[CodeGen] Emit alignment "Max Skip" operand for align directives

[NickGuy]

Summary

The current AsmPrinter has support to emit the "Max Skip" operand
(the 3rd of .p2align), however has no support for it to actually be specified.
Adding MaxBytesForAlignment to MachineBasicBlock provides this capability on a
per-block basis. Leaving the value as default (0) causes no observable differences
in behaviour.

[dmgreen]

Alignments are used in a lot of places, where you wouldn't really need to specify a maximum bytes added parameter, that's mostly use in aligning code sections as far as I understand. Do we really want to add this to the base Align class? Or should it just be part of whatever controls the code alignment?

[NickGuy]

In D114590#3153774, @dmgreen wrote:

Do we really want to add this to the base Align class? Or should it just be part of whatever controls the code alignment?

As far as I can tell, AsmPrinter::emitAlignment is what controls the code alignment. Beyond that point, all the parameters are either unsigned or int64_t types. I was hesitant to add it to Align, but this was the least-invasive solution I could come up with.

[gchatelet]

Align is a very generic class used in many contexts. Enriching it with MaxBytesForAlignment will break the mental model of everyone using it.
If you need the constraint for the AsmPrinter then pass it along manually.

[SjoerdMeijer]

Do you have a follow up patch in the pipeline that uses this? Or in other words, how could this be tested?

[SjoerdMeijer]

llvm/include/llvm/CodeGen/MachineBasicBlock.h

139

Some comments/explanations what this exactly would be good, like the other members here.

Not sure what the maximum value can be, if there is any, and if we can or should use a smaller int type to reduce the size of MBBs.

[NickGuy]

I've got a follow-up patch nearly ready with tests, yep. I want to settle on an approach here before implementing the other side of it.

[SjoerdMeijer]

In D114590#3163777, @NickGuy wrote:

I've got a follow-up patch nearly ready with tests, yep. I want to settle on an approach here before implementing the other side of it.

Okay, that's fine. Perhaps upload your WIP patches too then, stack them, so that we can actually see the (full) approach?

[NickGuy]

In D114590#3163780, @SjoerdMeijer wrote:

Okay, that's fine. Perhaps upload your WIP patches too then, stack them, so that we can actually see the (full) approach?

Done, D114879 has been added as a child revision. I couldn't figure out how to mark it as a draft revision though, so I've left it with no reviewers for now.

[dmgreen]

It may be better to fold this into D114879 so that there is a single patch that can be tested properly.

llvm/include/llvm/CodeGen/AsmPrinter.h

435–436

Can you make sure you clang-format the changes.

llvm/include/llvm/CodeGen/MachineBasicBlock.h

141

1<<Alignment or 2^Alignment. It may be better to use the wording from elsewhere and say "If the alignment cannot be reached in this many bytes, no bytes are emitted."

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

2466

What about this code path?

[courbet]

llvm/include/llvm/CodeGen/MachineBasicBlock.h

143

Is there any reason for the inconsistency between unsigned in emitAlignment and uint8_t here ? I don't think this struct is particularly memory-constrained, so I would just use unsigned here.

[NickGuy]

In D114590#3166112, @dmgreen wrote:

It may be better to fold this into D114879 so that there is a single patch that can be tested properly.

I've partially folded D114879 into this, I want to keep the AArch64-specific stuff separate though.

llvm/include/llvm/CodeGen/MachineBasicBlock.h

141

Right, Alignment^2 is not the same as 2^Alignment... Fixed

143

No particular reason beyond memory saving, if that's not a concern here then unsigned makes more sense.

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

2466

I haven't seen anything take this path, but I've now added MaxBytesToEmit to this call too, following the logic that if MaxBytesToEmit is provided, it should be used.

[dmgreen]

Can you add a test to show that the alignment when emitting an object file is correct too?

I've partially folded D114879 into this, I want to keep the AArch64-specific stuff separate though.

OK. I'm not sure I understand why. It is easier to review patches that are complete and do something specific :)

llvm/include/llvm/CodeGen/MachineBasicBlock.h

143

Using unsigned sounds good.

The "Values greater than..." sentence is already covered by the "no bytes are emitted" part above, and can be removed.

[NickGuy]

Fixes the test to pass with this patch in isolation (i.e. without the child revision on top)

[dmgreen]

llvm/lib/CodeGen/MachineBlockPlacement.cpp

3416

Should the MaxBytes option be used for these alignments too, if we are adding the option?

llvm/test/CodeGen/AArch64/aarch64-p2align-max-bytes.ll

5–6

Can you add a comment explaining what this is testing. It might be a bit awkward to maintain if the codegen changes and a comment explaining what this is testing will help with that.

[dmgreen]

Thanks. Together with D114879 this LGTM.

llvm/include/llvm/CodeGen/TargetLowering.h

2957

Add a quick comment for the MaxBytesForAlignment?

[SjoerdMeijer]

llvm/include/llvm/CodeGen/MachineBasicBlock.h

533

Just adding this nit before you commit this: missing punctuation here and in most other comments.

[vitalybuka]

This patch is likely responsible for https://lab.llvm.org/buildbot/#/builders/5/builds/16829 and some other bots.

[vitalybuka]

Here is report with origins from https://lab.llvm.org/buildbot/#/builders/74/builds/8671/steps/18/logs/stdio

********************
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 
FAIL: LLVM :: Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll (65880 of 81093)
******************** TEST 'LLVM :: Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll' FAILED ********************
Script:
--
: 'RUN: at line 1';   /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build_msan_track_origins/bin/opt < /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/test/Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll -mcpu=core-avx2 -loop-vectorize -S | /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build_msan_track_origins/bin/llc -mcpu=core-avx2 | /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build_msan_track_origins/bin/FileCheck /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/test/Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll
--
Exit Code: 2
Command Output (stderr):
--
==74358==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0xbf34382 in (anonymous namespace)::MCAsmStreamer::emitValueToAlignment(unsigned int, long, unsigned int, unsigned int) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/MC/MCAsmStreamer.cpp:1425:11
    #1 0x96f5097 in emitAlignment /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp:2494:18
    #2 0x96f5097 in llvm::AsmPrinter::emitBasicBlockStart(llvm::MachineBasicBlock const&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp:3290:5
    #3 0x96c528d in llvm::AsmPrinter::emitFunctionBody() /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp:1275:5
    #4 0x745fbdb in llvm::X86AsmPrinter::runOnMachineFunction(llvm::MachineFunction&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/Target/X86/X86AsmPrinter.cpp:82:3
    #5 0xa22715a in llvm::MachineFunctionPass::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/MachineFunctionPass.cpp:72:13
    #6 0xb9c2518 in llvm::FPPassManager::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1437:27
    #7 0xb9ec856 in llvm::FPPassManager::runOnModule(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1483:16
    #8 0xb9c49c2 in runOnModule /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1552:27
    #9 0xb9c49c2 in llvm::legacy::PassManagerImpl::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:542:44
    #10 0xb9ed62b in llvm::legacy::PassManager::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1679:14
    #11 0x2d964b9 in compileModule(char**, llvm::LLVMContext&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:719:8
    #12 0x2d8cbd8 in main /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:417:22
    #13 0x7ff258bad09a in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2409a) (BuildId: 18b9a9a8c523e5cfe5b5d946d605d09242f09798)
    #14 0x2cebea9 in _start (/b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build_msan_track_origins/bin/llc+0x2cebea9)

  Uninitialized value was stored to memory at
    #0 0xa06bbc3 in alignBlocks /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/MachineBlockPlacement.cpp
    #1 0xa06bbc3 in (anonymous namespace)::MachineBlockPlacement::runOnMachineFunction(llvm::MachineFunction&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/MachineBlockPlacement.cpp:3434:3
    #2 0xa22715a in llvm::MachineFunctionPass::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/MachineFunctionPass.cpp:72:13
    #3 0xb9c2518 in llvm::FPPassManager::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1437:27
    #4 0xb9ec856 in llvm::FPPassManager::runOnModule(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1483:16
    #5 0xb9c49c2 in runOnModule /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1552:27
    #6 0xb9c49c2 in llvm::legacy::PassManagerImpl::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:542:44
    #7 0xb9ed62b in llvm::legacy::PassManager::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1679:14
    #8 0x2d964b9 in compileModule(char**, llvm::LLVMContext&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:719:8
    #9 0x2d8cbd8 in main /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:417:22
    #10 0x7ff258bad09a in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2409a) (BuildId: 18b9a9a8c523e5cfe5b5d946d605d09242f09798)

  Uninitialized value was created by a heap allocation
    #0 0x2d7d613 in operator new(unsigned long) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/compiler-rt/lib/msan/msan_new_delete.cpp:45:35
    #1 0x7e2ba8d in make_unique<llvm::X86Subtarget, const llvm::Triple &, llvm::StringRef &, llvm::StringRef &, llvm::StringRef &, const llvm::X86TargetMachine &, llvm::MaybeAlign, unsigned int &, unsigned int &> /b/sanitizer-x86_64-linux-bootstrap-msan/build/libcxx_build_msan_track_origins/include/c++/v1/__memory/unique_ptr.h:725:28
    #2 0x7e2ba8d in llvm::X86TargetMachine::getSubtargetImpl(llvm::Function const&) const /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/Target/X86/X86TargetMachine.cpp:315:9
    #3 0x9a32a92 in (anonymous namespace)::AtomicExpand::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/CodeGen/AtomicExpandPass.cpp:175:11
    #4 0xb9c2518 in llvm::FPPassManager::runOnFunction(llvm::Function&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1437:27
    #5 0xb9ec856 in llvm::FPPassManager::runOnModule(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1483:16
    #6 0xb9c49c2 in runOnModule /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1552:27
    #7 0xb9c49c2 in llvm::legacy::PassManagerImpl::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:542:44
    #8 0xb9ed62b in llvm::legacy::PassManager::run(llvm::Module&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/IR/LegacyPassManager.cpp:1679:14
    #9 0x2d964b9 in compileModule(char**, llvm::LLVMContext&) /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:719:8
    #10 0x2d8cbd8 in main /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/tools/llc/llc.cpp:417:22
    #11 0x7ff258bad09a in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2409a) (BuildId: 18b9a9a8c523e5cfe5b5d946d605d09242f09798)
SUMMARY: MemorySanitizer: use-of-uninitialized-value /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/lib/MC/MCAsmStreamer.cpp:1425:11 in (anonymous namespace)::MCAsmStreamer::emitValueToAlignment(unsigned int, long, unsigned int, unsigned int)
Exiting
FileCheck error: '<stdin>' is empty.
FileCheck command line:  /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build_msan_track_origins/bin/FileCheck /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/test/Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll

[dmgreen]

Thanks for the report. Sounds like the sanitizer is doing its job admirably.

I'll try and make a fix. I think that MaxBytesForAlignment in TLI is the thing that isn't being initialized, but need to test it.

llvm/include/llvm/CodeGen/AsmPrinter.h

public:
   /// This method emits `llvm.global_ctors` or `llvm.global_dtors` list.
   virtual void emitXXStructorList(const DataLayout &DL, const Constant *List,
                                   bool IsCtor);
 
   /// Emit an alignment directive to the specified power of two boundary. If a
   /// global value is specified, and if that global has an explicit alignment
   /// requested, it will override the alignment request if required for
   /// correctness.
-  void emitAlignment(Align Alignment, const GlobalObject *GV = nullptr) const;
+  void emitAlignment(Align Alignment, const GlobalObject *GV = nullptr,
+                     unsigned MaxBytesToEmit = 0) const;
 
   /// Lower the specified LLVM Constant to an MCExpr.
   virtual const MCExpr *lowerConstant(const Constant *CV);
 
   /// Print a general LLVM constant to the .s file.
   void emitGlobalConstant(const DataLayout &DL, const Constant *CV);
 
   /// Unnamed constant global variables solely contaning a pointer to

llvm/include/llvm/CodeGen/MachineBasicBlock.h

private:
 
   /// Keep track of the physical registers that are livein of the basicblock.
   using LiveInVector = std::vector<RegisterMaskPair>;
   LiveInVector LiveIns;
 
   /// Alignment of the basic block. One if the basic block does not need to be
   /// aligned.
   Align Alignment;
+  /// Maximum amount of bytes that can be added to align the basic block. If the
+  /// alignment cannot be reached in this many bytes, no bytes are emitted.
+  /// Zero to represent no maximum.
+  uint8_t MaxBytesForAlignment = 0;
 
   /// Indicate that this basic block is entered via an exception handler.
   bool IsEHPad = false;
 
   /// Indicate that this basic block is potentially the target of an indirect
   /// branch.
   bool AddressTaken = false;
 
   /// Indicate that this basic block needs its symbol be emitted regardless of
#endif
   const uint32_t *getEndClobberMask(const TargetRegisterInfo *TRI) const;
 
   /// Return alignment of the basic block.
   Align getAlignment() const { return Alignment; }
 
   /// Set alignment of the basic block.
   void setAlignment(Align A) { Alignment = A; }
 
+  /// Return the maximum amount of padding allowed for aligning the basic block
+  unsigned getMaxBytesForAlignment() const { return MaxBytesForAlignment; }
+
+  /// Set the maximum amount of padding allowed for aligning the basic block
+  void setMaxBytesForAlignment(unsigned MaxBytes) {
+    MaxBytesForAlignment = MaxBytes;
+  }
+
   /// Returns true if the block is a landing pad. That is this basic block is
   /// entered via an exception handler.
   bool isEHPad() const { return IsEHPad; }
 
   /// Indicates the block is a landing pad.  That is this basic block is entered
   /// via an exception handler.
   void setIsEHPad(bool V = true) { IsEHPad = V; }
 

llvm/include/llvm/CodeGen/TargetLowering.h

#include "llvm/IR/ConstrainedOps.def"
   Align getMinFunctionAlignment() const { return MinFunctionAlignment; }
 
   /// Return the preferred function alignment.
   Align getPrefFunctionAlignment() const { return PrefFunctionAlignment; }
 
   /// Return the preferred loop alignment.
   virtual Align getPrefLoopAlignment(MachineLoop *ML = nullptr) const;
 
+  /// Return the maximum amount of bytes allowed to be emitted when padding for
+  /// alignment
+  virtual unsigned
+  getMaxPermittedBytesForAlignment(MachineBasicBlock *MBB) const;
+
   /// Should loops be aligned even when the function is marked OptSize (but not
   /// MinSize).
-  virtual bool alignLoopsWithOptSize() const {
-    return false;
-  }
+  virtual bool alignLoopsWithOptSize() const { return false; }
 
   /// If the target has a standard location for the stack protector guard,
   /// returns the address of that location. Otherwise, returns nullptr.
   /// DEPRECATED: please override useLoadStackGuardNode and customize
   ///             LOAD_STACK_GUARD, or customize \@llvm.stackguard().
   virtual Value *getIRStackGuard(IRBuilderBase &IRB) const;
 
   /// Inserts necessary declarations for SSP (stack protection) purpose.
private:
   Align MinFunctionAlignment;
 
   /// The preferred function alignment (used when alignment unspecified and
   /// optimizing for speed).
   Align PrefFunctionAlignment;
 
   /// The preferred loop alignment (in log2 bot in bytes).
   Align PrefLoopAlignment;
 
   /// Size in bits of the maximum atomics size the backend supports.
   /// Accesses larger than this will be expanded by AtomicExpandPass.
   unsigned MaxAtomicSizeInBitsSupported;
 
   /// Size in bits of the minimum cmpxchg or ll/sc operation the
   /// backend supports.
   unsigned MinCmpXchgSizeInBits;
 

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

 }
 
 //===----------------------------------------------------------------------===//
 
 // EmitAlignment - Emit an alignment directive to the specified power of
 // two boundary.  If a global value is specified, and if that global has
 // an explicit alignment requested, it will override the alignment request
 // if required for correctness.
-void AsmPrinter::emitAlignment(Align Alignment, const GlobalObject *GV) const {
+void AsmPrinter::emitAlignment(Align Alignment, const GlobalObject *GV,
+                               unsigned MaxBytesToEmit) const {
   if (GV)
     Alignment = getGVAlignment(GV, GV->getParent()->getDataLayout(), Alignment);
 
   if (Alignment == Align(1))
     return; // 1-byte aligned: no need to emit alignment.
 
   if (getCurrentSection()->getKind().isText()) {
     const MCSubtargetInfo *STI = nullptr;
     if (this->MF)
       STI = &getSubtargetInfo();
     else
       STI = TM.getMCSubtargetInfo();
-    OutStreamer->emitCodeAlignment(Alignment.value(), STI);
+    OutStreamer->emitCodeAlignment(Alignment.value(), STI, MaxBytesToEmit);
   } else
-    OutStreamer->emitValueToAlignment(Alignment.value());
+    OutStreamer->emitValueToAlignment(Alignment.value(), 0, 1, MaxBytesToEmit);
 }
 
 //===----------------------------------------------------------------------===//
 // Constant emission.
 //===----------------------------------------------------------------------===//
 
 const MCExpr *AsmPrinter::lowerConstant(const Constant *CV) {
   MCContext &Ctx = OutContext;
for (const HandlerInfo &HI : Handlers) {
       HI.Handler->endFunclet();
       HI.Handler->beginFunclet(MBB);
     }
   }
 
   // Emit an alignment directive for this block, if needed.
   const Align Alignment = MBB.getAlignment();
   if (Alignment != Align(1))
-    emitAlignment(Alignment);
+    emitAlignment(Alignment, nullptr, MBB.getMaxBytesForAlignment());
 
   // Switch to a new section if this basic block must begin a section. The
   // entry block is always placed in the function section and is handled
   // separately.
   if (MBB.isBeginSection() && !MBB.isEntryBlock()) {
     OutStreamer->SwitchSection(
         getObjFileLowering().getSectionForMachineBasicBlock(MF->getFunction(),
                                                             MBB, TM));

llvm/lib/CodeGen/MachineBlockPlacement.cpp

 
 static cl::opt<unsigned> AlignAllNonFallThruBlocks(
     "align-all-nofallthru-blocks",
     cl::desc("Force the alignment of all blocks that have no fall-through "
              "predecessors (i.e. don't add nops that are executed). In log2 "
              "format (e.g 4 means align on 16B boundaries)."),
     cl::init(0), cl::Hidden);
 
+static cl::opt<unsigned> MaxBytesForAlignmentOverride(
+    "max-bytes-for-alignment",
+    cl::desc("Forces the maximum bytes allowed to be emitted when padding for "
+             "alignment"),
+    cl::init(0), cl::Hidden);
+
 // FIXME: Find a good default for this flag and remove the flag.
 static cl::opt<unsigned> ExitBlockBias(
     "block-placement-exit-block-bias",
     cl::desc("Block frequency percentage a loop exit block needs "
              "over the original exit to be considered the new exit."),
     cl::init(0), cl::Hidden);
 
 // Definition:
if (llvm::shouldOptimizeForSize(ChainBB, PSI, MBFI.get()) &&
         !TLI->alignLoopsWithOptSize())
       continue;
 
     // Check for the existence of a non-layout predecessor which would benefit
     // from aligning this block.
     MachineBasicBlock *LayoutPred =
         &*std::prev(MachineFunction::iterator(ChainBB));
 
+    auto DetermineMaxAlignmentPadding = [&]() {
+      // Set the maximum bytes allowed to be emitted for alignment.
+      unsigned MaxBytes;
+      if (MaxBytesForAlignmentOverride.getNumOccurrences() > 0)
+        MaxBytes = MaxBytesForAlignmentOverride;
+      else
+        MaxBytes = TLI->getMaxPermittedBytesForAlignment(ChainBB);
+      ChainBB->setMaxBytesForAlignment(MaxBytes);
+    };
+
     // Force alignment if all the predecessors are jumps. We already checked
     // that the block isn't cold above.
     if (!LayoutPred->isSuccessor(ChainBB)) {
       ChainBB->setAlignment(Align);
+      DetermineMaxAlignmentPadding();
       continue;
     }
 
     // Align this block if the layout predecessor's edge into this block is
     // cold relative to the block. When this is true, other predecessors make up
     // all of the hot entries into the block and thus alignment is likely to be
     // important.
     BranchProbability LayoutProb =
         MBPI->getEdgeProbability(LayoutPred, ChainBB);
     BlockFrequency LayoutEdgeFreq = MBFI->getBlockFreq(LayoutPred) * LayoutProb;
-    if (LayoutEdgeFreq <= (Freq * ColdProb))
+    if (LayoutEdgeFreq <= (Freq * ColdProb)) {
       ChainBB->setAlignment(Align);
+      DetermineMaxAlignmentPadding();
+    }
   }
 }
 
 /// Tail duplicate \p BB into (some) predecessors if profitable, repeating if
 /// it was duplicated into its chain predecessor and removed.
 /// \p BB    - Basic block that may be duplicated.
 ///
 /// \p LPred - Chosen layout predecessor of \p BB.
bool MachineBlockPlacement::runOnMachineFunction(MachineFunction &MF) {
 
   optimizeBranches();
   alignBlocks();
 
   BlockToChain.clear();
   ComputedEdges.clear();
   ChainAllocator.DestroyAll();
 
   if (AlignAllBlock)
     // Align all of the blocks in the function to a specific alignment.
     for (MachineBasicBlock &MBB : MF)
       MBB.setAlignment(Align(1ULL << AlignAllBlock));
   else if (AlignAllNonFallThruBlocks) {
     // Align all of the blocks that have no fall-through predecessors to a
     // specific alignment.
     for (auto MBI = std::next(MF.begin()), MBE = MF.end(); MBI != MBE; ++MBI) {
       auto LayoutPred = std::prev(MBI);

llvm/lib/CodeGen/TargetLoweringBase.cpp

 }
 
 Align TargetLoweringBase::getPrefLoopAlignment(MachineLoop *ML) const {
   if (TM.Options.LoopAlignment)
     return Align(TM.Options.LoopAlignment);
   return PrefLoopAlignment;
 }
 
+unsigned TargetLoweringBase::getMaxPermittedBytesForAlignment(
+    MachineBasicBlock *MBB) const {
+  return 0;
+}
+
 //===----------------------------------------------------------------------===//
 //  Reciprocal Estimates
 //===----------------------------------------------------------------------===//
 
 /// Get the reciprocal estimate attribute string for a function that will
 /// override the target defaults.
 static StringRef getRecipEstimateForFunc(MachineFunction &MF) {
   const Function &F = MF.getFunction();

llvm/test/CodeGen/AArch64/aarch64-p2align-max-bytes.ll

+; RUN: llc -mtriple=aarch64-none-linux-gnu -max-bytes-for-alignment=8 --align-loops=32 < %s -o -| FileCheck %s --check-prefixes=CHECK,CHECK-EXPLICIT
+; RUN: llc -mtriple=aarch64-none-linux-gnu --align-loops=32 < %s -o -| FileCheck %s --check-prefixes=CHECK,CHECK-IMPLICIT
+; RUN: llc -mtriple=aarch64-none-linux-gnu --align-loops=32 < %s -o - --filetype=obj | llvm-objdump --arch=aarch64  -d -| FileCheck %s --check-prefixes=CHECK-OBJ,CHECK-OBJ-EXPLICIT
+; RUN: llc -mtriple=aarch64-none-linux-gnu -max-bytes-for-alignment=8 --align-loops=32 < %s -o - --filetype=obj | llvm-objdump --arch=aarch64  -d -| FileCheck %s --check-prefixes=CHECK-OBJ,CHECK-OBJ-IMPLICIT
+
+define i32 @a(i32 %x, i32* nocapture readonly %y, i32* nocapture readonly %z) {
+; CHECK-LABEL: a:
+; CHECK-EXPLICIT:    .p2align 5, 0x0, 8
+; CHECK-IMPLICIT:    .p2align 5
+; CHECK-NEXT:  .LBB0_5: // %vector.body
+; CHECK-EXPLICIT:    .p2align 5, 0x0, 8
+; CHECK-IMPLICIT:    .p2align 5
+; CHECK-NEXT:  .LBB0_8: // %for.body
+; CHECK-OBJ;Disassembly of section .text:
+; CHECK-OBJ:               88: 2a 00 0a 8b   add
+; CHECK-OBJ-EXPLICIT-NEXT: 8c: 1f 20 03 d5   nop
+; CHECK-OBJ-EXPLICIT-NEXT: 90: 1f 20 03 d5   nop
+; CHECK-OBJ-EXPLICIT-NEXT: 94: 1f 20 03 d5   nop
+; CHECK-OBJ-EXPLICIT-NEXT: 98: 1f 20 03 d5   nop
+; CHECK-OBJ-EXPLICIT-NEXT: 9c: 1f 20 03 d5   nop
+; CHECK-OBJ-EXPLICIT-NEXT: a0: 4b 45 40 b8   ldr
+; CHECK-OBJ-IMPLICIT-NEXT: 8c: 4b 45 40 b8   ldr
+entry:
+  %cmp10 = icmp sgt i32 %x, 0
+  br i1 %cmp10, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %x to i64
+  %min.iters.check = icmp ult i32 %x, 8
+  br i1 %min.iters.check, label %for.body.preheader17, label %vector.ph
+
+vector.ph:                                        ; preds = %for.body.preheader
+  %n.vec = and i64 %wide.trip.count, 4294967288
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body, %vector.ph
+  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+  %vec.phi = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ %10, %vector.body ]
+  %vec.phi13 = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ %11, %vector.body ]
+  %0 = getelementptr inbounds i32, i32* %y, i64 %index
+  %1 = bitcast i32* %0 to <4 x i32>*
+  %wide.load = load <4 x i32>, <4 x i32>* %1, align 4
+  %2 = getelementptr inbounds i32, i32* %0, i64 4
+  %3 = bitcast i32* %2 to <4 x i32>*
+  %wide.load14 = load <4 x i32>, <4 x i32>* %3, align 4
+  %4 = getelementptr inbounds i32, i32* %z, i64 %index
+  %5 = bitcast i32* %4 to <4 x i32>*
+  %wide.load15 = load <4 x i32>, <4 x i32>* %5, align 4
+  %6 = getelementptr inbounds i32, i32* %4, i64 4
+  %7 = bitcast i32* %6 to <4 x i32>*
+  %wide.load16 = load <4 x i32>, <4 x i32>* %7, align 4
+  %8 = add <4 x i32> %wide.load, %vec.phi
+  %9 = add <4 x i32> %wide.load14, %vec.phi13
+  %10 = add <4 x i32> %8, %wide.load15
+  %11 = add <4 x i32> %9, %wide.load16
+  %index.next = add nuw i64 %index, 8
+  %12 = icmp eq i64 %index.next, %n.vec
+  br i1 %12, label %middle.block, label %vector.body
+
+middle.block:                                     ; preds = %vector.body
+  %bin.rdx = add <4 x i32> %11, %10
+  %13 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %bin.rdx)
+  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
+  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader17
+
+for.body.preheader17:                             ; preds = %for.body.preheader, %middle.block
+  %indvars.iv.ph = phi i64 [ 0, %for.body.preheader ], [ %n.vec, %middle.block ]
+  %b.011.ph = phi i32 [ 0, %for.body.preheader ], [ %13, %middle.block ]
+  br label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %entry
+  %b.0.lcssa = phi i32 [ 0, %entry ], [ %13, %middle.block ], [ %add3, %for.body ]
+  ret i32 %b.0.lcssa
+
+for.body:                                         ; preds = %for.body.preheader17, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader17 ]
+  %b.011 = phi i32 [ %add3, %for.body ], [ %b.011.ph, %for.body.preheader17 ]
+  %arrayidx = getelementptr inbounds i32, i32* %y, i64 %indvars.iv
+  %14 = load i32, i32* %arrayidx, align 4
+  %arrayidx2 = getelementptr inbounds i32, i32* %z, i64 %indvars.iv
+  %15 = load i32, i32* %arrayidx2, align 4
+  %add = add i32 %14, %b.011
+  %add3 = add i32 %add, %15
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)