Force*/ false, /*AllowRuntime*/ false,<br>
>> >> - /*AllowExpensiveTripCount*/ false, /*PreserveCondBr*/ true,<br>
>> >> - /*PreserveOnlyFirst*/ false, /*TripMultiple*/ 1,<br>
>> >> - /*PeelCount*/ 0, /*UnrollRemainder*/ false, LI, SE, DT, AC,<br>
>> >> - /*ORE*/ nullptr, PreserveLCSSA);<br>
>> >> + UnrollResult =<br>
>> >> + UnrollLoop(remainderLoop, /*Count*/ Count - 1, /*TripCount*/ Count - 1,<br>
>> >> + /*Force*/ false, /*AllowRuntime*/ false,<br>
>> >> + /*AllowExpensiveTripCount*/ false, /*PreserveCondBr*/ true,<br>
>> >> + /*PreserveOnlyFirst*/ false, /*TripMultiple*/ 1,<br>
>> >> + /*PeelCount*/ 0, /*UnrollRemainder*/ false, LI, SE, DT, AC,<br>
>> >> + /*ORE*/ nullptr, PreserveLCSSA);<br>
>> >> }<br>
>> >><br>
>> >> + if (ResultLoop && UnrollResult != LoopUnrollResult::FullyUnrolled)<br>
>> >> + *ResultLoop = remainderLoop;<br>
>> >> NumRuntimeUnrolled++;<br>
>> >> return true;<br>
>> >> }<br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Utils/LoopUtils.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -42,6 +42,8 @@ using namespace llvm::PatternMatch;<br>
>> >><br>
>> >> #define DEBUG_TYPE "loop-utils"<br>
>> >><br>
>> >> +static const char *LLVMLoopDisableNonforced = "llvm.loop.disable_nonforced";<br>
>> >> +<br>
>> >> bool llvm::formDedicatedExitBlocks(Loop *L, DominatorTree *DT, LoopInfo *LI,<br>
>> >> bool PreserveLCSSA) {<br>
>> >> bool Changed = false;<br>
>> >> @@ -183,14 +185,8 @@ void llvm::initializeLoopPassPass(PassRe<br>
>> >> INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass)<br>
>> >> }<br>
>> >><br>
>> >> -/// Find string metadata for loop<br>
>> >> -///<br>
>> >> -/// If it has a value (e.g. {"llvm.distribute", 1} return the value as an<br>
>> >> -/// operand or null otherwise. If the string metadata is not found return<br>
>> >> -/// Optional's not-a-value.<br>
>> >> -Optional<const MDOperand *> llvm::findStringMetadataForLoop(Loop *TheLoop,<br>
>> >> - StringRef Name) {<br>
>> >> - MDNode *LoopID = TheLoop->getLoopID();<br>
>> >> +static Optional<MDNode *> findOptionMDForLoopID(MDNode *LoopID,<br>
>> >> + StringRef Name) {<br>
>> >> // Return none if LoopID is false.<br>
>> >> if (!LoopID)<br>
>> >> return None;<br>
>> >> @@ -209,18 +205,253 @@ Optional<const MDOperand *> llvm::findSt<br>
>> >> continue;<br>
>> >> // Return true if MDString holds expected MetaData.<br>
>> >> if (Name.equals(S->getString()))<br>
>> >> - switch (MD->getNumOperands()) {<br>
>> >> - case 1:<br>
>> >> - return nullptr;<br>
>> >> - case 2:<br>
>> >> - return &MD->getOperand(1);<br>
>> >> - default:<br>
>> >> - llvm_unreachable("loop metadata has 0 or 1 operand");<br>
>> >> - }<br>
>> >> + return MD;<br>
>> >> }<br>
>> >> return None;<br>
>> >> }<br>
>> >><br>
>> >> +static Optional<MDNode *> findOptionMDForLoop(const Loop *TheLoop,<br>
>> >> + StringRef Name) {<br>
>> >> + return findOptionMDForLoopID(TheLoop->getLoopID(), Name);<br>
>> >> +}<br>
>> >> +<br>
>> >> +/// Find string metadata for loop<br>
>> >> +///<br>
>> >> +/// If it has a value (e.g. {"llvm.distribute", 1} return the value as an<br>
>> >> +/// operand or null otherwise. If the string metadata is not found return<br>
>> >> +/// Optional's not-a-value.<br>
>> >> +Optional<const MDOperand *> llvm::findStringMetadataForLoop(Loop *TheLoop,<br>
>> >> + StringRef Name) {<br>
>> >> + auto MD = findOptionMDForLoop(TheLoop, Name).getValueOr(nullptr);<br>
>> >> + if (!MD)<br>
>> >> + return None;<br>
>> >> + switch (MD->getNumOperands()) {<br>
>> >> + case 1:<br>
>> >> + return nullptr;<br>
>> >> + case 2:<br>
>> >> + return &MD->getOperand(1);<br>
>> >> + default:<br>
>> >> + llvm_unreachable("loop metadata has 0 or 1 operand");<br>
>> >> + }<br>
>> >> +}<br>
>> >> +<br>
>> >> +static Optional<bool> getOptionalBoolLoopAttribute(const Loop *TheLoop,<br>
>> >> + StringRef Name) {<br>
>> >> + Optional<MDNode *> MD = findOptionMDForLoop(TheLoop, Name);<br>
>> >> + if (!MD.hasValue())<br>
>> >> + return None;<br>
>> >> + MDNode *OptionNode = MD.getValue();<br>
>> >> + if (OptionNode == nullptr)<br>
>> >> + return None;<br>
>> >> + switch (OptionNode->getNumOperands()) {<br>
>> >> + case 1:<br>
>> >> + // When the value is absent it is interpreted as 'attribute set'.<br>
>> >> + return true;<br>
>> >> + case 2:<br>
>> >> + return mdconst::extract_or_null<ConstantInt>(<br>
>> >> + OptionNode->getOperand(1).get());<br>
>> >> + }<br>
>> >> + llvm_unreachable("unexpected number of options");<br>
>> >> +}<br>
>> >> +<br>
>> >> +static bool getBooleanLoopAttribute(const Loop *TheLoop, StringRef Name) {<br>
>> >> + return getOptionalBoolLoopAttribute(TheLoop, Name).getValueOr(false);<br>
>> >> +}<br>
>> >> +<br>
>> >> +llvm::Optional<int> llvm::getOptionalIntLoopAttribute(Loop *TheLoop,<br>
>> >> + StringRef Name) {<br>
>> >> + const MDOperand *AttrMD =<br>
>> >> + findStringMetadataForLoop(TheLoop, Name).getValueOr(nullptr);<br>
>> >> + if (!AttrMD)<br>
>> >> + return None;<br>
>> >> +<br>
>> >> + ConstantInt *IntMD = mdconst::extract_or_null<ConstantInt>(AttrMD->get());<br>
>> >> + if (!IntMD)<br>
>> >> + return None;<br>
>> >> +<br>
>> >> + return IntMD->getSExtValue();<br>
>> >> +}<br>
>> >> +<br>
>> >> +Optional<MDNode *> llvm::makeFollowupLoopID(<br>
>> >> + MDNode *OrigLoopID, ArrayRef<StringRef> FollowupOptions,<br>
>> >> + const char *InheritOptionsExceptPrefix, bool AlwaysNew) {<br>
>> >> + if (!OrigLoopID) {<br>
>> >> + if (AlwaysNew)<br>
>> >> + return nullptr;<br>
>> >> + return None;<br>
>> >> + }<br>
>> >> +<br>
>> >> + assert(OrigLoopID->getOperand(0) == OrigLoopID);<br>
>> >> +<br>
>> >> + bool InheritAllAttrs = !InheritOptionsExceptPrefix;<br>
>> >> + bool InheritSomeAttrs =<br>
>> >> + InheritOptionsExceptPrefix && InheritOptionsExceptPrefix[0] != '\0';<br>
>> >> + SmallVector<Metadata *, 8> MDs;<br>
>> >> + MDs.push_back(nullptr);<br>
>> >> +<br>
>> >> + bool Changed = false;<br>
>> >> + if (InheritAllAttrs || InheritSomeAttrs) {<br>
>> >> + for (const MDOperand &Existing : drop_begin(OrigLoopID->operands(), 1)) {<br>
>> >> + MDNode *Op = cast<MDNode>(Existing.get());<br>
>> >> +<br>
>> >> + auto InheritThisAttribute = [InheritSomeAttrs,<br>
>> >> + InheritOptionsExceptPrefix](MDNode *Op) {<br>
>> >> + if (!InheritSomeAttrs)<br>
>> >> + return false;<br>
>> >> +<br>
>> >> + // Skip malformatted attribute metadata nodes.<br>
>> >> + if (Op->getNumOperands() == 0)<br>
>> >> + return true;<br>
>> >> + Metadata *NameMD = Op->getOperand(0).get();<br>
>> >> + if (!isa<MDString>(NameMD))<br>
>> >> + return true;<br>
>> >> + StringRef AttrName = cast<MDString>(NameMD)->getString();<br>
>> >> +<br>
>> >> + // Do not inherit excluded attributes.<br>
>> >> + return !AttrName.startswith(InheritOptionsExceptPrefix);<br>
>> >> + };<br>
>> >> +<br>
>> >> + if (InheritThisAttribute(Op))<br>
>> >> + MDs.push_back(Op);<br>
>> >> + else<br>
>> >> + Changed = true;<br>
>> >> + }<br>
>> >> + } else {<br>
>> >> + // Modified if we dropped at least one attribute.<br>
>> >> + Changed = OrigLoopID->getNumOperands() > 1;<br>
>> >> + }<br>
>> >> +<br>
>> >> + bool HasAnyFollowup = false;<br>
>> >> + for (StringRef OptionName : FollowupOptions) {<br>
>> >> + MDNode *FollowupNode =<br>
>> >> + findOptionMDForLoopID(OrigLoopID, OptionName).getValueOr(nullptr);<br>
>> >> + if (!FollowupNode)<br>
>> >> + continue;<br>
>> >> +<br>
>> >> + HasAnyFollowup = true;<br>
>> >> + for (const MDOperand &Option : drop_begin(FollowupNode->operands(), 1)) {<br>
>> >> + MDs.push_back(Option.get());<br>
>> >> + Changed = true;<br>
>> >> + }<br>
>> >> + }<br>
>> >> +<br>
>> >> + // Attributes of the followup loop not specified explicity, so signal to the<br>
>> >> + // transformation pass to add suitable attributes.<br>
>> >> + if (!AlwaysNew && !HasAnyFollowup)<br>
>> >> + return None;<br>
>> >> +<br>
>> >> + // If no attributes were added or remove, the previous loop Id can be reused.<br>
>> >> + if (!AlwaysNew && !Changed)<br>
>> >> + return OrigLoopID;<br>
>> >> +<br>
>> >> + // No attributes is equivalent to having no !llvm.loop metadata at all.<br>
>> >> + if (MDs.size() == 1)<br>
>> >> + return nullptr;<br>
>> >> +<br>
>> >> + // Build the new loop ID.<br>
>> >> + MDTuple *FollowupLoopID = MDNode::get(OrigLoopID->getContext(), MDs);<br>
>> >> + FollowupLoopID->replaceOperandWith(0, FollowupLoopID);<br>
>> >> + return FollowupLoopID;<br>
>> >> +}<br>
>> >> +<br>
>> >> +bool llvm::hasDisableAllTransformsHint(const Loop *L) {<br>
>> >> + return getBooleanLoopAttribute(L, LLVMLoopDisableNonforced);<br>
>> >> +}<br>
>> >> +<br>
>> >> +TransformationMode llvm::hasUnrollTransformation(Loop *L) {<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.unroll.disable"))<br>
>> >> + return TM_SuppressedByUser;<br>
>> >> +<br>
>> >> + Optional<int> Count =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.unroll.count");<br>
>> >> + if (Count.hasValue())<br>
>> >> + return Count.getValue() == 1 ? TM_SuppressedByUser : TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.unroll.enable"))<br>
>> >> + return TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.unroll.full"))<br>
>> >> + return TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (hasDisableAllTransformsHint(L))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + return TM_Unspecified;<br>
>> >> +}<br>
>> >> +<br>
>> >> +TransformationMode llvm::hasUnrollAndJamTransformation(Loop *L) {<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.unroll_and_jam.disable"))<br>
>> >> + return TM_SuppressedByUser;<br>
>> >> +<br>
>> >> + Optional<int> Count =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.unroll_and_jam.count");<br>
>> >> + if (Count.hasValue())<br>
>> >> + return Count.getValue() == 1 ? TM_SuppressedByUser : TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.unroll_and_jam.enable"))<br>
>> >> + return TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (hasDisableAllTransformsHint(L))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + return TM_Unspecified;<br>
>> >> +}<br>
>> >> +<br>
>> >> +TransformationMode llvm::hasVectorizeTransformation(Loop *L) {<br>
>> >> + Optional<bool> Enable =<br>
>> >> + getOptionalBoolLoopAttribute(L, "llvm.loop.vectorize.enable");<br>
>> >> +<br>
>> >> + if (Enable == false)<br>
>> >> + return TM_SuppressedByUser;<br>
>> >> +<br>
>> >> + Optional<int> VectorizeWidth =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.vectorize.width");<br>
>> >> + Optional<int> InterleaveCount =<br>
>> >> + getOptionalIntLoopAttribute(L, "llvm.loop.interleave.count");<br>
>> >> +<br>
>> >> + if (Enable == true) {<br>
>> >> + // 'Forcing' vector width and interleave count to one effectively disables<br>
>> >> + // this tranformation.<br>
>> >> + if (VectorizeWidth == 1 && InterleaveCount == 1)<br>
>> >> + return TM_SuppressedByUser;<br>
>> >> + return TM_ForcedByUser;<br>
>> >> + }<br>
>> >> +<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.isvectorized"))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + if (VectorizeWidth == 1 && InterleaveCount == 1)<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + if (VectorizeWidth > 1 || InterleaveCount > 1)<br>
>> >> + return TM_Enable;<br>
>> >> +<br>
>> >> + if (hasDisableAllTransformsHint(L))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + return TM_Unspecified;<br>
>> >> +}<br>
>> >> +<br>
>> >> +TransformationMode llvm::hasDistributeTransformation(Loop *L) {<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.distribute.enable"))<br>
>> >> + return TM_ForcedByUser;<br>
>> >> +<br>
>> >> + if (hasDisableAllTransformsHint(L))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + return TM_Unspecified;<br>
>> >> +}<br>
>> >> +<br>
>> >> +TransformationMode llvm::hasLICMVersioningTransformation(Loop *L) {<br>
>> >> + if (getBooleanLoopAttribute(L, "llvm.loop.licm_versioning.disable"))<br>
>> >> + return TM_SuppressedByUser;<br>
>> >> +<br>
>> >> + if (hasDisableAllTransformsHint(L))<br>
>> >> + return TM_Disable;<br>
>> >> +<br>
>> >> + return TM_Unspecified;<br>
>> >> +}<br>
>> >> +<br>
>> >> /// Does a BFS from a given node to all of its children inside a given loop.<br>
>> >> /// The returned vector of nodes includes the starting point.<br>
>> >> SmallVector<DomTreeNode *, 16><br>
>> >><br>
>> >> Modified: llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp (original)<br>
>> >> +++ llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp Wed Dec 12 09:32:52 2018<br>
>> >> @@ -152,6 +152,16 @@ using namespace llvm;<br>
>> >> #define LV_NAME "loop-vectorize"<br>
>> >> #define DEBUG_TYPE LV_NAME<br>
>> >><br>
>> >> +/// @{<br>
>> >> +/// Metadata attribute names<br>
>> >> +static const char *const LLVMLoopVectorizeFollowupAll =<br>
>> >> + "llvm.loop.vectorize.followup_all";<br>
>> >> +static const char *const LLVMLoopVectorizeFollowupVectorized =<br>
>> >> + "llvm.loop.vectorize.followup_vectorized";<br>
>> >> +static const char *const LLVMLoopVectorizeFollowupEpilogue =<br>
>> >> + "llvm.loop.vectorize.followup_epilogue";<br>
>> >> +/// @}<br>
>> >> +<br>
>> >> STATISTIC(LoopsVectorized, "Number of loops vectorized");<br>
>> >> STATISTIC(LoopsAnalyzed, "Number of loops analyzed for vectorization");<br>
>> >><br>
>> >> @@ -796,27 +806,6 @@ void InnerLoopVectorizer::addMetadata(Ar<br>
>> >> }<br>
>> >> }<br>
>> >><br>
>> >> -static void emitMissedWarning(Function *F, Loop *L,<br>
>> >> - const LoopVectorizeHints &LH,<br>
>> >> - OptimizationRemarkEmitter *ORE) {<br>
>> >> - LH.emitRemarkWithHints();<br>
>> >> -<br>
>> >> - if (LH.getForce() == LoopVectorizeHints::FK_Enabled) {<br>
>> >> - if (LH.getWidth() != 1)<br>
>> >> - ORE->emit(DiagnosticInfoOptimizationFailure(<br>
>> >> - DEBUG_TYPE, "FailedRequestedVectorization",<br>
>> >> - L->getStartLoc(), L->getHeader())<br>
>> >> - << "loop not vectorized: "<br>
>> >> - << "failed explicitly specified loop vectorization");<br>
>> >> - else if (LH.getInterleave() != 1)<br>
>> >> - ORE->emit(DiagnosticInfoOptimizationFailure(<br>
>> >> - DEBUG_TYPE, "FailedRequestedInterleaving", L->getStartLoc(),<br>
>> >> - L->getHeader())<br>
>> >> - << "loop not interleaved: "<br>
>> >> - << "failed explicitly specified loop interleaving");<br>
>> >> - }<br>
>> >> -}<br>
>> >> -<br>
>> >> namespace llvm {<br>
>> >><br>
>> >> /// LoopVectorizationCostModel - estimates the expected speedups due to<br>
>> >> @@ -1377,7 +1366,7 @@ static bool isExplicitVecOuterLoop(Loop<br>
>> >><br>
>> >> if (!Hints.getWidth()) {<br>
>> >> LLVM_DEBUG(dbgs() << "LV: Not vectorizing: No user vector width.\n");<br>
>> >> - emitMissedWarning(Fn, OuterLp, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -1385,7 +1374,7 @@ static bool isExplicitVecOuterLoop(Loop<br>
>> >> // TODO: Interleave support is future work.<br>
>> >> LLVM_DEBUG(dbgs() << "LV: Not vectorizing: Interleave is not supported for "<br>
>> >> "outer loops.\n");<br>
>> >> - emitMissedWarning(Fn, OuterLp, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -2739,6 +2728,7 @@ BasicBlock *InnerLoopVectorizer::createV<br>
>> >> BasicBlock *OldBasicBlock = OrigLoop->getHeader();<br>
>> >> BasicBlock *VectorPH = OrigLoop->getLoopPreheader();<br>
>> >> BasicBlock *ExitBlock = OrigLoop->getExitBlock();<br>
>> >> + MDNode *OrigLoopID = OrigLoop->getLoopID();<br>
>> >> assert(VectorPH && "Invalid loop structure");<br>
>> >> assert(ExitBlock && "Must have an exit block");<br>
>> >><br>
>> >> @@ -2882,6 +2872,17 @@ BasicBlock *InnerLoopVectorizer::createV<br>
>> >> LoopVectorBody = VecBody;<br>
>> >> LoopScalarBody = OldBasicBlock;<br>
>> >><br>
>> >> + Optional<MDNode *> VectorizedLoopID =<br>
>> >> + makeFollowupLoopID(OrigLoopID, {LLVMLoopVectorizeFollowupAll,<br>
>> >> + LLVMLoopVectorizeFollowupVectorized});<br>
>> >> + if (VectorizedLoopID.hasValue()) {<br>
>> >> + Lp->setLoopID(VectorizedLoopID.getValue());<br>
>> >> +<br>
>> >> + // Do not setAlreadyVectorized if loop attributes have been defined<br>
>> >> + // explicitly.<br>
>> >> + return LoopVectorPreHeader;<br>
>> >> + }<br>
>> >> +<br>
>> >> // Keep all loop hints from the original loop on the vector loop (we'll<br>
>> >> // replace the vectorizer-specific hints below).<br>
>> >> if (MDNode *LID = OrigLoop->getLoopID())<br>
>> >> @@ -7177,7 +7178,7 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> &Requirements, &Hints, DB, AC);<br>
>> >> if (!LVL.canVectorize(EnableVPlanNativePath)) {<br>
>> >> LLVM_DEBUG(dbgs() << "LV: Not vectorizing: Cannot prove legality.\n");<br>
>> >> - emitMissedWarning(F, L, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -7250,7 +7251,7 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> ORE->emit(createLVMissedAnalysis(Hints.vectorizeAnalysisPassName(),<br>
>> >> "NoImplicitFloat", L)<br>
>> >> << "loop not vectorized due to NoImplicitFloat attribute");<br>
>> >> - emitMissedWarning(F, L, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -7265,7 +7266,7 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> ORE->emit(<br>
>> >> createLVMissedAnalysis(Hints.vectorizeAnalysisPassName(), "UnsafeFP", L)<br>
>> >> << "loop not vectorized due to unsafe FP support.");<br>
>> >> - emitMissedWarning(F, L, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -7307,7 +7308,7 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> if (Requirements.doesNotMeet(F, L, Hints)) {<br>
>> >> LLVM_DEBUG(dbgs() << "LV: Not vectorizing: loop did not meet vectorization "<br>
>> >> "requirements.\n");<br>
>> >> - emitMissedWarning(F, L, Hints, ORE);<br>
>> >> + Hints.emitRemarkWithHints();<br>
>> >> return false;<br>
>> >> }<br>
>> >><br>
>> >> @@ -7384,6 +7385,8 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> LVP.setBestPlan(VF.Width, IC);<br>
>> >><br>
>> >> using namespace ore;<br>
>> >> + bool DisableRuntimeUnroll = false;<br>
>> >> + MDNode *OrigLoopID = L->getLoopID();<br>
>> >><br>
>> >> if (!VectorizeLoop) {<br>
>> >> assert(IC > 1 && "interleave count should not be 1 or 0");<br>
>> >> @@ -7410,7 +7413,7 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> // no runtime checks about strides and memory. A scalar loop that is<br>
>> >> // rarely used is not worth unrolling.<br>
>> >> if (!LB.areSafetyChecksAdded())<br>
>> >> - AddRuntimeUnrollDisableMetaData(L);<br>
>> >> + DisableRuntimeUnroll = true;<br>
>> >><br>
>> >> // Report the vectorization decision.<br>
>> >> ORE->emit([&]() {<br>
>> >> @@ -7422,8 +7425,18 @@ bool LoopVectorizePass::processLoop(Loop<br>
>> >> });<br>
>> >> }<br>
>> >><br>
>> >> - // Mark the loop as already vectorized to avoid vectorizing again.<br>
>> >> - Hints.setAlreadyVectorized();<br>
>> >> + Optional<MDNode *> RemainderLoopID =<br>
>> >> + makeFollowupLoopID(OrigLoopID, {LLVMLoopVectorizeFollowupAll,<br>
>> >> + LLVMLoopVectorizeFollowupEpilogue});<br>
>> >> + if (RemainderLoopID.hasValue()) {<br>
>> >> + L->setLoopID(RemainderLoopID.getValue());<br>
>> >> + } else {<br>
>> >> + if (DisableRuntimeUnroll)<br>
>> >> + AddRuntimeUnrollDisableMetaData(L);<br>
>> >> +<br>
>> >> + // Mark the loop as already vectorized to avoid vectorizing again.<br>
>> >> + Hints.setAlreadyVectorized();<br>
>> >> + }<br>
>> >><br>
>> >> LLVM_DEBUG(verifyFunction(*L->getHeader()->getParent()));<br>
>> >> return true;<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/new-pm-defaults.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/new-pm-defaults.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/new-pm-defaults.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/new-pm-defaults.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/new-pm-defaults.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -246,6 +246,7 @@<br>
>> >> ; CHECK-O-NEXT: Running pass: InstCombinePass<br>
>> >> ; CHECK-O-NEXT: Running pass: LoopUnrollPass<br>
>> >> ; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<br>
>> >> +; CHECK-O-NEXT: Running pass: WarnMissedTransformationsPass<br>
>> >> ; CHECK-O-NEXT: Running pass: InstCombinePass<br>
>> >> ; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}OptimizationRemarkEmitterAnalysis<br>
>> >> ; CHECK-O-NEXT: Running pass: FunctionToLoopPassAdaptor<{{.*}}LICMPass<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/new-pm-thinlto-defaults.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/new-pm-thinlto-defaults.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/new-pm-thinlto-defaults.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/new-pm-thinlto-defaults.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/new-pm-thinlto-defaults.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -224,6 +224,7 @@<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running pass: LoopUnrollPass<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<br>
>> >> +; CHECK-POSTLINK-O-NEXT: Running pass: WarnMissedTransformationsPass<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}OptimizationRemarkEmitterAnalysis<br>
>> >> ; CHECK-POSTLINK-O-NEXT: Running pass: FunctionToLoopPassAdaptor<{{.*}}LICMPass<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/opt-O2-pipeline.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-O2-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-O2-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/opt-O2-pipeline.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/opt-O2-pipeline.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -250,6 +250,10 @@<br>
>> >> ; CHECK-NEXT: Scalar Evolution Analysis<br>
>> >> ; CHECK-NEXT: Loop Pass Manager<br>
>> >> ; CHECK-NEXT: Loop Invariant Code Motion<br>
>> >> +; CHECK-NEXT: Lazy Branch Probability Analysis<br>
>> >> +; CHECK-NEXT: Lazy Block Frequency Analysis<br>
>> >> +; CHECK-NEXT: Optimization Remark Emitter<br>
>> >> +; CHECK-NEXT: Warn about non-applied transformations<br>
>> >> ; CHECK-NEXT: Alignment from assumptions<br>
>> >> ; CHECK-NEXT: Strip Unused Function Prototypes<br>
>> >> ; CHECK-NEXT: Dead Global Elimination<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/opt-O3-pipeline.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-O3-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-O3-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/opt-O3-pipeline.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/opt-O3-pipeline.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -255,6 +255,10 @@<br>
>> >> ; CHECK-NEXT: Scalar Evolution Analysis<br>
>> >> ; CHECK-NEXT: Loop Pass Manager<br>
>> >> ; CHECK-NEXT: Loop Invariant Code Motion<br>
>> >> +; CHECK-NEXT: Lazy Branch Probability Analysis<br>
>> >> +; CHECK-NEXT: Lazy Block Frequency Analysis<br>
>> >> +; CHECK-NEXT: Optimization Remark Emitter<br>
>> >> +; CHECK-NEXT: Warn about non-applied transformations<br>
>> >> ; CHECK-NEXT: Alignment from assumptions<br>
>> >> ; CHECK-NEXT: Strip Unused Function Prototypes<br>
>> >> ; CHECK-NEXT: Dead Global Elimination<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/opt-Os-pipeline.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-Os-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-Os-pipeline.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/opt-Os-pipeline.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/opt-Os-pipeline.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -237,6 +237,10 @@<br>
>> >> ; CHECK-NEXT: Scalar Evolution Analysis<br>
>> >> ; CHECK-NEXT: Loop Pass Manager<br>
>> >> ; CHECK-NEXT: Loop Invariant Code Motion<br>
>> >> +; CHECK-NEXT: Lazy Branch Probability Analysis<br>
>> >> +; CHECK-NEXT: Lazy Block Frequency Analysis<br>
>> >> +; CHECK-NEXT: Optimization Remark Emitter<br>
>> >> +; CHECK-NEXT: Warn about non-applied transformations<br>
>> >> ; CHECK-NEXT: Alignment from assumptions<br>
>> >> ; CHECK-NEXT: Strip Unused Function Prototypes<br>
>> >> ; CHECK-NEXT: Dead Global Elimination<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Other/opt-hot-cold-split.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-hot-cold-split.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Other/opt-hot-cold-split.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Other/opt-hot-cold-split.ll (original)<br>
>> >> +++ llvm/trunk/test/Other/opt-hot-cold-split.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -236,6 +236,10 @@<br>
>> >> ; CHECK-NEXT: Scalar Evolution Analysis<br>
>> >> ; CHECK-NEXT: Loop Pass Manager<br>
>> >> ; CHECK-NEXT: Loop Invariant Code Motion<br>
>> >> +; CHECK-NEXT: Lazy Branch Probability Analysis<br>
>> >> +; CHECK-NEXT: Lazy Block Frequency Analysis<br>
>> >> +; CHECK-NEXT: Optimization Remark Emitter<br>
>> >> +; CHECK-NEXT: Warn about non-applied transformations<br>
>> >> ; CHECK-NEXT: Alignment from assumptions<br>
>> >> ; CHECK-NEXT: Strip Unused Function Prototypes<br>
>> >> ; CHECK-NEXT: Dead Global Elimination<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,50 @@<br>
>> >> +; RUN: opt -loop-distribute -enable-loop-distribute=1 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the disable_nonforced is honored by loop distribution.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced(<br>
>> >> +; CHECK-NOT: for.body.ldist1:<br>
>> >> +define void @disable_nonforced(i32* noalias %a,<br>
>> >> + i32* noalias %b,<br>
>> >> + i32* noalias %c,<br>
>> >> + i32* noalias %d,<br>
>> >> + i32* noalias %e) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]<br>
>> >> +<br>
>> >> + %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind<br>
>> >> + %loadA = load i32, i32* %arrayidxA, align 4<br>
>> >> +<br>
>> >> + %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind<br>
>> >> + %loadB = load i32, i32* %arrayidxB, align 4<br>
>> >> +<br>
>> >> + %mulA = mul i32 %loadB, %loadA<br>
>> >> +<br>
>> >> + %add = add nuw nsw i64 %ind, 1<br>
>> >> + %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add<br>
>> >> + store i32 %mulA, i32* %arrayidxA_plus_4, align 4<br>
>> >> +<br>
>> >> + %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind<br>
>> >> + %loadD = load i32, i32* %arrayidxD, align 4<br>
>> >> +<br>
>> >> + %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind<br>
>> >> + %loadE = load i32, i32* %arrayidxE, align 4<br>
>> >> +<br>
>> >> + %mulC = mul i32 %loadD, %loadE<br>
>> >> +<br>
>> >> + %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind<br>
>> >> + store i32 %mulC, i32* %arrayidxC, align 4<br>
>> >> +<br>
>> >> + %exitcond = icmp eq i64 %add, 20<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !{!"llvm.loop.disable_nonforced"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopDistribute/disable_nonforced_enable.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,51 @@<br>
>> >> +; RUN: opt -loop-distribute -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that llvm.loop.distribute.enable overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced(<br>
>> >> +; CHECK: for.body.ldist1:<br>
>> >> +define void @disable_nonforced(i32* noalias %a,<br>
>> >> + i32* noalias %b,<br>
>> >> + i32* noalias %c,<br>
>> >> + i32* noalias %d,<br>
>> >> + i32* noalias %e) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]<br>
>> >> +<br>
>> >> + %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind<br>
>> >> + %loadA = load i32, i32* %arrayidxA, align 4<br>
>> >> +<br>
>> >> + %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind<br>
>> >> + %loadB = load i32, i32* %arrayidxB, align 4<br>
>> >> +<br>
>> >> + %mulA = mul i32 %loadB, %loadA<br>
>> >> +<br>
>> >> + %add = add nuw nsw i64 %ind, 1<br>
>> >> + %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add<br>
>> >> + store i32 %mulA, i32* %arrayidxA_plus_4, align 4<br>
>> >> +<br>
>> >> + %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind<br>
>> >> + %loadD = load i32, i32* %arrayidxD, align 4<br>
>> >> +<br>
>> >> + %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind<br>
>> >> + %loadE = load i32, i32* %arrayidxE, align 4<br>
>> >> +<br>
>> >> + %mulC = mul i32 %loadD, %loadE<br>
>> >> +<br>
>> >> + %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind<br>
>> >> + store i32 %mulC, i32* %arrayidxC, align 4<br>
>> >> +<br>
>> >> + %exitcond = icmp eq i64 %add, 20<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.distribute.enable", i32 1}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopDistribute/followup.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/followup.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDistribute/followup.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopDistribute/followup.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopDistribute/followup.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,66 @@<br>
>> >> +; RUN: opt -basicaa -loop-distribute -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that followup loop-attributes are applied to the loops after<br>
>> >> +; loop distribution.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @f(i32* %a, i32* %b, i32* %c, i32* %d, i32* %e) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]<br>
>> >> +<br>
>> >> + %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind<br>
>> >> + %loadA = load i32, i32* %arrayidxA, align 4<br>
>> >> +<br>
>> >> + %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind<br>
>> >> + %loadB = load i32, i32* %arrayidxB, align 4<br>
>> >> +<br>
>> >> + %mulA = mul i32 %loadB, %loadA<br>
>> >> +<br>
>> >> + %add = add nuw nsw i64 %ind, 1<br>
>> >> + %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add<br>
>> >> + store i32 %mulA, i32* %arrayidxA_plus_4, align 4<br>
>> >> +<br>
>> >> + %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind<br>
>> >> + %loadD = load i32, i32* %arrayidxD, align 4<br>
>> >> +<br>
>> >> + %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind<br>
>> >> + %loadE = load i32, i32* %arrayidxE, align 4<br>
>> >> +<br>
>> >> + %mulC = mul i32 %loadD, %loadE<br>
>> >> +<br>
>> >> + %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind<br>
>> >> + store i32 %mulC, i32* %arrayidxC, align 4<br>
>> >> +<br>
>> >> + %exitcond = icmp eq i64 %add, 20<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !1, !2, !3, !4, !5}<br>
>> >> +!1 = !{!"llvm.loop.distribute.enable", i1 true}<br>
>> >> +!2 = !{!"llvm.loop.distribute.followup_all", !{!"FollowupAll"}}<br>
>> >> +!3 = !{!"llvm.loop.distribute.followup_coincident", !{!"FollowupCoincident", i1 false}}<br>
>> >> +!4 = !{!"llvm.loop.distribute.followup_sequential", !{!"FollowupSequential", i32 8}}<br>
>> >> +!5 = !{!"llvm.loop.distribute.followup_fallback", !{!"FollowupFallback"}}<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK-LABEL: for.body.lver.orig:<br>
>> >> +; CHECK: br i1 %exitcond.lver.orig, label %for.end, label %for.body.lver.orig, !llvm.loop ![[LOOP_ORIG:[0-9]+]]<br>
>> >> +; CHECK-LABEL: for.body.ldist1:<br>
>> >> +; CHECK: br i1 %exitcond.ldist1, label %<a href="http://for.body.ph" rel="noreferrer" target="_blank">for.body.ph</a>, label %for.body.ldist1, !llvm.loop ![[LOOP_SEQUENTIAL:[0-9]+]]<br>
>> >> +; CHECK-LABEL: for.body:<br>
>> >> +; CHECK: br i1 %exitcond, label %for.end, label %for.body, !llvm.loop ![[LOOP_COINCIDENT:[0-9]+]]<br>
>> >> +<br>
>> >> +; CHECK: ![[LOOP_ORIG]] = distinct !{![[LOOP_ORIG]], ![[FOLLOWUP_ALL:[0-9]+]], ![[FOLLOUP_FALLBACK:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_ALL]] = !{!"FollowupAll"}<br>
>> >> +; CHECK: ![[FOLLOUP_FALLBACK]] = !{!"FollowupFallback"}<br>
>> >> +; CHECK: ![[LOOP_SEQUENTIAL]] = distinct !{![[LOOP_SEQUENTIAL]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_SEQUENTIAL:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_SEQUENTIAL]] = !{!"FollowupSequential", i32 8}<br>
>> >> +; CHECK: ![[LOOP_COINCIDENT]] = distinct !{![[LOOP_COINCIDENT]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_COINCIDENT:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_COINCIDENT]] = !{!"FollowupCoincident", i1 false}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopTransformWarning/distribution-remarks-missed.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,99 @@<br>
>> >> +; Legacy pass manager<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +; New pass manager<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK: warning: source.cpp:19:5: loop not distributed: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >> +<br>
>> >> +; YAML: --- !Failure<br>
>> >> +; YAML-NEXT: Pass: transform-warning<br>
>> >> +; YAML-NEXT: Name: FailedRequestedDistribution<br>
>> >> +; YAML-NEXT: DebugLoc: { File: source.cpp, Line: 19, Column: 5 }<br>
>> >> +; YAML-NEXT: Function: _Z17test_array_boundsPiS_i<br>
>> >> +; YAML-NEXT: Args:<br>
>> >> +; YAML-NEXT: - String: 'loop not distributed: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering'<br>
>> >> +; YAML-NEXT: ...<br>
>> >> +<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @_Z17test_array_boundsPiS_i(i32* nocapture %A, i32* nocapture readonly %B, i32 %Length) !dbg !8 {<br>
>> >> +entry:<br>
>> >> + %cmp9 = icmp sgt i32 %Length, 0, !dbg !32<br>
>> >> + br i1 %cmp9, label %for.body.preheader, label %for.end, !dbg !32<br>
>> >> +<br>
>> >> +for.body.preheader:<br>
>> >> + br label %for.body, !dbg !35<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv, !dbg !35<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4, !dbg !35, !tbaa !18<br>
>> >> + %idxprom1 = sext i32 %0 to i64, !dbg !35<br>
>> >> + %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %idxprom1, !dbg !35<br>
>> >> + %1 = load i32, i32* %arrayidx2, align 4, !dbg !35, !tbaa !18<br>
>> >> + %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !35<br>
>> >> + store i32 %1, i32* %arrayidx4, align 4, !dbg !35, !tbaa !18<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !32<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !32<br>
>> >> + br i1 %exitcond, label %for.end.loopexit, label %for.body, !dbg !32, !llvm.loop !50<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void, !dbg !36<br>
>> >> +}<br>
>> >> +<br>
>> >> +!<a href="http://llvm.dbg.cu" rel="noreferrer" target="_blank">llvm.dbg.cu</a> = !{!0}<br>
>> >> +!llvm.module.flags = !{!9, !10}<br>
>> >> +!llvm.ident = !{!11}<br>
>> >> +<br>
>> >> +!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, producer: "clang version 3.5.0", isOptimized: true, runtimeVersion: 6, emissionKind: LineTablesOnly, file: !1, enums: !2, retainedTypes: !2, globals: !2, imports: !2)<br>
>> >> +!1 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!2 = !{}<br>
>> >> +!4 = distinct !DISubprogram(name: "test", line: 1, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 1, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!5 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!6 = !DISubroutineType(types: !2)<br>
>> >> +!7 = distinct !DISubprogram(name: "test_disabled", line: 10, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 10, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!8 = distinct !DISubprogram(name: "test_array_bounds", line: 16, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 16, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!9 = !{i32 2, !"Dwarf Version", i32 2}<br>
>> >> +!10 = !{i32 2, !"Debug Info Version", i32 3}<br>
>> >> +!11 = !{!"clang version 3.5.0"}<br>
>> >> +!12 = !DILocation(line: 3, column: 8, scope: !13)<br>
>> >> +!13 = distinct !DILexicalBlock(line: 3, column: 3, file: !1, scope: !4)<br>
>> >> +!16 = !DILocation(line: 4, column: 5, scope: !17)<br>
>> >> +!17 = distinct !DILexicalBlock(line: 3, column: 36, file: !1, scope: !13)<br>
>> >> +!18 = !{!19, !19, i64 0}<br>
>> >> +!19 = !{!"int", !20, i64 0}<br>
>> >> +!20 = !{!"omnipotent char", !21, i64 0}<br>
>> >> +!21 = !{!"Simple C/C++ TBAA"}<br>
>> >> +!22 = !DILocation(line: 5, column: 9, scope: !23)<br>
>> >> +!23 = distinct !DILexicalBlock(line: 5, column: 9, file: !1, scope: !17)<br>
>> >> +!24 = !DILocation(line: 8, column: 1, scope: !4)<br>
>> >> +!25 = !DILocation(line: 12, column: 8, scope: !26)<br>
>> >> +!26 = distinct !DILexicalBlock(line: 12, column: 3, file: !1, scope: !7)<br>
>> >> +!30 = !DILocation(line: 13, column: 5, scope: !26)<br>
>> >> +!31 = !DILocation(line: 14, column: 1, scope: !7)<br>
>> >> +!32 = !DILocation(line: 18, column: 8, scope: !33)<br>
>> >> +!33 = distinct !DILexicalBlock(line: 18, column: 3, file: !1, scope: !8)<br>
>> >> +!35 = !DILocation(line: 19, column: 5, scope: !33)<br>
>> >> +!36 = !DILocation(line: 20, column: 1, scope: !8)<br>
>> >> +!37 = distinct !DILexicalBlock(line: 24, column: 3, file: !1, scope: !46)<br>
>> >> +!38 = !DILocation(line: 27, column: 3, scope: !37)<br>
>> >> +!39 = !DILocation(line: 31, column: 3, scope: !37)<br>
>> >> +!40 = !DILocation(line: 28, column: 9, scope: !37)<br>
>> >> +!41 = !DILocation(line: 29, column: 11, scope: !37)<br>
>> >> +!42 = !DILocation(line: 29, column: 7, scope: !37)<br>
>> >> +!43 = !DILocation(line: 27, column: 32, scope: !37)<br>
>> >> +!44 = !DILocation(line: 27, column: 30, scope: !37)<br>
>> >> +!45 = !DILocation(line: 27, column: 21, scope: !37)<br>
>> >> +!46 = distinct !DISubprogram(name: "test_multiple_failures", line: 26, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 26, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +<br>
>> >> +!50 = !{!50, !{!"llvm.loop.distribute.enable"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopTransformWarning/unrollandjam-remarks-missed.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,99 @@<br>
>> >> +; Legacy pass manager<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +; New pass manager<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK: warning: source.cpp:19:5: loop not unroll-and-jammed: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >> +<br>
>> >> +; YAML: --- !Failure<br>
>> >> +; YAML-NEXT: Pass: transform-warning<br>
>> >> +; YAML-NEXT: Name: FailedRequestedUnrollAndJamming<br>
>> >> +; YAML-NEXT: DebugLoc: { File: source.cpp, Line: 19, Column: 5 }<br>
>> >> +; YAML-NEXT: Function: _Z17test_array_boundsPiS_i<br>
>> >> +; YAML-NEXT: Args:<br>
>> >> +; YAML-NEXT: - String: 'loop not unroll-and-jammed: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering'<br>
>> >> +; YAML-NEXT: ...<br>
>> >> +<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @_Z17test_array_boundsPiS_i(i32* nocapture %A, i32* nocapture readonly %B, i32 %Length) !dbg !8 {<br>
>> >> +entry:<br>
>> >> + %cmp9 = icmp sgt i32 %Length, 0, !dbg !32<br>
>> >> + br i1 %cmp9, label %for.body.preheader, label %for.end, !dbg !32<br>
>> >> +<br>
>> >> +for.body.preheader:<br>
>> >> + br label %for.body, !dbg !35<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv, !dbg !35<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4, !dbg !35, !tbaa !18<br>
>> >> + %idxprom1 = sext i32 %0 to i64, !dbg !35<br>
>> >> + %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %idxprom1, !dbg !35<br>
>> >> + %1 = load i32, i32* %arrayidx2, align 4, !dbg !35, !tbaa !18<br>
>> >> + %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !35<br>
>> >> + store i32 %1, i32* %arrayidx4, align 4, !dbg !35, !tbaa !18<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !32<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !32<br>
>> >> + br i1 %exitcond, label %for.end.loopexit, label %for.body, !dbg !32, !llvm.loop !50<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void, !dbg !36<br>
>> >> +}<br>
>> >> +<br>
>> >> +!<a href="http://llvm.dbg.cu" rel="noreferrer" target="_blank">llvm.dbg.cu</a> = !{!0}<br>
>> >> +!llvm.module.flags = !{!9, !10}<br>
>> >> +!llvm.ident = !{!11}<br>
>> >> +<br>
>> >> +!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, producer: "clang version 3.5.0", isOptimized: true, runtimeVersion: 6, emissionKind: LineTablesOnly, file: !1, enums: !2, retainedTypes: !2, globals: !2, imports: !2)<br>
>> >> +!1 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!2 = !{}<br>
>> >> +!4 = distinct !DISubprogram(name: "test", line: 1, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 1, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!5 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!6 = !DISubroutineType(types: !2)<br>
>> >> +!7 = distinct !DISubprogram(name: "test_disabled", line: 10, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 10, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!8 = distinct !DISubprogram(name: "test_array_bounds", line: 16, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 16, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!9 = !{i32 2, !"Dwarf Version", i32 2}<br>
>> >> +!10 = !{i32 2, !"Debug Info Version", i32 3}<br>
>> >> +!11 = !{!"clang version 3.5.0"}<br>
>> >> +!12 = !DILocation(line: 3, column: 8, scope: !13)<br>
>> >> +!13 = distinct !DILexicalBlock(line: 3, column: 3, file: !1, scope: !4)<br>
>> >> +!16 = !DILocation(line: 4, column: 5, scope: !17)<br>
>> >> +!17 = distinct !DILexicalBlock(line: 3, column: 36, file: !1, scope: !13)<br>
>> >> +!18 = !{!19, !19, i64 0}<br>
>> >> +!19 = !{!"int", !20, i64 0}<br>
>> >> +!20 = !{!"omnipotent char", !21, i64 0}<br>
>> >> +!21 = !{!"Simple C/C++ TBAA"}<br>
>> >> +!22 = !DILocation(line: 5, column: 9, scope: !23)<br>
>> >> +!23 = distinct !DILexicalBlock(line: 5, column: 9, file: !1, scope: !17)<br>
>> >> +!24 = !DILocation(line: 8, column: 1, scope: !4)<br>
>> >> +!25 = !DILocation(line: 12, column: 8, scope: !26)<br>
>> >> +!26 = distinct !DILexicalBlock(line: 12, column: 3, file: !1, scope: !7)<br>
>> >> +!30 = !DILocation(line: 13, column: 5, scope: !26)<br>
>> >> +!31 = !DILocation(line: 14, column: 1, scope: !7)<br>
>> >> +!32 = !DILocation(line: 18, column: 8, scope: !33)<br>
>> >> +!33 = distinct !DILexicalBlock(line: 18, column: 3, file: !1, scope: !8)<br>
>> >> +!35 = !DILocation(line: 19, column: 5, scope: !33)<br>
>> >> +!36 = !DILocation(line: 20, column: 1, scope: !8)<br>
>> >> +!37 = distinct !DILexicalBlock(line: 24, column: 3, file: !1, scope: !46)<br>
>> >> +!38 = !DILocation(line: 27, column: 3, scope: !37)<br>
>> >> +!39 = !DILocation(line: 31, column: 3, scope: !37)<br>
>> >> +!40 = !DILocation(line: 28, column: 9, scope: !37)<br>
>> >> +!41 = !DILocation(line: 29, column: 11, scope: !37)<br>
>> >> +!42 = !DILocation(line: 29, column: 7, scope: !37)<br>
>> >> +!43 = !DILocation(line: 27, column: 32, scope: !37)<br>
>> >> +!44 = !DILocation(line: 27, column: 30, scope: !37)<br>
>> >> +!45 = !DILocation(line: 27, column: 21, scope: !37)<br>
>> >> +!46 = distinct !DISubprogram(name: "test_multiple_failures", line: 26, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 26, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +<br>
>> >> +!50 = !{!50, !{!"llvm.loop.unroll_and_jam.enable"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopTransformWarning/unrolling-remarks-missed.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,99 @@<br>
>> >> +; Legacy pass manager<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +; New pass manager<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK: warning: source.cpp:19:5: loop not unrolled: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >> +<br>
>> >> +; YAML: --- !Failure<br>
>> >> +; YAML-NEXT: Pass: transform-warning<br>
>> >> +; YAML-NEXT: Name: FailedRequestedUnrolling<br>
>> >> +; YAML-NEXT: DebugLoc: { File: source.cpp, Line: 19, Column: 5 }<br>
>> >> +; YAML-NEXT: Function: _Z17test_array_boundsPiS_i<br>
>> >> +; YAML-NEXT: Args:<br>
>> >> +; YAML-NEXT: - String: 'loop not unrolled: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering'<br>
>> >> +; YAML-NEXT: ...<br>
>> >> +<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @_Z17test_array_boundsPiS_i(i32* nocapture %A, i32* nocapture readonly %B, i32 %Length) !dbg !8 {<br>
>> >> +entry:<br>
>> >> + %cmp9 = icmp sgt i32 %Length, 0, !dbg !32<br>
>> >> + br i1 %cmp9, label %for.body.preheader, label %for.end, !dbg !32<br>
>> >> +<br>
>> >> +for.body.preheader:<br>
>> >> + br label %for.body, !dbg !35<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv, !dbg !35<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4, !dbg !35, !tbaa !18<br>
>> >> + %idxprom1 = sext i32 %0 to i64, !dbg !35<br>
>> >> + %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %idxprom1, !dbg !35<br>
>> >> + %1 = load i32, i32* %arrayidx2, align 4, !dbg !35, !tbaa !18<br>
>> >> + %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !35<br>
>> >> + store i32 %1, i32* %arrayidx4, align 4, !dbg !35, !tbaa !18<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !32<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !32<br>
>> >> + br i1 %exitcond, label %for.end.loopexit, label %for.body, !dbg !32, !llvm.loop !50<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void, !dbg !36<br>
>> >> +}<br>
>> >> +<br>
>> >> +!<a href="http://llvm.dbg.cu" rel="noreferrer" target="_blank">llvm.dbg.cu</a> = !{!0}<br>
>> >> +!llvm.module.flags = !{!9, !10}<br>
>> >> +!llvm.ident = !{!11}<br>
>> >> +<br>
>> >> +!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, producer: "clang version 3.5.0", isOptimized: true, runtimeVersion: 6, emissionKind: LineTablesOnly, file: !1, enums: !2, retainedTypes: !2, globals: !2, imports: !2)<br>
>> >> +!1 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!2 = !{}<br>
>> >> +!4 = distinct !DISubprogram(name: "test", line: 1, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 1, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!5 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!6 = !DISubroutineType(types: !2)<br>
>> >> +!7 = distinct !DISubprogram(name: "test_disabled", line: 10, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 10, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!8 = distinct !DISubprogram(name: "test_array_bounds", line: 16, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 16, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!9 = !{i32 2, !"Dwarf Version", i32 2}<br>
>> >> +!10 = !{i32 2, !"Debug Info Version", i32 3}<br>
>> >> +!11 = !{!"clang version 3.5.0"}<br>
>> >> +!12 = !DILocation(line: 3, column: 8, scope: !13)<br>
>> >> +!13 = distinct !DILexicalBlock(line: 3, column: 3, file: !1, scope: !4)<br>
>> >> +!16 = !DILocation(line: 4, column: 5, scope: !17)<br>
>> >> +!17 = distinct !DILexicalBlock(line: 3, column: 36, file: !1, scope: !13)<br>
>> >> +!18 = !{!19, !19, i64 0}<br>
>> >> +!19 = !{!"int", !20, i64 0}<br>
>> >> +!20 = !{!"omnipotent char", !21, i64 0}<br>
>> >> +!21 = !{!"Simple C/C++ TBAA"}<br>
>> >> +!22 = !DILocation(line: 5, column: 9, scope: !23)<br>
>> >> +!23 = distinct !DILexicalBlock(line: 5, column: 9, file: !1, scope: !17)<br>
>> >> +!24 = !DILocation(line: 8, column: 1, scope: !4)<br>
>> >> +!25 = !DILocation(line: 12, column: 8, scope: !26)<br>
>> >> +!26 = distinct !DILexicalBlock(line: 12, column: 3, file: !1, scope: !7)<br>
>> >> +!30 = !DILocation(line: 13, column: 5, scope: !26)<br>
>> >> +!31 = !DILocation(line: 14, column: 1, scope: !7)<br>
>> >> +!32 = !DILocation(line: 18, column: 8, scope: !33)<br>
>> >> +!33 = distinct !DILexicalBlock(line: 18, column: 3, file: !1, scope: !8)<br>
>> >> +!35 = !DILocation(line: 19, column: 5, scope: !33)<br>
>> >> +!36 = !DILocation(line: 20, column: 1, scope: !8)<br>
>> >> +!37 = distinct !DILexicalBlock(line: 24, column: 3, file: !1, scope: !46)<br>
>> >> +!38 = !DILocation(line: 27, column: 3, scope: !37)<br>
>> >> +!39 = !DILocation(line: 31, column: 3, scope: !37)<br>
>> >> +!40 = !DILocation(line: 28, column: 9, scope: !37)<br>
>> >> +!41 = !DILocation(line: 29, column: 11, scope: !37)<br>
>> >> +!42 = !DILocation(line: 29, column: 7, scope: !37)<br>
>> >> +!43 = !DILocation(line: 27, column: 32, scope: !37)<br>
>> >> +!44 = !DILocation(line: 27, column: 30, scope: !37)<br>
>> >> +!45 = !DILocation(line: 27, column: 21, scope: !37)<br>
>> >> +!46 = distinct !DISubprogram(name: "test_multiple_failures", line: 26, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 26, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +<br>
>> >> +!50 = !{!50, !{!"llvm.loop.unroll.enable"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopTransformWarning/vectorization-remarks-missed.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,113 @@<br>
>> >> +; Legacy pass manager<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +; New pass manager<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-missed=transform-warning -pass-remarks-analysis=transform-warning 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -passes=transform-warning -disable-output -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >> +<br>
>> >> +<br>
>> >> +; C/C++ code for tests<br>
>> >> +; void test(int *A, int Length) {<br>
>> >> +; #pragma clang loop vectorize(enable) interleave(enable)<br>
>> >> +; for (int i = 0; i < Length; i++) {<br>
>> >> +; A[i] = i;<br>
>> >> +; if (A[i] > Length)<br>
>> >> +; break;<br>
>> >> +; }<br>
>> >> +; }<br>
>> >> +; File, line, and column should match those specified in the metadata<br>
>> >> +; CHECK: warning: source.cpp:19:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >> +<br>
>> >> +; YAML: --- !Failure<br>
>> >> +; YAML-NEXT: Pass: transform-warning<br>
>> >> +; YAML-NEXT: Name: FailedRequestedVectorization<br>
>> >> +; YAML-NEXT: DebugLoc: { File: source.cpp, Line: 19, Column: 5 }<br>
>> >> +; YAML-NEXT: Function: _Z17test_array_boundsPiS_i<br>
>> >> +; YAML-NEXT: Args:<br>
>> >> +; YAML-NEXT: - String: 'loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering'<br>
>> >> +; YAML-NEXT: ...<br>
>> >> +<br>
>> >> +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @_Z17test_array_boundsPiS_i(i32* nocapture %A, i32* nocapture readonly %B, i32 %Length) !dbg !8 {<br>
>> >> +entry:<br>
>> >> + %cmp9 = icmp sgt i32 %Length, 0, !dbg !32<br>
>> >> + br i1 %cmp9, label %for.body.preheader, label %for.end, !dbg !32, !llvm.loop !34<br>
>> >> +<br>
>> >> +for.body.preheader:<br>
>> >> + br label %for.body, !dbg !35<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv, !dbg !35<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4, !dbg !35, !tbaa !18<br>
>> >> + %idxprom1 = sext i32 %0 to i64, !dbg !35<br>
>> >> + %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %idxprom1, !dbg !35<br>
>> >> + %1 = load i32, i32* %arrayidx2, align 4, !dbg !35, !tbaa !18<br>
>> >> + %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !35<br>
>> >> + store i32 %1, i32* %arrayidx4, align 4, !dbg !35, !tbaa !18<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !32<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !32<br>
>> >> + br i1 %exitcond, label %for.end.loopexit, label %for.body, !dbg !32, !llvm.loop !34<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void, !dbg !36<br>
>> >> +}<br>
>> >> +<br>
>> >> +!<a href="http://llvm.dbg.cu" rel="noreferrer" target="_blank">llvm.dbg.cu</a> = !{!0}<br>
>> >> +!llvm.module.flags = !{!9, !10}<br>
>> >> +!llvm.ident = !{!11}<br>
>> >> +<br>
>> >> +!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, producer: "clang version 3.5.0", isOptimized: true, runtimeVersion: 6, emissionKind: LineTablesOnly, file: !1, enums: !2, retainedTypes: !2, globals: !2, imports: !2)<br>
>> >> +!1 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!2 = !{}<br>
>> >> +!4 = distinct !DISubprogram(name: "test", line: 1, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 1, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!5 = !DIFile(filename: "source.cpp", directory: ".")<br>
>> >> +!6 = !DISubroutineType(types: !2)<br>
>> >> +!7 = distinct !DISubprogram(name: "test_disabled", line: 10, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 10, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!8 = distinct !DISubprogram(name: "test_array_bounds", line: 16, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 16, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >> +!9 = !{i32 2, !"Dwarf Version", i32 2}<br>
>> >> +!10 = !{i32 2, !"Debug Info Version", i32 3}<br>
>> >> +!11 = !{!"clang version 3.5.0"}<br>
>> >> +!12 = !DILocation(line: 3, column: 8, scope: !13)<br>
>> >> +!13 = distinct !DILexicalBlock(line: 3, column: 3, file: !1, scope: !4)<br>
>> >> +!14 = !{!14, !15, !15}<br>
>> >> +!15 = !{!"llvm.loop.vectorize.enable", i1 true}<br>
>> >> +!16 = !DILocation(line: 4, column: 5, scope: !17)<br>
>> >> +!17 = distinct !DILexicalBlock(line: 3, column: 36, file: !1, scope: !13)<br>
>> >> +!18 = !{!19, !19, i64 0}<br>
>> >> +!19 = !{!"int", !20, i64 0}<br>
>> >> +!20 = !{!"omnipotent char", !21, i64 0}<br>
>> >> +!21 = !{!"Simple C/C++ TBAA"}<br>
>> >> +!22 = !DILocation(line: 5, column: 9, scope: !23)<br>
>> >> +!23 = distinct !DILexicalBlock(line: 5, column: 9, file: !1, scope: !17)<br>
>> >> +!24 = !DILocation(line: 8, column: 1, scope: !4)<br>
>> >> +!25 = !DILocation(line: 12, column: 8, scope: !26)<br>
>> >> +!26 = distinct !DILexicalBlock(line: 12, column: 3, file: !1, scope: !7)<br>
>> >> +!27 = !{!27, !28, !29}<br>
>> >> +!28 = !{!"llvm.loop.interleave.count", i32 1}<br>
>> >> +!29 = !{!"llvm.loop.vectorize.width", i32 1}<br>
>> >> +!30 = !DILocation(line: 13, column: 5, scope: !26)<br>
>> >> +!31 = !DILocation(line: 14, column: 1, scope: !7)<br>
>> >> +!32 = !DILocation(line: 18, column: 8, scope: !33)<br>
>> >> +!33 = distinct !DILexicalBlock(line: 18, column: 3, file: !1, scope: !8)<br>
>> >> +!34 = !{!34, !15}<br>
>> >> +!35 = !DILocation(line: 19, column: 5, scope: !33)<br>
>> >> +!36 = !DILocation(line: 20, column: 1, scope: !8)<br>
>> >> +!37 = distinct !DILexicalBlock(line: 24, column: 3, file: !1, scope: !46)<br>
>> >> +!38 = !DILocation(line: 27, column: 3, scope: !37)<br>
>> >> +!39 = !DILocation(line: 31, column: 3, scope: !37)<br>
>> >> +!40 = !DILocation(line: 28, column: 9, scope: !37)<br>
>> >> +!41 = !DILocation(line: 29, column: 11, scope: !37)<br>
>> >> +!42 = !DILocation(line: 29, column: 7, scope: !37)<br>
>> >> +!43 = !DILocation(line: 27, column: 32, scope: !37)<br>
>> >> +!44 = !DILocation(line: 27, column: 30, scope: !37)<br>
>> >> +!45 = !DILocation(line: 27, column: 21, scope: !37)<br>
>> >> +!46 = distinct !DISubprogram(name: "test_multiple_failures", line: 26, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 26, file: !1, scope: !5, type: !6, retainedNodes: !2)<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,29 @@<br>
>> >> +; RUN: opt -loop-unroll -unroll-count=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the disable_nonforced loop property is honored by<br>
>> >> +; loop unroll.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced(<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK-NOT: load<br>
>> >> +define void @disable_nonforced(i32* nocapture %a) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4<br>
>> >> + %inc = add nsw i32 %0, 1<br>
>> >> + store i32 %inc, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
>> >> + %exitcond = icmp eq i64 %indvars.iv.next, 64<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_count.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,30 @@<br>
>> >> +; RUN: opt -loop-unroll -unroll-count=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check whether the llvm.loop.unroll.count loop property overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced_count(<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK-NOT: store<br>
>> >> +define void @disable_nonforced_count(i32* nocapture %a) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4<br>
>> >> + %inc = add nsw i32 %0, 1<br>
>> >> + store i32 %inc, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
>> >> + %exitcond = icmp eq i64 %indvars.iv.next, 64<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.unroll.count", i32 2}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_enable.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,30 @@<br>
>> >> +; RUN: opt -loop-unroll -unroll-count=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the llvm.loop.unroll.enable loop property overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced_enable(<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK-NOT: store<br>
>> >> +define void @disable_nonforced_enable(i32* nocapture %a) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4<br>
>> >> + %inc = add nsw i32 %0, 1<br>
>> >> + store i32 %inc, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
>> >> + %exitcond = icmp eq i64 %indvars.iv.next, 64<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.unroll.enable"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnroll/disable_nonforced_full.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,32 @@<br>
>> >> +; RUN: opt -loop-unroll -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the llvm.loop.unroll.full loop property overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced_full(<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK: store<br>
>> >> +; CHECK-NOT: store<br>
>> >> +define void @disable_nonforced_full(i32* nocapture %a) {<br>
>> >> +entry:<br>
>> >> + br label %for.body<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4<br>
>> >> + %inc = add nsw i32 %0, 1<br>
>> >> + store i32 %inc, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1<br>
>> >> + %exitcond = icmp eq i64 %indvars.iv.next, 4<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.unroll.full"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnroll/followup.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/followup.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/followup.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnroll/followup.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnroll/followup.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,63 @@<br>
>> >> +; RUN: opt < %s -S -loop-unroll -unroll-count=2 | FileCheck %s -check-prefixes=COUNT,COMMON<br>
>> >> +; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true | FileCheck %s -check-prefixes=EPILOG,COMMON<br>
>> >> +; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=PROLOG,COMMON<br>
>> >> +;<br>
>> >> +; Check that followup-attributes are applied after LoopUnroll.<br>
>> >> +;<br>
>> >> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define i32 @test(i32* nocapture %a, i32 %n) nounwind uwtable readonly {<br>
>> >> +entry:<br>
>> >> + %cmp1 = icmp eq i32 %n, 0<br>
>> >> + br i1 %cmp1, label %for.end, label %for.body<br>
>> >> +<br>
>> >> +for.body: ; preds = %for.body, %entry<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]<br>
>> >> + %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = load i32, i32* %arrayidx, align 4<br>
>> >> + %add = add nsw i32 %0, %sum.02<br>
>> >> + %indvars.iv.next = add i64 %indvars.iv, 1<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %n<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !4<br>
>> >> +<br>
>> >> +for.end: ; preds = %for.body, %entry<br>
>> >> + %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]<br>
>> >> + ret i32 %sum.0.lcssa<br>
>> >> +}<br>
>> >> +<br>
>> >> +!1 = !{!"llvm.loop.unroll.followup_all", !{!"FollowupAll"}}<br>
>> >> +!2 = !{!"llvm.loop.unroll.followup_unrolled", !{!"FollowupUnrolled"}}<br>
>> >> +!3 = !{!"llvm.loop.unroll.followup_remainder", !{!"FollowupRemainder"}}<br>
>> >> +!4 = distinct !{!4, !1, !2, !3}<br>
>> >> +<br>
>> >> +<br>
>> >> +; COMMON-LABEL: @test(<br>
>> >> +<br>
>> >> +<br>
>> >> +; COUNT: br i1 %exitcond.1, label %for.end.loopexit, label %for.body, !llvm.loop ![[LOOP:[0-9]+]]<br>
>> >> +<br>
>> >> +; COUNT: ![[FOLLOWUP_ALL:[0-9]+]] = !{!"FollowupAll"}<br>
>> >> +; COUNT: ![[FOLLOWUP_UNROLLED:[0-9]+]] = !{!"FollowupUnrolled"}<br>
>> >> +; COUNT: ![[LOOP]] = distinct !{![[LOOP]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_UNROLLED]]}<br>
>> >> +<br>
>> >> +<br>
>> >> +; EPILOG: br i1 %niter.ncmp.7, label %for.end.loopexit.unr-lcssa.loopexit, label %for.body, !llvm.loop ![[LOOP_0:[0-9]+]]<br>
>> >> +; EPILOG: br i1 %epil.iter.cmp, label %for.body.epil, label %for.end.loopexit.epilog-lcssa, !llvm.loop ![[LOOP_2:[0-9]+]]<br>
>> >> +<br>
>> >> +; EPILOG: ![[LOOP_0]] = distinct !{![[LOOP_0]], ![[FOLLOWUP_ALL:[0-9]+]], ![[FOLLOWUP_UNROLLED:[0-9]+]]}<br>
>> >> +; EPILOG: ![[FOLLOWUP_ALL]] = !{!"FollowupAll"}<br>
>> >> +; EPILOG: ![[FOLLOWUP_UNROLLED]] = !{!"FollowupUnrolled"}<br>
>> >> +; EPILOG: ![[LOOP_2]] = distinct !{![[LOOP_2]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_REMAINDER:[0-9]+]]}<br>
>> >> +; EPILOG: ![[FOLLOWUP_REMAINDER]] = !{!"FollowupRemainder"}<br>
>> >> +<br>
>> >> +<br>
>> >> +; PROLOG: br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit.unr-lcssa, !llvm.loop ![[LOOP_0:[0-9]+]]<br>
>> >> +; PROLOG: br i1 %exitcond.7, label %for.end.loopexit.unr-lcssa, label %for.body, !llvm.loop ![[LOOP_2:[0-9]+]]<br>
>> >> +<br>
>> >> +; PROLOG: ![[LOOP_0]] = distinct !{![[LOOP_0]], ![[FOLLOWUP_ALL:[0-9]+]], ![[FOLLOWUP_REMAINDER:[0-9]+]]}<br>
>> >> +; PROLOG: ![[FOLLOWUP_ALL]] = !{!"FollowupAll"}<br>
>> >> +; PROLOG: ![[FOLLOWUP_REMAINDER]] = !{!"FollowupRemainder"}<br>
>> >> +; PROLOG: ![[LOOP_2]] = distinct !{![[LOOP_2]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_UNROLLED:[0-9]+]]}<br>
>> >> +; PROLOG: ![[FOLLOWUP_UNROLLED]] = !{!"FollowupUnrolled"}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,50 @@<br>
>> >> +; RUN: opt -loop-unroll-and-jam -allow-unroll-and-jam -unroll-and-jam-count=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the disable_nonforced loop property is honored by<br>
>> >> +; loop unroll-and-jam.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: disable_nonforced<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK-NOT: load<br>
>> >> +define void @disable_nonforced(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) {<br>
>> >> +entry:<br>
>> >> + %cmp = icmp ne i32 %J, 0<br>
>> >> + %cmp122 = icmp ne i32 %I, 0<br>
>> >> + %or.cond = and i1 %cmp, %cmp122<br>
>> >> + br i1 %or.cond, label %for.outer.preheader, label %for.end<br>
>> >> +<br>
>> >> +for.outer.preheader:<br>
>> >> + br label %for.outer<br>
>> >> +<br>
>> >> +for.outer:<br>
>> >> + %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a> = phi i32 [ %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %for.latch ], [ 0, %for.outer.preheader ]<br>
>> >> + br label %for.inner<br>
>> >> +<br>
>> >> +for.inner:<br>
>> >> + %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %for.inner ]<br>
>> >> + %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a> = getelementptr inbounds i32, i32* %B, i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a><br>
>> >> + %0 = load i32, i32* %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a>, align 4<br>
>> >> + %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a> = add i32 %0, %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a><br>
>> >> + %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a> = add nuw i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a>, 1<br>
>> >> + %exitcond = icmp eq i32 %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %J<br>
>> >> + br i1 %exitcond, label %for.latch, label %for.inner<br>
>> >> +<br>
>> >> +for.latch:<br>
>> >> + %add.us.lcssa = phi i32 [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a> = getelementptr inbounds i32, i32* %A, i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a><br>
>> >> + store i32 %add.us.lcssa, i32* %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a>, align 4<br>
>> >> + %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a> = add nuw i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a>, 1<br>
>> >> + %exitcond25 = icmp eq i32 %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %I<br>
>> >> + br i1 %exitcond25, label %for.end.loopexit, label %for.outer, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !{!"llvm.loop.disable_nonforced"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_count.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,52 @@<br>
>> >> +; RUN: opt -loop-unroll-and-jam -allow-unroll-and-jam -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Verify that the llvm.loop.unroll_and_jam.count loop property overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced_enable(<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK-NOT: load<br>
>> >> +; CHECK: br i1<br>
>> >> +define void @disable_nonforced_enable(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) {<br>
>> >> +entry:<br>
>> >> + %cmp = icmp ne i32 %J, 0<br>
>> >> + %cmp122 = icmp ne i32 %I, 0<br>
>> >> + %or.cond = and i1 %cmp, %cmp122<br>
>> >> + br i1 %or.cond, label %for.outer.preheader, label %for.end<br>
>> >> +<br>
>> >> +for.outer.preheader:<br>
>> >> + br label %for.outer<br>
>> >> +<br>
>> >> +for.outer:<br>
>> >> + %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a> = phi i32 [ %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %for.latch ], [ 0, %for.outer.preheader ]<br>
>> >> + br label %for.inner<br>
>> >> +<br>
>> >> +for.inner:<br>
>> >> + %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %for.inner ]<br>
>> >> + %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a> = getelementptr inbounds i32, i32* %B, i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a><br>
>> >> + %0 = load i32, i32* %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a>, align 4<br>
>> >> + %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a> = add i32 %0, %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a><br>
>> >> + %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a> = add nuw i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a>, 1<br>
>> >> + %exitcond = icmp eq i32 %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %J<br>
>> >> + br i1 %exitcond, label %for.latch, label %for.inner<br>
>> >> +<br>
>> >> +for.latch:<br>
>> >> + %add.us.lcssa = phi i32 [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a> = getelementptr inbounds i32, i32* %A, i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a><br>
>> >> + store i32 %add.us.lcssa, i32* %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a>, align 4<br>
>> >> + %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a> = add nuw i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a>, 1<br>
>> >> + %exitcond25 = icmp eq i32 %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %I<br>
>> >> + br i1 %exitcond25, label %for.end.loopexit, label %for.outer, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.unroll_and_jam.count", i32 2}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnrollAndJam/disable_nonforced_enable.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,52 @@<br>
>> >> +; RUN: opt -loop-unroll-and-jam -allow-unroll-and-jam -unroll-and-jam-count=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Verify that the llvm.loop.unroll_and_jam.enable loop property<br>
>> >> +; overrides llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: disable_nonforced_enable<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK: load<br>
>> >> +; CHECK-NOT: load<br>
>> >> +; CHECK: br i1<br>
>> >> +define void @disable_nonforced_enable(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) {<br>
>> >> +entry:<br>
>> >> + %cmp = icmp ne i32 %J, 0<br>
>> >> + %cmp122 = icmp ne i32 %I, 0<br>
>> >> + %or.cond = and i1 %cmp, %cmp122<br>
>> >> + br i1 %or.cond, label %for.outer.preheader, label %for.end<br>
>> >> +<br>
>> >> +for.outer.preheader:<br>
>> >> + br label %for.outer<br>
>> >> +<br>
>> >> +for.outer:<br>
>> >> + %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a> = phi i32 [ %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %for.latch ], [ 0, %for.outer.preheader ]<br>
>> >> + br label %for.inner<br>
>> >> +<br>
>> >> +for.inner:<br>
>> >> + %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %for.inner ]<br>
>> >> + %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a> = getelementptr inbounds i32, i32* %B, i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a><br>
>> >> + %0 = load i32, i32* %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a>, align 4<br>
>> >> + %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a> = add i32 %0, %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a><br>
>> >> + %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a> = add nuw i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a>, 1<br>
>> >> + %exitcond = icmp eq i32 %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %J<br>
>> >> + br i1 %exitcond, label %for.latch, label %for.inner<br>
>> >> +<br>
>> >> +for.latch:<br>
>> >> + %add.us.lcssa = phi i32 [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a> = getelementptr inbounds i32, i32* %A, i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a><br>
>> >> + store i32 %add.us.lcssa, i32* %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a>, align 4<br>
>> >> + %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a> = add nuw i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a>, 1<br>
>> >> + %exitcond25 = icmp eq i32 %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %I<br>
>> >> + br i1 %exitcond25, label %for.end.loopexit, label %for.outer, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.unroll_and_jam.enable"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnrollAndJam/followup.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,66 @@<br>
>> >> +; RUN: opt -basicaa -tbaa -loop-unroll-and-jam -allow-unroll-and-jam -unroll-and-jam-count=4 -unroll-remainder < %s -S | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that followup attributes are set in the new loops.<br>
>> >> +;<br>
>> >> +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"<br>
>> >> +<br>
>> >> +define void @followup(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) {<br>
>> >> +entry:<br>
>> >> + %cmp = icmp ne i32 %J, 0<br>
>> >> + %cmp122 = icmp ne i32 %I, 0<br>
>> >> + %or.cond = and i1 %cmp, %cmp122<br>
>> >> + br i1 %or.cond, label %for.outer.preheader, label %for.end<br>
>> >> +<br>
>> >> +for.outer.preheader:<br>
>> >> + br label %for.outer<br>
>> >> +<br>
>> >> +for.outer:<br>
>> >> + %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a> = phi i32 [ %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %for.latch ], [ 0, %for.outer.preheader ]<br>
>> >> + br label %for.inner<br>
>> >> +<br>
>> >> +for.inner:<br>
>> >> + %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %for.inner ]<br>
>> >> + %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a> = phi i32 [ 0, %for.outer ], [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a> = getelementptr inbounds i32, i32* %B, i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a><br>
>> >> + %0 = load i32, i32* %<a href="http://arrayidx.us" rel="noreferrer" target="_blank">arrayidx.us</a>, align 4<br>
>> >> + %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a> = add i32 %0, %<a href="http://sum1.us" rel="noreferrer" target="_blank">sum1.us</a><br>
>> >> + %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a> = add nuw i32 %<a href="http://j.us" rel="noreferrer" target="_blank">j.us</a>, 1<br>
>> >> + %exitcond = icmp eq i32 %<a href="http://inc.us" rel="noreferrer" target="_blank">inc.us</a>, %J<br>
>> >> + br i1 %exitcond, label %for.latch, label %for.inner<br>
>> >> +<br>
>> >> +for.latch:<br>
>> >> + %add.us.lcssa = phi i32 [ %<a href="http://add.us" rel="noreferrer" target="_blank">add.us</a>, %for.inner ]<br>
>> >> + %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a> = getelementptr inbounds i32, i32* %A, i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a><br>
>> >> + store i32 %add.us.lcssa, i32* %<a href="http://arrayidx6.us" rel="noreferrer" target="_blank">arrayidx6.us</a>, align 4<br>
>> >> + %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a> = add nuw i32 %<a href="http://i.us" rel="noreferrer" target="_blank">i.us</a>, 1<br>
>> >> + %exitcond25 = icmp eq i32 %<a href="http://add8.us" rel="noreferrer" target="_blank">add8.us</a>, %I<br>
>> >> + br i1 %exitcond25, label %for.end.loopexit, label %for.outer, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end.loopexit:<br>
>> >> + br label %for.end<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !1, !2, !3, !4, !6}<br>
>> >> +!1 = !{!"llvm.loop.unroll_and_jam.enable"}<br>
>> >> +!2 = !{!"llvm.loop.unroll_and_jam.followup_outer", !{!"FollowupOuter"}}<br>
>> >> +!3 = !{!"llvm.loop.unroll_and_jam.followup_inner", !{!"FollowupInner"}}<br>
>> >> +!4 = !{!"llvm.loop.unroll_and_jam.followup_all", !{!"FollowupAll"}}<br>
>> >> +!6 = !{!"llvm.loop.unroll_and_jam.followup_remainder_inner", !{!"FollowupRemainderInner"}}<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK: br i1 %exitcond.3, label %for.latch, label %for.inner, !llvm.loop ![[LOOP_INNER:[0-9]+]]<br>
>> >> +; CHECK: br i1 %niter.ncmp.3, label %for.end.loopexit.unr-lcssa.loopexit, label %for.outer, !llvm.loop ![[LOOP_OUTER:[0-9]+]]<br>
>> >> +; CHECK: br i1 %exitcond.epil, label %for.latch.epil, label %for.inner.epil, !llvm.loop ![[LOOP_REMAINDER_INNER:[0-9]+]]<br>
>> >> +; CHECK: br i1 %exitcond.epil.1, label %for.latch.epil.1, label %for.inner.epil.1, !llvm.loop ![[LOOP_REMAINDER_INNER]]<br>
>> >> +; CHECK: br i1 %exitcond.epil.2, label %for.latch.epil.2, label %for.inner.epil.2, !llvm.loop ![[LOOP_REMAINDER_INNER]]<br>
>> >> +<br>
>> >> +; CHECK: ![[LOOP_INNER]] = distinct !{![[LOOP_INNER]], ![[FOLLOWUP_ALL:[0-9]+]], ![[FOLLOWUP_INNER:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_ALL]] = !{!"FollowupAll"}<br>
>> >> +; CHECK: ![[FOLLOWUP_INNER]] = !{!"FollowupInner"}<br>
>> >> +; CHECK: ![[LOOP_OUTER]] = distinct !{![[LOOP_OUTER]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_OUTER:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_OUTER]] = !{!"FollowupOuter"}<br>
>> >> +; CHECK: ![[LOOP_REMAINDER_INNER]] = distinct !{![[LOOP_REMAINDER_INNER]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_REMAINDER_INNER:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_REMAINDER_INNER]] = !{!"FollowupRemainderInner"}<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll (original)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopUnrollAndJam/pragma.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -316,4 +316,4 @@ for.end:<br>
>> >> !8 = distinct !{!"llvm.loop.unroll.disable"}<br>
>> >> !9 = distinct !{!9, !10}<br>
>> >> !10 = distinct !{!"llvm.loop.unroll.enable"}<br>
>> >> -!11 = distinct !{!11, !8, !6}<br>
>> >> \ No newline at end of file<br>
>> >> +!11 = distinct !{!11, !8, !6}<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll (original)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/X86/vectorization-remarks-missed.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -1,9 +1,9 @@<br>
>> >> -; RUN: opt < %s -loop-vectorize -S -pass-remarks-missed='loop-vectorize' -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s<br>
>> >> -; RUN: opt < %s -loop-vectorize -o /dev/null -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: opt < %s -loop-vectorize -transform-warning -S -pass-remarks-missed='loop-vectorize' -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -loop-vectorize -transform-warning -o /dev/null -pass-remarks-output=%t.yaml<br>
>> >> ; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >><br>
>> >> -; RUN: opt < %s -passes=loop-vectorize -S -pass-remarks-missed='loop-vectorize' -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s<br>
>> >> -; RUN: opt < %s -passes=loop-vectorize -o /dev/null -pass-remarks-output=%t.yaml<br>
>> >> +; RUN: opt < %s -passes=loop-vectorize,transform-warning -S -pass-remarks-missed='loop-vectorize' -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -passes=loop-vectorize,transform-warning -o /dev/null -pass-remarks-output=%t.yaml<br>
>> >> ; RUN: cat %t.yaml | FileCheck -check-prefix=YAML %s<br>
>> >><br>
>> >> ; C/C++ code for tests<br>
>> >> @@ -33,7 +33,7 @@<br>
>> >> ; }<br>
>> >> ; CHECK: remark: source.cpp:19:5: loop not vectorized: cannot identify array bounds<br>
>> >> ; CHECK: remark: source.cpp:19:5: loop not vectorized<br>
>> >> -; CHECK: warning: source.cpp:19:5: loop not vectorized: failed explicitly specified loop vectorization<br>
>> >> +; CHECK: warning: source.cpp:19:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >><br>
>> >> ; int foo();<br>
>> >> ; void test_multiple_failures(int *A) {<br>
>> >> @@ -94,13 +94,12 @@<br>
>> >> ; YAML-NEXT: - String: ')'<br>
>> >> ; YAML-NEXT: ...<br>
>> >> ; YAML-NEXT: --- !Failure<br>
>> >> -; YAML-NEXT: Pass: loop-vectorize<br>
>> >> +; YAML-NEXT: Pass: transform-warning<br>
>> >> ; YAML-NEXT: Name: FailedRequestedVectorization<br>
>> >> ; YAML-NEXT: DebugLoc: { File: source.cpp, Line: 19, Column: 5 }<br>
>> >> ; YAML-NEXT: Function: _Z17test_array_boundsPiS_i<br>
>> >> ; YAML-NEXT: Args:<br>
>> >> -; YAML-NEXT: - String: 'loop not vectorized: '<br>
>> >> -; YAML-NEXT: - String: failed explicitly specified loop vectorization<br>
>> >> +; YAML-NEXT: - String: 'loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering'<br>
>> >> ; YAML-NEXT: ...<br>
>> >> ; YAML-NEXT: --- !Analysis<br>
>> >> ; YAML-NEXT: Pass: loop-vectorize<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,29 @@<br>
>> >> +; RUN: opt -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the disable_nonforced loop property is honored by the<br>
>> >> +; loop vectorizer.<br>
>> >> +;<br>
>> >> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced(<br>
>> >> +; CHECK-NOT: x i32><br>
>> >> +define void @disable_nonforced(i32* nocapture %a, i32 %n) {<br>
>> >> +entry:<br>
>> >> + %cmp4 = icmp sgt i32 %n, 0<br>
>> >> + br i1 %cmp4, label %for.body, label %for.end<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = trunc i64 %indvars.iv to i32<br>
>> >> + store i32 %0, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add i64 %indvars.iv, 1<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %n<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/disable_nonforced_enable.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,29 @@<br>
>> >> +; RUN: opt -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check whether the llvm.loop.vectorize.enable loop property overrides<br>
>> >> +; llvm.loop.disable_nonforced.<br>
>> >> +;<br>
>> >> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +; CHECK-LABEL: @disable_nonforced_enable(<br>
>> >> +; CHECK: store <2 x i32><br>
>> >> +define void @disable_nonforced_enable(i32* nocapture %a, i32 %n) {<br>
>> >> +entry:<br>
>> >> + %cmp4 = icmp sgt i32 %n, 0<br>
>> >> + br i1 %cmp4, label %for.body, label %for.end<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = trunc i64 %indvars.iv to i32<br>
>> >> + store i32 %0, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add i64 %indvars.iv, 1<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %n<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = !{!0, !{!"llvm.loop.disable_nonforced"}, !{!"llvm.loop.vectorize.enable", i32 1}}<br>
>> >><br>
>> >> Added: llvm/trunk/test/Transforms/LoopVectorize/followup.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/followup.ll?rev=348944&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/followup.ll?rev=348944&view=auto</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/followup.ll (added)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/followup.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -0,0 +1,43 @@<br>
>> >> +; RUN: opt -loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S < %s | FileCheck %s<br>
>> >> +;<br>
>> >> +; Check that the followup loop attributes are applied.<br>
>> >> +;<br>
>> >> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"<br>
>> >> +<br>
>> >> +define void @followup(i32* nocapture %a, i32 %n) {<br>
>> >> +entry:<br>
>> >> + %cmp4 = icmp sgt i32 %n, 0<br>
>> >> + br i1 %cmp4, label %for.body, label %for.end<br>
>> >> +<br>
>> >> +for.body:<br>
>> >> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]<br>
>> >> + %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv<br>
>> >> + %0 = trunc i64 %indvars.iv to i32<br>
>> >> + store i32 %0, i32* %arrayidx, align 4<br>
>> >> + %indvars.iv.next = add i64 %indvars.iv, 1<br>
>> >> + %lftr.wideiv = trunc i64 %indvars.iv.next to i32<br>
>> >> + %exitcond = icmp eq i32 %lftr.wideiv, %n<br>
>> >> + br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0<br>
>> >> +<br>
>> >> +for.end:<br>
>> >> + ret void<br>
>> >> +}<br>
>> >> +<br>
>> >> +!0 = distinct !{!0, !3, !4, !5}<br>
>> >> +!3 = !{!"llvm.loop.vectorize.followup_vectorized", !{!"FollowupVectorized"}}<br>
>> >> +!4 = !{!"llvm.loop.vectorize.followup_epilogue", !{!"FollowupEpilogue"}}<br>
>> >> +!5 = !{!"llvm.loop.vectorize.followup_all", !{!"FollowupAll"}}<br>
>> >> +<br>
>> >> +<br>
>> >> +; CHECK-LABEL @followup(<br>
>> >> +<br>
>> >> +; CHECK-LABEL: vector.body:<br>
>> >> +; CHECK: br i1 %13, label %middle.block, label %vector.body, !llvm.loop ![[LOOP_VECTOR:[0-9]+]]<br>
>> >> +; CHECK-LABEL: for.body:<br>
>> >> +; CHECK: br i1 %exitcond, label %for.end.loopexit, label %for.body, !llvm.loop ![[LOOP_EPILOGUE:[0-9]+]]<br>
>> >> +<br>
>> >> +; CHECK: ![[LOOP_VECTOR]] = distinct !{![[LOOP_VECTOR]], ![[FOLLOWUP_ALL:[0-9]+]], ![[FOLLOWUP_VECTORIZED:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_ALL]] = !{!"FollowupAll"}<br>
>> >> +; CHECK: ![[FOLLOWUP_VECTORIZED:[0-9]+]] = !{!"FollowupVectorized"}<br>
>> >> +; CHECK: ![[LOOP_EPILOGUE]] = distinct !{![[LOOP_EPILOGUE]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_EPILOGUE:[0-9]+]]}<br>
>> >> +; CHECK: ![[FOLLOWUP_EPILOGUE]] = !{!"FollowupEpilogue"}<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll (original)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/no_array_bounds.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -1,8 +1,8 @@<br>
>> >> -; RUN: opt < %s -loop-vectorize -S 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -loop-vectorize -transform-warning -S 2>&1 | FileCheck %s<br>
>> >><br>
>> >> ; Verify warning is generated when vectorization/ interleaving is explicitly specified and fails to occur.<br>
>> >> -; CHECK: warning: no_array_bounds.cpp:5:5: loop not vectorized: failed explicitly specified loop vectorization<br>
>> >> -; CHECK: warning: no_array_bounds.cpp:10:5: loop not interleaved: failed explicitly specified loop interleaving<br>
>> >> +; CHECK: warning: no_array_bounds.cpp:5:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >> +; CHECK: warning: no_array_bounds.cpp:10:5: loop not interleaved: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >><br>
>> >> ; #pragma clang loop vectorize(enable)<br>
>> >> ; for (int i = 0; i < number; i++) {<br>
>> >><br>
>> >> Modified: llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll<br>
>> >> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll?rev=348944&r1=348943&r2=348944&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll?rev=348944&r1=348943&r2=348944&view=diff</a><br>
>> >> ==============================================================================<br>
>> >> --- llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll (original)<br>
>> >> +++ llvm/trunk/test/Transforms/LoopVectorize/no_switch.ll Wed Dec 12 09:32:52 2018<br>
>> >> @@ -1,16 +1,16 @@<br>
>> >> -; RUN: opt < %s -loop-vectorize -force-vector-width=4 -S 2>&1 | FileCheck %s<br>
>> >> -; RUN: opt < %s -loop-vectorize -force-vector-width=1 -S 2>&1 | FileCheck %s -check-prefix=NOANALYSIS<br>
>> >> -; RUN: opt < %s -loop-vectorize -force-vector-width=4 -pass-remarks-missed='loop-vectorize' -S 2>&1 | FileCheck %s -check-prefix=MOREINFO<br>
>> >> +; RUN: opt < %s -loop-vectorize -force-vector-width=4 -transform-warning -S 2>&1 | FileCheck %s<br>
>> >> +; RUN: opt < %s -loop-vectorize -force-vector-width=1 -transform-warning -S 2>&1 | FileCheck %s -check-prefix=NOANALYSIS<br>
>> >> +; RUN: opt < %s -loop-vectorize -force-vector-width=4 -transform-warning -pass-remarks-missed='loop-vectorize' -S 2>&1 | FileCheck %s -check-prefix=MOREINFO<br>
>> >><br>
>> >> ; CHECK: remark: source.cpp:4:5: loop not vectorized: loop contains a switch statement<br>
>> >> -; CHECK: warning: source.cpp:4:5: loop not vectorized: failed explicitly specified loop vectorization<br>
>> >> +; CHECK: warning: source.cpp:4:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >><br>
>> >> ; NOANALYSIS-NOT: remark: {{.*}}<br>
>> >> -; NOANALYSIS: warning: source.cpp:4:5: loop not interleaved: failed explicitly specified loop interleaving<br>
>> >> +; NOANALYSIS: warning: source.cpp:4:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >><br>
>> >> ; MOREINFO: remark: source.cpp:4:5: loop not vectorized: loop contains a switch statement<br>
>> >> ; MOREINFO: remark: source.cpp:4:5: loop not vectorized (Force=true, Vector Width=4)<br>
>> >> -; MOREINFO: warning: source.cpp:4:5: loop not vectorized: failed explicitly specified loop vectorization<br>
>> >> +; MOREINFO: warning: source.cpp:4:5: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering<br>
>> >><br>
>> >> ; CHECK: _Z11test_switchPii<br>
>> >> ; CHECK-NOT: x i32><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>