[AArch64] Implement pseudo LDR <reg>, =<literal/label>

[weimingz]

Summary

See bugzilla: http://llvm.org/bugs/show_bug.cgi?id=20056

[dpeixott]

Overall this looks good to me, but I think we should split this into two commits: 1 factoring commit that moves the constant pools to a place that can be accessed by both AArch64 and ARM backends, and a second commit to implement the ldr= psuedo for AArch64.

test/MC/AArch64/ldr-pseudo.s

2

Either add the corresponding test for darwin, or delete this comment. I'm not sure if it is applicable for aarch64, or if the syntax is now unified between linux and darwin.

[dpeixott]

Adding llvm-commits to CC.

[weimingz]

This is the refactoring part.

[dpeixott]

LGTM with a suitable commit message explaining why we want to split out the implementation.

[weimingz]

In D4163#9, @dpeixott wrote:

LGTM with a suitable commit message explaining why we want to split out the implementation.

Thanks for reviewing, David. I committed the refactoring part as r211198.

[weimingz]

This is the second part (for aarch64).

[dpeixott]

Minor cleanup, otherwise LGTM.

As a suggestion for the future I think it would be better to start a new differential revision for each patch. Also, if you add "Differential Revison: <url>" to the commit message it will close it automatically on commit.

lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp

3031

use cast<> instead of dyn_cast<> when not checking the return value for null.

[dpeixott]

LGTM.

[weimingz]

Closed by commit rL211533 (authored by @weimingz).

include/llvm/MC/MCStreamer.h

   virtual void finish();
 };
 
+class AArch64TargetStreamer : public MCTargetStreamer {
+public:
+  AArch64TargetStreamer(MCStreamer &S);
+  ~AArch64TargetStreamer();
+
+
+  void finish() override;
+
+  /// Callback used to implement the ldr= pseudo.
+  /// Add a new entry to the constant pool for the current section and return an
+  /// MCExpr that can be used to refer to the constant pool location.
+  const MCExpr *addConstantPoolEntry(const MCExpr *);
+
+  /// Callback used to implemnt the .ltorg directive.
+  /// Emit contents of constant pool for the current section.
+  void emitCurrentConstantPool();
+
+private:
+  std::unique_ptr<AssemblerConstantPools> ConstantPools;
+};
+
 // FIXME: declared here because it is used from
 // lib/CodeGen/AsmPrinter/ARMException.cpp.
 class ARMTargetStreamer : public MCTargetStreamer {

lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp

   MCSubtargetInfo &STI;
   MCAsmParser &Parser;
 
+  AArch64TargetStreamer &getTargetStreamer() {
+    MCTargetStreamer &TS = *getParser().getStreamer().getTargetStreamer();
+    return static_cast<AArch64TargetStreamer &>(TS);
+  }
+
   MCAsmParser &getParser() const { return Parser; }
   MCAsmLexer &getLexer() const { return Parser.getLexer(); }
 

   bool parseDirectiveTLSDescCall(SMLoc L);
 
   bool parseDirectiveLOH(StringRef LOH, SMLoc L);
+  bool parseDirectiveLtorg(SMLoc L);
 
   bool validateInstruction(MCInst &Inst, SmallVectorImpl<SMLoc> &Loc);
   bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,

                  const MCTargetOptions &Options)
       : MCTargetAsmParser(), STI(_STI), Parser(_Parser) {
     MCAsmParserExtension::Initialize(_Parser);
+    if (Parser.getStreamer().getTargetStreamer() == nullptr)
+      new AArch64TargetStreamer(Parser.getStreamer());
 
     // Initialize the set of available features.
     setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits()));

     Operands.push_back(AArch64Operand::CreateImm(ImmVal, S, E, getContext()));
     return false;
   }
+  case AsmToken::Equal: {
+    SMLoc Loc = Parser.getTok().getLoc();
+    if (Mnemonic != "ldr") // only parse for ldr pseudo (e.g. ldr r0, =val)
+      return Error(Loc, "unexpected token in operand");
+    Parser.Lex(); // Eat '='
+    const MCExpr *SubExprVal;
+    if (getParser().parseExpression(SubExprVal))
+      return true;
+
+    MCContext& Ctx = getContext();
+    E = SMLoc::getFromPointer(Loc.getPointer() - 1);
+    // If the op is an imm and can be fit into a mov, then replace ldr with mov.
+    if (isa<MCConstantExpr>(SubExprVal) && Operands.size() >= 2 &&
+        static_cast<AArch64Operand &>(*Operands[1]).isReg()) {
+      bool IsXReg =  AArch64MCRegisterClasses[AArch64::GPR64allRegClassID].contains(
+            Operands[1]->getReg());
+      uint64_t Imm = (dyn_cast<MCConstantExpr>(SubExprVal))->getValue();
+      uint32_t ShiftAmt = 0, MaxShiftAmt = IsXReg ? 48 : 16;
+      while(Imm > 0xFFFF && countTrailingZeros(Imm) >= 16) {
+        ShiftAmt += 16;
+        Imm >>= 16;
+      }
+      if (ShiftAmt <= MaxShiftAmt && Imm <= 0xFFFF) {
+          Operands[0] = AArch64Operand::CreateToken("movz", false, Loc, Ctx);
+          Operands.push_back(AArch64Operand::CreateImm(
+                     MCConstantExpr::Create(Imm, Ctx), S, E, Ctx));
+        if (ShiftAmt)
+          Operands.push_back(AArch64Operand::CreateShiftExtend(AArch64_AM::LSL,
+                     ShiftAmt, true, S, E, Ctx));
+        return false;
+      }
+    }
+    // If it is a label or an imm that cannot fit in a movz, put it into CP.
+    const MCExpr *CPLoc = getTargetStreamer().addConstantPoolEntry(SubExprVal);
+    Operands.push_back(AArch64Operand::CreateImm(CPLoc, S, E, Ctx));
+    return false;
+  }
   }
 }
 

     return parseDirectiveWord(8, Loc);
   if (IDVal == ".tlsdesccall")
     return parseDirectiveTLSDescCall(Loc);
-
+  if (IDVal == ".ltorg" || IDVal == ".pool")
+    return parseDirectiveLtorg(Loc);
   return parseDirectiveLOH(IDVal, Loc);
 }
 

   return false;
 }
 
+/// parseDirectiveLtorg
+///  ::= .ltorg | .pool
+bool AArch64AsmParser::parseDirectiveLtorg(SMLoc L) {
+  getTargetStreamer().emitCurrentConstantPool();
+  return false;
+}
+
 bool
 AArch64AsmParser::classifySymbolRef(const MCExpr *Expr,
                                     AArch64MCExpr::VariantKind &ELFRefKind,

lib/Target/AArch64/MCTargetDesc/AArch64TargetStreamer.cpp

+//===- AArch64TargetStreamer.cpp - AArch64TargetStreamer class --*- C++ -*---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the AArch64TargetStreamer class.
+//
+//===----------------------------------------------------------------------===//
+#include "llvm/ADT/MapVector.h"
+#include "llvm/MC/ConstantPools.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCStreamer.h"
+
+using namespace llvm;
+
+//
+// AArch64TargetStreamer Implemenation
+//
+AArch64TargetStreamer::AArch64TargetStreamer(MCStreamer &S)
+    : MCTargetStreamer(S), ConstantPools(new AssemblerConstantPools()) {}
+
+AArch64TargetStreamer::~AArch64TargetStreamer() {}
+
+// The constant pool handling is shared by all AArch64TargetStreamer
+// implementations.
+const MCExpr *AArch64TargetStreamer::addConstantPoolEntry(const MCExpr *Expr) {
+  return ConstantPools->addEntry(Streamer, Expr);
+}
+
+void AArch64TargetStreamer::emitCurrentConstantPool() {
+  ConstantPools->emitForCurrentSection(Streamer);
+}
+
+// finish() - write out any non-empty assembler constant pools.
+void AArch64TargetStreamer::finish() { ConstantPools->emitAll(Streamer); }

lib/Target/AArch64/MCTargetDesc/CMakeLists.txt

   AArch64MCExpr.cpp
   AArch64MCTargetDesc.cpp
   AArch64MachObjectWriter.cpp
+  AArch64TargetStreamer.cpp
 )
 add_dependencies(LLVMAArch64Desc AArch64CommonTableGen)
 

test/CodeGen/AArch64/inlineasm-ldr-pseudo.ll

+; We actually need to use -filetype=obj in this test because if we output
+; assembly, the current code path will bypass the parser and just write the
+; raw text out to the Streamer. We need to actually parse the inlineasm to
+; demonstrate the bug. Going the asm->obj route does not show the issue.
+; RUN: llc -mtriple=aarch64   < %s -filetype=obj | llvm-objdump -arch=aarch64 -d - | FileCheck %s
+
+; CHECK-LABEL: foo:
+; CHECK:       a0 79 95 d2   movz x0, #0xabcd
+; CHECK:       c0 03 5f d6   ret
+define i32 @foo() nounwind {
+entry:
+  %0 = tail call i32 asm sideeffect "ldr $0,=0xabcd", "=r"() nounwind
+  ret i32 %0
+}
+; CHECK-LABEL: bar:
+; CHECK:        40 00 00 58                                      ldr    x0, #8
+; CHECK:        c0 03 5f d6                                      ret
+; Make sure the constant pool entry comes after the return
+; CHECK-LABEL:        $d.1:
+define i32 @bar() nounwind {
+entry:
+  %0 = tail call i32 asm sideeffect "ldr $0,=0x10001", "=r"() nounwind
+  ret i32 %0
+}
+
+

test/MC/AArch64/ldr-pseudo-obj-errors.s

+//RUN: not llvm-mc -arch aarch64 -filetype=obj %s -o %t1 2> %t2
+//RUN: cat %t2 | FileCheck %s
+
+//These tests look for errors that should be reported for invalid object layout
+//with the ldr pseudo. They are tested separately from parse errors because they
+//only trigger when the file has successfully parsed and the object file is about
+//to be written out.
+
+.text
+foo:
+  ldr x0, =0x10111
+  .space 0xdeadb0
+// CHECK: LVM ERROR: fixup value out of range

test/MC/AArch64/ldr-pseudo.s

+//RUN: llvm-mc -arch aarch64    %s | FileCheck %s
+
+//
+// Check that large constants are converted to ldr from constant pool
+//
+// simple test
+.section a, "ax", @progbits
+// CHECK-LABEL: f1:
+f1:
+  ldr x0, =0x1234
+// CHECK: movz    x0, #0x1234
+  ldr w1, =0x4567
+// CHECK:  movz    w1, #0x4567
+  ldr x0, =0x12340000
+// CHECK:  movz    x0, #0x1234, lsl #16
+  ldr w1, =0x45670000
+// CHECK: movz    w1, #0x4567, lsl #16
+  ldr x0, =0xabc00000000
+// CHECK: movz    x0, #0xabc, lsl #32
+  ldr x0, =0xbeef000000000000
+// CHECK: movz    x0, #0xbeef, lsl #48
+
+.section b,"ax",@progbits
+// CHECK-LABEL: f3:
+f3:
+  ldr x0, =0x10001
+// CHECK: ldr x0, .Ltmp[[TMP0:[0-9]+]]
+
+// loading multiple constants
+.section c,"ax",@progbits
+// CHECK-LABEL: f4:
+f4:
+  ldr x0, =0x10002
+// CHECK: ldr x0, .Ltmp[[TMP1:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  ldr x0, =0x10003
+// CHECK: ldr x0, .Ltmp[[TMP2:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+// TODO: the same constants should have the same constant pool location
+.section d,"ax",@progbits
+// CHECK-LABEL: f5:
+f5:
+  ldr x0, =0x10004
+// CHECK: ldr x0, .Ltmp[[TMP3:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  ldr x0, =0x10004
+// CHECK: ldr x0, .Ltmp[[TMP4:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+// a section defined in multiple pieces should be merged and use a single constant pool
+.section e,"ax",@progbits
+// CHECK-LABEL: f6:
+f6:
+  ldr x0, =0x10006
+// CHECK: ldr x0, .Ltmp[[TMP5:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+.section f, "ax", @progbits
+// CHECK-LABEL: f7:
+f7:
+  adds x0, x0, #1
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+.section e, "ax", @progbits
+// CHECK-LABEL: f8:
+f8:
+  adds x0, x0, #1
+  ldr x0, =0x10007
+// CHECK: ldr x0, .Ltmp[[TMP6:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+//
+// Check that symbols can be loaded using ldr pseudo
+//
+
+// load an undefined symbol
+.section g,"ax",@progbits
+// CHECK-LABEL: f9:
+f9:
+  ldr x0, =foo
+// CHECK: ldr x0, .Ltmp[[TMP7:[0-9]+]]
+
+// load a symbol from another section
+.section h,"ax",@progbits
+// CHECK-LABEL: f10:
+f10:
+  ldr x0, =f5
+// CHECK: ldr x0, .Ltmp[[TMP8:[0-9]+]]
+
+// load a symbol from the same section
+.section i,"ax",@progbits
+// CHECK-LABEL: f11:
+f11:
+  ldr x0, =f12
+// CHECK: ldr x0, .Ltmp[[TMP9:[0-9]+]]
+  ldr w0,=0x3C000
+// CHECK: ldr     w0, .Ltmp[[TMP10:[0-9]+]]
+
+// CHECK-LABEL: f12:
+f12:
+  adds x0, x0, #1
+  adds x0, x0, #1
+
+.section j,"ax",@progbits
+// mix of symbols and constants
+// CHECK-LABEL: f13:
+f13:
+  adds x0, x0, #1
+  adds x0, x0, #1
+  ldr x0, =0x101
+// CHECK: movz x0, #0x101
+  adds x0, x0, #1
+  adds x0, x0, #1
+  ldr x0, =bar
+// CHECK: ldr x0, .Ltmp[[TMP11:[0-9]+]]
+  adds x0, x0, #1
+  adds x0, x0, #1
+//
+// Check for correct usage in other contexts
+//
+
+// usage in macro
+.macro useit_in_a_macro
+  ldr x0, =0x10008
+  ldr x0, =baz
+.endm
+.section k,"ax",@progbits
+// CHECK-LABEL: f14:
+f14:
+  useit_in_a_macro
+// CHECK: ldr x0, .Ltmp[[TMP12:[0-9]+]]
+// CHECK: ldr x0, .Ltmp[[TMP13:[0-9]+]]
+
+// usage with expressions
+.section l, "ax", @progbits
+// CHECK-LABEL: f15:
+f15:
+  ldr x0, =0x10001+8
+// CHECK: ldr x0, .Ltmp[[TMP14:[0-9]+]]
+  adds x0, x0, #1
+  ldr x0, =bar+4
+// CHECK: ldr x0, .Ltmp[[TMP15:[0-9]+]]
+  adds x0, x0, #1
+
+//
+// Constant Pools
+//
+// CHECK: .section b,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP0]]
+// CHECK: .word 65537
+
+// CHECK: .section c,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP1]]
+// CHECK: .word 65538
+// CHECK: .Ltmp[[TMP2]]
+// CHECK: .word 65539
+
+// CHECK: .section d,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP3]]
+// CHECK: .word 65540
+// CHECK: .Ltmp[[TMP4]]
+// CHECK: .word 65540
+
+// CHECK: .section e,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP5]]
+// CHECK: .word 65542
+// CHECK: .Ltmp[[TMP6]]
+// CHECK: .word 65543
+
+// Should not switch to section because it has no constant pool
+// CHECK-NOT: .section f,"ax",@progbits
+
+// CHECK: .section g,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP7]]
+// CHECK: .word foo
+
+// CHECK: .section h,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP8]]
+// CHECK: .word f5
+
+// CHECK: .section i,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP9]]
+// CHECK: .word f12
+// CHECK: .Ltmp[[TMP10]]
+// CHECK: .word 245760
+
+// CHECK: .section j,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP11]]
+// CHECK: .word bar
+
+// CHECK: .section k,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP12]]
+// CHECK: .word 65544
+// CHECK: .Ltmp[[TMP13]]
+// CHECK: .word baz
+
+// CHECK: .section l,"ax",@progbits
+// CHECK: .align 2
+// CHECK: .Ltmp[[TMP14]]
+// CHECK: .word 65545
+// CHECK: .Ltmp[[TMP15]]
+// CHECK: .word bar+4