[PowerPC][Future] Prefixed Instructions 64 Byte Boundary Support

[stefanp]

Summary

A known limitation for Future CPU is that the new prefixed instructions may not cross 64 Byte boundaries.
All instructions are already 4 byte aligned so the only situation where this can occur is when the prefix is in one 64 byte block and the instruction that is prefixed is at the top of the next 64 byte block. To fix this case PPCELFStreamer was added to intercept EmitInstruction. When a prefixed instruction is emitted we try to align it to 64 Bytes by adding a maximum of 4 bytes. If the prefixed instruction crosses the 64 Byte boundary then the alignment would trigger and a 4 byte nop would be added to push the instruction into the next 64 byte block.

[nemanjai]

Other than a few minor nits, LGTM.

llvm/lib/Target/PowerPC/MCTargetDesc/PPCELFStreamer.cpp

14

Maybe this should
s/if possible/if required
I suppose you are trying to communicate that we will only align an instruction on a 64B boundary if it is crossing the boundary.

llvm/lib/Target/PowerPC/MCTargetDesc/PPCELFStreamer.h

29

Perhaps a comment about why we need this:

// We need to keep track of the last label we emitted (only one) because
// depending on whether the label is on the same line as an aligned
// instruction or not, the label may refer to the instruction or the nop.

llvm/test/MC/PowerPC/ppc64-prefix-align.s

58

Would it be possible to add branches to LAB1 and LAB2 to show that one of them will branch to the nop and the other one won't?

[stefanp]

Added full context.

[stefanp]

Address review comments.

[Yi-Hong.Lyu]

llvm/lib/Target/PowerPC/MCTargetDesc/PPCELFStreamer.cpp

24

PPCMCCodeEmitter.h should be put later than PPCInstrInfo.h

46

You can just initialized LastLabel in member initializer lists so the constructor could be empty.

107

Remove parameter bool RelaxAll since it is unused.

llvm/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp

17

"PPCELFStreamer.h" should be put before "PPCTargetStreamer.h".

112

Argument RelaxAll is not necessary since you no longer use it.

llvm/lib/Target/PowerPC/PPCInstrInfo.h

70

Why do you want to shift left NewDef_Shift+3 instead of NewDef_Shift+2?

[stefanp]

Address review comments.

llvm/lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt

   PPCMachObjectWriter.cpp
   PPCELFObjectWriter.cpp
   PPCXCOFFObjectWriter.cpp
+  PPCELFStreamer.cpp
   )

llvm/lib/Target/PowerPC/MCTargetDesc/PPCELFStreamer.h

+//===- PPCELFStreamer.h - ELF Object Output --------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This is a custom MCELFStreamer for PowerPC.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_PPC_MCELFSTREAMER_PPCELFSTREAMER_H
+#define LLVM_LIB_TARGET_PPC_MCELFSTREAMER_PPCELFSTREAMER_H
+
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/MC/MCELFStreamer.h"
+#include <memory>
+
+namespace llvm {
+
+class MCAsmBackend;
+class MCCodeEmitter;
+class MCContext;
+class MCSubtargetInfo;
+
+class PPCELFStreamer : public MCELFStreamer {
+  MCSymbol *LastLabel;
+  SMLoc LastLabelLoc;
+
+public:
+  PPCELFStreamer(MCContext &Context, std::unique_ptr<MCAsmBackend> MAB,
+                 std::unique_ptr<MCObjectWriter> OW,
+                 std::unique_ptr<MCCodeEmitter> Emitter);
+
+  void EmitInstruction(const MCInst &Inst, const MCSubtargetInfo &STI) override;
+
+  void EmitLabel(MCSymbol *Symbol, SMLoc Loc = SMLoc()) override;
+};
+
+MCELFStreamer *createPPCELFStreamer(MCContext &Context,
+                                    std::unique_ptr<MCAsmBackend> MAB,
+                                    std::unique_ptr<MCObjectWriter> OW,
+                                    std::unique_ptr<MCCodeEmitter> Emitter,
+                                    bool RelaxAll);
+} // end namespace llvm
+
+#endif // LLVM_LIB_TARGET_PPC_MCELFSTREAMER_PPCELFSTREAMER_H

llvm/lib/Target/PowerPC/MCTargetDesc/PPCELFStreamer.cpp

+//===-------- PPCELFStreamer.cpp - ELF Object Output ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This is a custom MCELFStreamer for PowerPC.
+//
+// The purpose of the custom ELF streamer is to allow us to intercept
+// instructions as they are being emitted and align all 8 byte instructions
+// to a 64 byte boundary if possible (by adding a 4 byte nop). This is important
+// because 8 byte instructions are not allowed to cross 64 byte boundaries
+// and by aliging anything that is within 4 bytes of the boundary we can
+// guarantee that the 8 byte instructions do not cross that boundary.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "PPCELFStreamer.h"
+#include "PPCMCCodeEmitter.h"
+#include "PPCInstrInfo.h"
+#include "llvm/BinaryFormat/ELF.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrDesc.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCSymbolELF.h"
+#include "llvm/Support/Casting.h"
+#include "llvm/Support/SourceMgr.h"
+
+using namespace llvm;
+
+PPCELFStreamer::PPCELFStreamer(MCContext &Context,
+                               std::unique_ptr<MCAsmBackend> MAB,
+                               std::unique_ptr<MCObjectWriter> OW,
+                               std::unique_ptr<MCCodeEmitter> Emitter)
+    : MCELFStreamer(Context, std::move(MAB), std::move(OW),
+                    std::move(Emitter)) {
+  LastLabel = NULL;
+}
+
+void PPCELFStreamer::EmitInstruction(const MCInst &Inst,
+                                     const MCSubtargetInfo &STI) {
+  PPCMCCodeEmitter *Emitter =
+      static_cast<PPCMCCodeEmitter*>(getAssembler().getEmitterPtr());
+
+  // Special handling is only for prefixed instructions.
+  if (!Emitter->isPrefixedInstruction(Inst)) {
+    MCELFStreamer::EmitInstruction(Inst, STI);
+    return;
+  }
+
+  // Prefixed instructions must not cross a 64-byte boundary (i.e. prefix is
+  // before the boundary and the remaining 4-bytes are after the boundary). In
+  // order to achieve this, a nop is added prior to any such boundary-crossing
+  // prefixed instruction. Align to 64 bytes if possible but add a maximum of 4
+  // bytes when trying to do that. If alignment requires adding more than 4
+  // bytes then the instruction won't be aligned. When emitting a code alignment
+  // a new fragment is created for this alignment. This fragment will contain
+  // all of the nops required as part of the alignment operation. In the cases
+  // when no nops are added then The fragment is still created but it remains
+  // empty.
+  EmitCodeAlignment(64, 4);
+
+  // Emit the instruction.
+  // Since the previous emit created a new fragment then adding this instruction
+  // also forces the addition of a new fragment. Inst is now the first
+  // instruction in that new fragment.
+  MCELFStreamer::EmitInstruction(Inst, STI);
+
+  // The above instruction is forced to start a new fragment because it
+  // comes after a code alignment fragment. Get that new fragment.
+  MCFragment *InstructionFragment = getCurrentFragment();
+  SMLoc InstLoc = Inst.getLoc();
+  // Check if there was a last label emitted.
+  if (LastLabel && !LastLabel->isUnset() && LastLabelLoc.isValid() &&
+      InstLoc.isValid()) {
+    const SourceMgr *SourceManager = getContext().getSourceManager();
+    unsigned InstLine = SourceManager->FindLineNumber(InstLoc);
+    unsigned LabelLine = SourceManager->FindLineNumber(LastLabelLoc);
+    // If the Label and the Instruction are on the same line then move the
+    // label to the top of the fragment containing the aligned instruction that
+    // was just added.
+    if (InstLine == LabelLine) {
+      AssignFragment(LastLabel, InstructionFragment);
+      LastLabel->setOffset(0);
+    }
+  }
+}
+
+void PPCELFStreamer::EmitLabel(MCSymbol *Symbol, SMLoc Loc) {
+  LastLabel = Symbol;
+  LastLabelLoc = Loc;
+  MCELFStreamer::EmitLabel(Symbol);
+}
+
+MCELFStreamer *llvm::createPPCELFStreamer(
+    MCContext &Context, std::unique_ptr<MCAsmBackend> MAB,
+    std::unique_ptr<MCObjectWriter> OW, std::unique_ptr<MCCodeEmitter> Emitter,
+    bool RelaxAll) {
+  return new PPCELFStreamer(Context, std::move(MAB), std::move(OW),
+                            std::move(Emitter));
+}

llvm/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.h

   // Get the number of bytes used to encode the given MCInst.
   unsigned getInstSizeInBytes(const MCInst &MI) const;
 
+  // Is this instruction a prefixed instruction.
+  bool isPrefixedInstruction(const MCInst &MI) const;
+
 private:
   FeatureBitset computeAvailableFeatures(const FeatureBitset &FB) const;
   void

llvm/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp

   return Desc.getSize();
 }
 
+bool PPCMCCodeEmitter::isPrefixedInstruction(const MCInst &MI) const {
+  unsigned Opcode = MI.getOpcode();
+  const PPCInstrInfo *InstrInfo = static_cast<const PPCInstrInfo*>(&MCII);
+  return InstrInfo->isPrefixed(Opcode);
+}
+
 #define ENABLE_INSTR_PREDICATE_VERIFIER
 #include "PPCGenMCCodeEmitter.inc"

llvm/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp

 #include "MCTargetDesc/PPCInstPrinter.h"
 #include "MCTargetDesc/PPCMCAsmInfo.h"
 #include "PPCTargetStreamer.h"
+#include "PPCELFStreamer.h"
 #include "TargetInfo/PowerPCTargetInfo.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/BinaryFormat/ELF.h"
 #include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCDwarf.h"
 #include "llvm/MC/MCELFStreamer.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCObjectWriter.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"

   return MAI;
 }
 
+static MCStreamer *createPPCMCStreamer(const Triple &T, MCContext &Context,
+                                       std::unique_ptr<MCAsmBackend> &&MAB,
+                                       std::unique_ptr<MCObjectWriter> &&OW,
+                                       std::unique_ptr<MCCodeEmitter> &&Emitter,
+                                       bool RelaxAll) {
+  return createPPCELFStreamer(Context, std::move(MAB), std::move(OW),
+                              std::move(Emitter), RelaxAll);
+}
+
 namespace {
 
 class PPCTargetAsmStreamer : public PPCTargetStreamer {

     // Register the asm backend.
     TargetRegistry::RegisterMCAsmBackend(*T, createPPCAsmBackend);
 
+    // Register the elf streamer.
+    TargetRegistry::RegisterELFStreamer(*T, createPPCMCStreamer);
+
     // Register the object target streamer.
     TargetRegistry::RegisterObjectTargetStreamer(*T,
                                                  createObjectTargetStreamer);

llvm/lib/Target/PowerPC/PPCInstrInfo.h

   NewDef_Shift = 6,
 
   /// This instruction is an X-Form memory operation.
-  XFormMemOp = 0x1 << (NewDef_Shift+1)
+  XFormMemOp = 0x1 << (NewDef_Shift+1),
+  /// This is a prefixed instruction
+  Prefixed = 0x1 << (NewDef_Shift+3)
 };
 } // end namespace PPCII
 

   bool isXFormMemOp(unsigned Opcode) const {
     return get(Opcode).TSFlags & PPCII::XFormMemOp;
   }
+  bool isPrefixed(unsigned Opcode) const {
+    return get(Opcode).TSFlags & PPCII::Prefixed;
+  }
+
   static bool isSameClassPhysRegCopy(unsigned Opcode) {
     unsigned CopyOpcodes[] =
       { PPC::OR, PPC::OR8, PPC::FMR, PPC::VOR, PPC::XXLOR, PPC::XXLORf,

llvm/test/MC/PowerPC/ppc64-prefix-align-labels.s

+# RUN: llvm-mc -triple powerpc64-unknown-linux-gnu --filetype=obj -o - %s | \
+# RUN:   llvm-objdump -D  -r - | FileCheck -check-prefix=CHECK-BE %s
+# RUN: llvm-mc -triple powerpc64le-unknown-linux-gnu --filetype=obj -o - %s | \
+# RUN:   llvm-objdump -D  -r - | FileCheck -check-prefix=CHECK-LE %s
+
+# The purpose of this test is to check that when an alignment nop is added
+# it is added correctly with resepect to the labels in the .s file.
+# The test contains 3 labels at the end (1:, 2:, 3:). The label 2: is on the
+# same line as an unaligned 8 byte instruction. The desired behaviour is to have
+# the alignment nop inserted after the 1: label but before the 2: label. The
+# branch to 1: should jump to 3c: and the branch to 2: should jump to 40:.
+
+  .text
+_start:
+  b 1f;
+  b 2f;
+  b 3f;
+# CHECK-BE:      0:       48 00 00 3c
+# CHECK-BE-NEXT: 4:       48 00 00 3c
+# CHECK-BE-NEXT: 8:       48 00 00 40
+# CHECK-LE:      0:       3c 00 00 48
+# CHECK-LE-NEXT: 4:       3c 00 00 48
+# CHECK-LE-NEXT: 8:       40 00 00 48
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+  trap
+1:
+2:  paddi 1, 2, 8589934576, 0  # 8 Byte Instruction
+3:
+  blr
+# CHECK-BE:      3c:       60 00 00 00     nop
+# CHECK-BE-NEXT: 40:       06 01 ff ff
+# CHECK-BE-NEXT: 44:       38 22 ff f0
+# CHECK-BE-NEXT: 48:       4e 80 00 20
+# CHECK-LE:      3c:       00 00 00 60     nop
+# CHECK-LE-NEXT: 40:       ff ff 01 06
+# CHECK-LE-NEXT: 44:       f0 ff 22 38
+# CHECK-LE-NEXT: 48:       20 00 80 4e
+

llvm/test/MC/PowerPC/ppc64-prefix-align.s

+# RUN: llvm-mc -triple powerpc64-unknown-linux-gnu --filetype=obj -o - %s | \
+# RUN:   llvm-objdump -D  -r - | FileCheck -check-prefix=CHECK-BE %s
+# RUN: llvm-mc -triple powerpc64le-unknown-linux-gnu --filetype=obj -o - %s | \
+# RUN:   llvm-objdump -D  -r - | FileCheck -check-prefix=CHECK-LE %s
+
+# The purpose of this test is to make sure that 8 byte instructions do not
+# cross 64 byte boundaries. If an 8 byte instruction is about to cross such
+# a boundary then a nop should be added so that the 8 byte instruction starts
+# 4 bytes later and does not cross the boundary.
+# This instruction is 8 bytes: paddi 1, 2, 8589934576, 0
+# This instruction is 4 bytes: addi 2, 3, 15
+
+paddi 1, 2, 8589934576, 0 # 8
+paddi 1, 2, 8589934576, 0 # 16
+paddi 1, 2, 8589934576, 0 # 24
+paddi 1, 2, 8589934576, 0 # 32
+paddi 1, 2, 8589934576, 0 # 40
+paddi 1, 2, 8589934576, 0 # 48
+paddi 1, 2, 8589934576, 0 # 56
+addi 2, 3, 15             # 60
+# Below the lines 40: and 44: contain the 8 byte instruction.
+# We check to make sure that the nop is added at 3c: so that the 8 byte
+# instruction can start at 40: which is 64 bytes aligned.
+# CHECK-BE:      38:  38 43 00 0f
+# CHECK-BE-NEXT: 3c:  60 00 00 00   nop
+# CHECK-BE-NEXT: 40:  06 01 ff ff
+# CHECK-BE-NEXT: 44:  38 22 ff f0
+# CHECK-LE:      38:  0f 00 43 38
+# CHECK-LE-NEXT: 3c:  00 00 00 60   nop
+# CHECK-LE-NEXT: 40:  ff ff 01 06
+# CHECK-LE-NEXT: 44:  f0 ff 22 38
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0 # 64
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+addi 2, 3, 15             # 60
+# CHECK-BE:      b8:  38 43 00 0f
+# CHECK-BE-NEXT: bc:  60 00 00 00   nop
+# CHECK-BE:      LAB1:
+# CHECK-BE-NEXT: c0:  06 01 ff ff
+# CHECK-BE-NEXT: c4:  38 22 ff f0
+# CHECK-LE:      b8:  0f 00 43 38
+# CHECK-LE-NEXT: bc:  00 00 00 60   nop
+# CHECK-LE:      LAB1:
+# CHECK-LE-NEXT: c0:  ff ff 01 06
+# CHECK-LE-NEXT: c4:  f0 ff 22 38
+LAB1: paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+paddi 1, 2, 8589934576, 0
+addi 2, 3, 15             # 60
+# CHECK-BE:      f8:  38 43 00 0f
+# CHECK-BE:      LAB2:
+# CHECK-BE-NEXT: fc:  60 00 00 00   nop
+# CHECK-BE-NEXT: 100: 06 01 ff ff
+# CHECK-BE-NEXT: 104: 38 22 ff f0
+# CHECK-LE:      f8:  0f 00 43 38
+# CHECK-LE:      LAB2:
+# CHECK-LE-NEXT: fc:  00 00 00 60   nop
+# CHECK-LE-NEXT: 100: ff ff 01 06
+# CHECK-LE-NEXT: 104: f0 ff 22 38
+LAB2:
+  paddi 1, 2, 8589934576, 0
+
+
+