[ARM] Emmit data symbol for constant pool data

[dnsampaio]

Summary

The ARM elf emitter would omit printing data symbol when constant data. This patch overrides the emitFill method as to enforce that the symbol is correctly printed.

[olista01]

lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp

1481

Why can't we just call EmitDataMappingSymbol, like EmitBytes does?

test/CodeGen/ARM/CheckDataSymbol.ll

1

I don't think this test is necessary, since this can be tested with an assembly file.

test/CodeGen/ARM/CheckDataSymbol.s

17

This test can be reduced greatly, to something like this:

.text
nop
.zero 4

[dnsampaio]

lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp

1481

Indeed it can.

[dnsampaio]

Simplified tests. Using the standard function for emitting data symbols, if required.

[olista01]

test/CodeGen/ARM/CheckDataSymbol.s

2

We can't use clang in LLVM tests (LLVM can be built without clang), this should use llvm-mc instead. Also, this should be in tests/MC/ARM, not tests/CodeGen/ARM, since it's testing stuff in the MC layer.

[dnsampaio]

Updated test to use llvm-mc instead of clang.

[olista01]

LGTM with one nit.

test/MC/ARM/CheckDataSymbol.s

1 ↗

(On Diff #167912)

The -arm-implicit-it=thumb option isn't needed here.

lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp

public:
   void emitPersonality(const MCSymbol *Per);
   void emitPersonalityIndex(unsigned index);
   void emitHandlerData();
   void emitSetFP(unsigned NewFpReg, unsigned NewSpReg, int64_t Offset = 0);
   void emitMovSP(unsigned Reg, int64_t Offset = 0);
   void emitPad(int64_t Offset);
   void emitRegSave(const SmallVectorImpl<unsigned> &RegList, bool isVector);
   void emitUnwindRaw(int64_t Offset, const SmallVectorImpl<uint8_t> &Opcodes);
+  void emitFill(const MCExpr &NumBytes, uint64_t FillValue, SMLoc Loc) override;
 
   void ChangeSection(MCSection *Section, const MCExpr *Subsection) override {
     LastMappingSymbols[getCurrentSection().first] = std::move(LastEMSInfo);
     MCELFStreamer::ChangeSection(Section, Subsection);
     auto LastMappingSymbol = LastMappingSymbols.find(Section);
     if (LastMappingSymbol != LastMappingSymbols.end()) {
       LastEMSInfo = std::move(LastMappingSymbol->second);
       return;

 
 void ARMELFStreamer::emitUnwindRaw(int64_t Offset,
                                    const SmallVectorImpl<uint8_t> &Opcodes) {
   FlushPendingOffset();
   SPOffset = SPOffset - Offset;
   UnwindOpAsm.EmitRaw(Opcodes);
 }
 
+void ARMELFStreamer::emitFill(const MCExpr &NumBytes, uint64_t FillValue,
+                              SMLoc Loc) {
+  if (LastEMSInfo->State == EMS_None) {
+    EmitMappingSymbol("$d");
+    LastEMSInfo->State = EMS_Data;
+  } else
+    EmitDataMappingSymbol();
+  MCObjectStreamer::emitFill(NumBytes, FillValue, Loc);
+}
+
 namespace llvm {
 
 MCTargetStreamer *createARMTargetAsmStreamer(MCStreamer &S,
                                              formatted_raw_ostream &OS,
                                              MCInstPrinter *InstPrint,
                                              bool isVerboseAsm) {
   return new ARMTargetAsmStreamer(S, OS, *InstPrint, isVerboseAsm);
 }

test/CodeGen/ARM/CheckDataSymbol.ll

+; RUN: llc -filetype=obj -mtriple=thumbv7-arm-none-eabi -o - %s | llvm-readobj -s -t - | FileCheck %s
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv7-arm-none-eabi"
+
+%struct.S = type { i32 }
+
+@.str = private unnamed_addr constant [4 x i8] c"%p\0A\00", align 1
+@defaultS = internal constant %struct.S zeroinitializer, align 4 #0
+
+define hidden void @f(i8* %p) local_unnamed_addr {
+entry:
+  %0 = tail call i32 (i8*, ...) @__2printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i32 0, i32 0), i8* %p)
+  ret void
+}
+
+define hidden void @g() local_unnamed_addr #1 {
+entry:
+  %0 = tail call i32 (i8*, ...) @__2printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i32 0, i32 0), i8* bitcast (%struct.S* @defaultS to i8*))
+  ret void
+}
+
+define hidden i32 @main() local_unnamed_addr  {
+entry:
+  %0 = tail call i32 (i8*, ...) @__2printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i32 0, i32 0), i8* bitcast (%struct.S* @defaultS to i8*))
+  ret i32 0
+}
+
+declare i32 @__2printf(i8*, ...)
+
+attributes #0 = { "rodata-section"="os_code" }
+attributes #1 = { "implicit-section-name"="os_code" }
+
+; CHECK: Name: $d.{{[0-9]+ \([1-9][0-9]*\)}}
+; CHECK-NEXT: Value: [[POSITION:0x[1-9][0-9]*]]
+; CHECK-NEXT: Size: 0
+; CHECK-NEXT: Binding: Local (0x0)
+; CHECK-NEXT: Type: None (0x0)
+; CHECK-NEXT: Other: 0
+; CHECK-NEXT: Section: os_code (0x6)
+; CHECK: Name: defaultS ({{[1-9][0-9]*}})
+; CHECK-NEXT: Value: [[POSITION]]
+; CHECK-NEXT: Size: 4
+; CHECK-NEXT: Binding: Local (0x0)
+; CHECK-NEXT: Type: Object (0x1)
+; CHECK-NEXT: Other: 0
+; CHECK-NEXT: Section: os_code (0x6)
+; CHECK-NEXT: }

test/CodeGen/ARM/CheckDataSymbol.s

+# RUN: clang -c -target arm-arm-none-eabi -mcpu=cortex-a9 -O2 -mthumb %s -o - | llvm-readobj -s -t - | FileCheck %s
+# CHECK: Name: $d.{{[0-9]+ \([1-9][0-9]*\)}}
+# CHECK-NEXT: Value: [[POSITION:0x[1-9][0-9]*]]
+# CHECK-NEXT: Size: 0
+# CHECK-NEXT: Binding: Local (0x0)
+# CHECK-NEXT: Type: None (0x0)
+# CHECK-NEXT: Other: 0
+# CHECK-NEXT: Section: os_code (0x6)
+# CHECK: Name: defaultS ({{[1-9][0-9]*}})
+# CHECK-NEXT: Value: [[POSITION]]
+# CHECK-NEXT: Size: 4
+# CHECK-NEXT: Binding: Local (0x0)
+# CHECK-NEXT: Type: Object (0x1)
+# CHECK-NEXT: Other: 0
+# CHECK-NEXT: Section: os_code (0x6)
+# CHECK-NEXT: }
+  .text
+  .syntax unified
+  .eabi_attribute 67, "2.09"
+  .cpu  cortex-a9
+  .eabi_attribute 6, 10
+  .eabi_attribute 7, 65
+  .eabi_attribute 8, 1
+  .eabi_attribute 9, 2
+  .fpu  neon-fp16
+  .eabi_attribute 36, 1
+  .eabi_attribute 42, 1
+  .eabi_attribute 34, 0
+  .eabi_attribute 68, 1
+  .eabi_attribute 17, 1
+  .eabi_attribute 20, 1
+  .eabi_attribute 21, 1
+  .eabi_attribute 23, 3
+  .eabi_attribute 24, 1
+  .eabi_attribute 25, 1
+  .eabi_attribute 38, 1
+  .eabi_attribute 18, 4
+  .eabi_attribute 26, 2
+  .eabi_attribute 14, 0
+  .file "thumb-data.c"
+  .globl  f
+  .p2align  1
+  .type f,%function
+  .code 16
+  .thumb_func
+f:
+  .fnstart
+  mov r1, r0
+  movw  r0, :lower16:.L.str
+  movt  r0, :upper16:.L.str
+  b printf
+.Lfunc_end0:
+  .size f, .Lfunc_end0-f
+  .cantunwind
+  .fnend
+
+  .section  os_code,"ax",%progbits
+  .globl  g
+  .p2align  1
+  .type g,%function
+  .code 16
+  .thumb_func
+g:
+  .fnstart
+  movw  r0, :lower16:.L.str
+  movw  r1, :lower16:defaultS
+  movt  r0, :upper16:.L.str
+  movt  r1, :upper16:defaultS
+  b printf
+.Lfunc_end1:
+  .size g, .Lfunc_end1-g
+  .cantunwind
+  .fnend
+
+  .text
+  .globl  main
+  .p2align  1
+  .type main,%function
+  .code 16
+  .thumb_func
+main:
+  .fnstart
+  .save {r7, lr}
+  push  {r7, lr}
+  .setfp  r7, sp
+  mov r7, sp
+  movw  r0, :lower16:.L.str
+  movw  r1, :lower16:defaultS
+  movt  r0, :upper16:.L.str
+  movt  r1, :upper16:defaultS
+  bl  printf
+  movs  r0, #0
+  pop {r7, pc}
+.Lfunc_end2:
+  .size main, .Lfunc_end2-main
+  .cantunwind
+  .fnend
+
+  .type .L.str,%object
+  .section  .rodata.str1.1,"aMS",%progbits,1
+.L.str:
+  .asciz  "%p\n"
+  .size .L.str, 4
+
+  .type defaultS,%object
+  .section  os_code,"ax",%progbits
+  .p2align  2
+defaultS:
+  .zero 4
+  .size defaultS, 4
+
+
+  .ident  "clang version 8.0.0 (trunk 340934)"
+  .section  ".note.GNU-stack","",%progbits
+  .addrsig
+  .addrsig_sym defaultS
+  .eabi_attribute 30, 1