[mips] Expand JAL instructions when PIC is enabled.

[dsanders]

Summary

This is the correct way to handle JAL instructions when PIC is enabled.

[tomatabacu]

-We now emit a correct JALR instruction, which makes us unable to add the R_(MICRO)MIPS_JALR relocation after it.
This is fine because that relocation is just an optimization hint for the linker and is not needed for correctness.
This means that we don't need to implement that relocation any more.

-Added a FIXME which explains this situation.

-Made tests match with the changes mentioned above.

[tomatabacu]

Fix a spelling mistake in a comment.

[tomatabacu]

Ping.

[tomatabacu]

Ping 2.

[tomatabacu]

Ping 3.

[tomatabacu]

Updated tests because of fix to the microMIPS disassembler.
Rebased.

[dsanders]

Sorry for taking so long on these. I had to find out how this is supposed to work before I could comment.

A couple general comments on the patch:

• There's no mention of -mxgot. I'm happy for -mxgot to be a follow up patch but could you insert assertions/unreachables to guard against it for now.
• Not required but consider refactoring into smaller functions. It's difficult to see the overall logic at the moment.
• You'll need to change the commands to use -target-abi

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1462–1465

Nit: This would be a little tidier as a single statement. We never use JalOpnd, and JalSymExpr again.

1467

I doubt this condition is right, although it's probably close enough to cover a lot of cases.

Things that spring to mind:

• Are forward-declared locals handled correctly?
• Are weak symbols handled correctly?
• What if the symbol is a section name like .text?

1479–1487

These paths should be merged and we should use JalSym in both cases. The .text you see in objdump's output stems from a GAS optimisation to reduce the number of symbols it has to maintain.

Also, use JalSym instead of JalSym.getName() so we don't lose the offset on things like 'jal foo+4'

1494–1500

Same as above

1511

LD for N64

1514–1523

Same as above

1526–1539

Is this doing the right thing for N64?

1542–1545

Nit: JalrInst.setOpcode(inMicroMipsMode() ? Mips::JALR_MM : Mips::JALR);

[dsanders]

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1514–1523

Same as above

(I mean the above JalSym comment, not the N64 one)

[tomatabacu]

Partially addressed review comments.
Changed tests to use -target-abi instead of -mattr.
Removed the .set reorder test (redundant).
Switched away from using llvm-objdump (doesn't work for microMIPS on N64).
Added support for local labels (1b, 2f, etc.).
Updated comments.

[tomatabacu]

Regarding -mxgot, AFAICT it is never checked for in the IAS. There is a command-line option used in llc, but I can't find a way to check for it from MipsAsmParser.
In the meantime, I've added a FIXME comment.

Also, note that offsets do not work (e.g. jal label+8). I can get the value of the offset, but I can't use it as an operand for the ADDIU, in the case of O32,
or store it in the relocation addendum, in the case of N32/N64. I think we would need to improve relocation emission in order to properly handle offsets.

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1462–1465

Done.

1467

Are forward-declared locals handled correctly?
Unfortunately, they are not (I assume you're talking about doing a "jal fwd_label" before the definition of fwd_label). They end up being treated as global symbols. I'm not sure how to fix this.

Are weak symbols handled correctly?
AFAICT weak symbols are treated the same way as global symbols.

What if the symbol is a section name like .text?
.text should be treated as a local symbol, but there is something weird happening: with llvm-objdump it's treated as a local symbol, but with plain llvm-mc, .text is treated as a global symbol.

1479–1487

Done.

1494–1500

Done.

1511

Fixed.

1514–1523

Done.

1526–1539

Except for not generating LD on N64 (now fixed), I believe so.

1542–1545

Done.

[dsanders]

Regarding -mxgot, AFAICT it is never checked for in the IAS. There is a
command-line option used in llc, but I can't find a way to check for it from
MipsAsmParser. In the meantime, I've added a FIXME comment.

Ok. We'll have to add that at some point.

Also, note that offsets do not work (e.g. jal label+8). I can get the value
of the offset, but I can't use it as an operand for the ADDIU, in the case
of O32, or store it in the relocation addendum, in the case of N32/N64.
I think we would need to improve relocation emission in order to properly handle offsets.

I believe the problem is that you are extracting the symbol from the expression, then creating a new expression from the symbol. The original expression should have the offset in it and will result in relocations for label+offset.

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1467

Are forward-declared locals handled correctly?
Unfortunately, they are not (I assume you're talking about doing a "jal fwd_label" before the definition of fwd_label). They end up being treated as global symbols. I'm not sure how to fix this.

Yes, I'm talking about cases like:

  jal foo
foo:

I've just experimented with it and it turns out that they do work as expected and are treated as local symbols.

Are weak symbols handled correctly?
AFAICT weak symbols are treated the same way as global symbols.

Apart from a duplicate nop which needs fixing, this is behaving the same as GAS too.

What if the symbol is a section name like .text?
.text should be treated as a local symbol, but there is something weird happening: with llvm-objdump it's treated as a local symbol, but with plain llvm-mc, .text is treated as a global symbol.

I see this odd behaviour too. We need llvm-mc to do the right thing for testing purposes so could you look into this? My guess is that there's a difference in how the .text section is created.

1526–1539

Thanks. I've experimented a bit with GAS too and my remaining doubts are now settled. The relocations tend to change for N64 but this is one case where it doesn't.

[seanbruno]

I think this needs a regen against head as it fails to apply cleanly as of today.

[tomatabacu]

Rebased on top of D9938 which fixes invalid names for temp symbol operands (e.g. $tmp00, $tmp10).
Added .set nomacro warning & test case.
Added the desired FIXME output for the .text test case.
Rebased.

[tomatabacu]

I have found a potential fix for the discrepancy between the asm and obj outputs when using .text in a PIC relocation,
but I don't have an arch-independent test for it yet and it breaks a lot of NVPTX tests, for some reason.

[tomatabacu]

Rebased (for Sean Bruno's convenience).

[dsanders]

In D6231#177071, @tomatabacu wrote:

I have found a potential fix for the discrepancy between the asm and obj outputs when using .text in a PIC relocation,
but I don't have an arch-independent test for it yet and it breaks a lot of NVPTX tests, for some reason.

If it's difficult or a significant amount of work then I'd prefer to get this patch committed and handle that as a follow-up.

Looking back through the comments, I believe the remaining issues are:

• Duplicated nops
• The handling of .text
• Offsets are lost for label+offset

Even without resolving those details this is still a significant step forwards.

Are there any other outstanding issues? If not, I think it's ok to commit whats done so far and fix the smaller details in follow-ups.

[tomatabacu]

Improved the code which gets the JalSym from the original JalExpr.
Moved relocation creation before MCInst creation in order to improve readability. (NFC)
Added FIXME comments for the remaining problems.

[tomatabacu]

In D6231#187264, @dsanders wrote:

Looking back through the comments, I believe the remaining issues are:

• Duplicated nops

I'm not sure if this is accurate. I've compared IAS and GAS with .set reorder/noreorder and
with/without manually added NOPs and I haven't seen any discrepancies.

• The handling of .text

I have found a fix for this, but it's a bit ugly and it only touches common code.
If I include it here, it might slow down this patch even more.

• Offsets are lost for label+offset

They used to be lost, but since I switched the code to use evaluateRelocExpr() offsets will cause a fatal error,
because it tries to inappropriately apply relocations to constants.

I'd rather have this than silently generating the wrong code.

However, evaluateRelocExpr() needs to be fixed.

Some other remaining problems are:

• no support for the LargeGOT case
• no support for forward-declared locals

Also, this patch will cause a recently added RuntimeDyld test to fail (ExecutionEngine/RuntimeDyld/Mips/ELF_Mips64r2N64_PIC_relocations.s).
I don't know how to fix this test.

[dsanders]

In D6231#190332, @tomatabacu wrote:

In D6231#187264, @dsanders wrote:

Looking back through the comments, I believe the remaining issues are:

• Duplicated nops

I'm not sure if this is accurate. I've compared IAS and GAS with .set reorder/noreorder and
with/without manually added NOPs and I haven't seen any discrepancies.

In that case it must have been fixed recently.

• The handling of .text

I have found a fix for this, but it's a bit ugly and it only touches common code.
If I include it here, it might slow down this patch even more.

I'm ok with leaving that for another patch.

• Offsets are lost for label+offset

They used to be lost, but since I switched the code to use evaluateRelocExpr() offsets will cause a fatal error,
because it tries to inappropriately apply relocations to constants.

I'd rather have this than silently generating the wrong code.
However, evaluateRelocExpr() needs to be fixed.

I agree that an error is better than silently producing bad code and that evaluateRelocExpr() is badly flawed and needs fixing..

Some other remaining problems are:

• no support for the LargeGOT case

That's ok for this patch, but following the principle above I'd suggest asserting that LargeGOT is not in use.

• no support for forward-declared locals

Didn't we establish earlier in the review that these worked as expected? Has that changed?

Also, this patch will cause a recently added RuntimeDyld test to fail (ExecutionEngine/RuntimeDyld/Mips/ELF_Mips64r2N64_PIC_relocations.s).
I don't know how to fix this test.

You need to disable pic for the 'jal foo' on line 44 using '.option pic0' and '.option pic2'. At the moment, it's trying to read the bottom 26-bits of a jal instruction but the new expansion means it's really reading the bottom 26-bits of an lw.

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1273–1274

Nit: Unnecessary braces

1461

Nit: psudo->pseudo

1468–1473

I'm not sure I understand this bit. Can you explain?

1511

Nit: isABI_N64() -> ABI.ArePtrs64bit()

[tomatabacu]

Fixed the RuntimeDyld test.
Addressed the other review comments.

[tomatabacu]

In D6231#191632, @dsanders wrote:

That's ok for this patch, but following the principle above I'd suggest asserting that LargeGOT is not in use.

I just realized that llvm-mc doesn't even accept -mxgot on the command line (it's also not listed in --help-hidden).
It seems that -mxgot is a hidden option only for llc.

Didn't we establish earlier in the review that these worked as expected? Has that changed?

I'm not sure we did. We're still treating forward-declared local symbols as global, while GAS treats them as local.
Just to be clear, I'm talking the following:

.option pic2
  jal foo
foo:

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1273–1274

Fixed.

1461

Fixed.

1468–1473

GAS spews all kinds of errors if it is given 2 symbols here, so we should report an error as well.
Also, if we didn't, the IAS would (currently) segfault.

1511

Fixed.

[seanbruno]

Thanks for the work on this!

[dsanders]

Commandeered since Toma has finished his placement with us.

In D6231#191632, @dsanders wrote:
That's ok for this patch, but following the principle above I'd suggest asserting that LargeGOT is not in use.
I just realized that llvm-mc doesn't even accept -mxgot on the command line (it's also not listed in --help-hidden).

It seems that -mxgot is a hidden option only for llc.

We'll need to add -mxgot in a later patch. For now, I'll just add a fixme.

Didn't we establish earlier in the review that these worked as expected? Has that changed?

I'm not sure we did. We're still treating forward-declared local symbols as global, while GAS treats them as local.
Just to be clear, I'm talking the following:

.option pic2

jal foo

foo:

I've decided not to worry about the handling of forward-declared local symbols. If we do still have issues, then we can fix them later. For now, it's more important to get Toma's remaining work committed.

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

1468–1473

I understand that Toma was checking for strange assembly like:

jal foo-bar

I agree that this should have a nice error message, but I wonder if this patch is checking in the right place. I think it may be a job for the code that figures out which reloc to use for the expression.

Having looked into this a bit, I think that handling this in the right place is related to representing our unary operators (%hi(), etc.) actual operators in MCExpr nodes. printExpr() ought to blindly print the expression it has whether valid or not, and the reloc selection should handle validity.

For now, I'll add a FIXME comment saying it should be removed once %hi(), etc are MCExpr operators.

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

switch (Opcode) {
     case Mips::BGEZALS_MM:
     case Mips::BLTZALS_MM:
       return true;
     default:
       return false;
   }
 }
 
+static const MCSymbol *getSingleMCSymbol(const MCExpr *Expr) {
+  if (const MCSymbolRefExpr *SRExpr = dyn_cast<MCSymbolRefExpr>(Expr)) {
+    return &(SRExpr->getSymbol());
+  }
+
+  if (const MCBinaryExpr *BExpr = dyn_cast<MCBinaryExpr>(Expr)) {
+    const MCSymbol *LHSSym = getSingleMCSymbol(BExpr->getLHS());
+    const MCSymbol *RHSSym = getSingleMCSymbol(BExpr->getRHS());
+
+    if (LHSSym)
+      return LHSSym;
+
+    if (RHSSym)
+      return RHSSym;
+
+    return nullptr;
+  }
+
+  if (const MCUnaryExpr *UExpr = dyn_cast<MCUnaryExpr>(Expr))
+    return getSingleMCSymbol(UExpr->getSubExpr());
+
+  return nullptr;
+}
+
+static unsigned countMCSymbolRefExpr(const MCExpr *Expr) {
+  if (isa<MCSymbolRefExpr>(Expr))
+    return 1;
+
+  if (const MCBinaryExpr *BExpr = dyn_cast<MCBinaryExpr>(Expr))
+    return countMCSymbolRefExpr(BExpr->getLHS()) +
+           countMCSymbolRefExpr(BExpr->getRHS());
+
+  if (const MCUnaryExpr *UExpr = dyn_cast<MCUnaryExpr>(Expr))
+    return countMCSymbolRefExpr(UExpr->getSubExpr());
+
+  return 0;
+}
+
 bool MipsAsmParser::processInstruction(MCInst &Inst, SMLoc IDLoc,
                                        SmallVectorImpl<MCInst> &Instructions) {
   const MCInstrDesc &MCID = getInstDesc(Inst.getOpcode());
 
   Inst.setLoc(IDLoc);
 
   if (MCID.isBranch() || MCID.isCall()) {
     const unsigned Opcode = Inst.getOpcode();
switch (Opcode) {
           return Error(IDLoc, "expected immediate operand kind");
         Imm = Opnd.getImm();
         if (!isInt<10>(Imm))
           return Error(IDLoc, "immediate operand value out of range");
         break;
     }
   }
 
+  // This expansion is not in a function called by expandInstruction() because
+  // the psudo-instruction doesn't have a distinct opcode.
+  if ((Inst.getOpcode() == Mips::JAL || Inst.getOpcode() == Mips::JAL_MM) &&
+      inPicMode()) {
+    warnIfNoMacro(IDLoc);
+
+    const MCExpr *JalExpr = Inst.getOperand(0).getExpr();
+
+    // We can do this expansion if there's only 1 symbol in the argument
+    // expression.
+    if (countMCSymbolRefExpr(JalExpr) > 1)
+      return Error(IDLoc, "jal doesn't support multiple symbols in PIC mode");
+
+    const MCSymbol *JalSym = getSingleMCSymbol(JalExpr);
+
+    // FIXME: Add support for label+offset operands (currently causes an error).
+    // FIXME: Add support for forward-declared local symbols.
+    // FIXME: Add expansion for when the LargeGOT option is enabled.
+    if (JalSym->isInSection() || JalSym->isTemporary()) {
+      if (isABI_O32()) {
+        // If it's a local symbol and the O32 ABI is being used, we expand to:
+        //  lw  ! $25, 0($gp)
+        //    R_(MICRO)MIPS_GOT16  label
+        //  addiu $25, $25, 0
+        //    R_(MICRO)MIPS_LO16   label
+        //  jalr  $25
+        const MCExpr *Got16RelocExpr = evaluateRelocExpr(JalExpr, "got");
+        const MCExpr *Lo16RelocExpr = evaluateRelocExpr(JalExpr, "lo");
+
+        MCInst LwInst;
+        LwInst.setOpcode(Mips::LW);
+        LwInst.addOperand(MCOperand::createReg(Mips::T9));
+        LwInst.addOperand(MCOperand::createReg(Mips::GP));
+        LwInst.addOperand(MCOperand::createExpr(Got16RelocExpr));
+        Instructions.push_back(LwInst);
+
+        MCInst AddiuInst;
+        AddiuInst.setOpcode(Mips::ADDiu);
+        AddiuInst.addOperand(MCOperand::createReg(Mips::T9));
+        AddiuInst.addOperand(MCOperand::createReg(Mips::T9));
+        AddiuInst.addOperand(MCOperand::createExpr(Lo16RelocExpr));
+        Instructions.push_back(AddiuInst);
+      } else if (isABI_N32() || isABI_N64()) {
+        // If it's a local symbol and the N32/N64 ABIs are being used,
+        // we expand to:
+        //  lw/ld  ! $25, 0($gp)
+        //    R_(MICRO)MIPS_GOT_DISP  label
+        //  jalr  $25
+        const MCExpr *GotDispRelocExpr = evaluateRelocExpr(JalExpr, "got_disp");
+
+        MCInst LoadInst;
+        LoadInst.setOpcode(isABI_N64() ? Mips::LD : Mips::LW);
+        LoadInst.addOperand(MCOperand::createReg(Mips::T9));
+        LoadInst.addOperand(MCOperand::createReg(Mips::GP));
+        LoadInst.addOperand(MCOperand::createExpr(GotDispRelocExpr));
+        Instructions.push_back(LoadInst);
+      }
+    } else {
+      // If it's an external/weak symbol, we expand to:
+      //  lw/ld    $25, 0($gp)
+      //    R_(MICRO)MIPS_CALL16  label
+      //  jalr  $25
+      const MCExpr *Call16RelocExpr = evaluateRelocExpr(JalExpr, "call16");
+
+      MCInst LoadInst;
+      LoadInst.setOpcode(isABI_N64() ? Mips::LD : Mips::LW);
+      LoadInst.addOperand(MCOperand::createReg(Mips::T9));
+      LoadInst.addOperand(MCOperand::createReg(Mips::GP));
+      LoadInst.addOperand(MCOperand::createExpr(Call16RelocExpr));
+      Instructions.push_back(LoadInst);
+    }
+
+    MCInst JalrInst;
+    JalrInst.setOpcode(inMicroMipsMode() ? Mips::JALR_MM : Mips::JALR);
+    JalrInst.addOperand(MCOperand::createReg(Mips::RA));
+    JalrInst.addOperand(MCOperand::createReg(Mips::T9));
+
+    // FIXME: Add an R_(MICRO)MIPS_JALR relocation after the JALR.
+    // This relocation is supposed to be an optimization hint for the linker
+    // and is not necessary for correctness.
+
+    Inst = JalrInst;
+  }
+
   if (MCID.mayLoad() || MCID.mayStore()) {
     // Check the offset of memory operand, if it is a symbol
     // reference or immediate we may have to expand instructions.
     for (unsigned i = 0; i < MCID.getNumOperands(); i++) {
       const MCOperandInfo &OpInfo = MCID.OpInfo[i];
       if ((OpInfo.OperandType == MCOI::OPERAND_MEMORY) ||
           (OpInfo.OperandType == MCOI::OPERAND_UNKNOWN)) {
         MCOperand &Op = Inst.getOperand(i);
         if (Op.isImm()) {
           int MemOffset = Op.getImm();

test/MC/Mips/expansion-jal-sym-pic.s

+# RUN: llvm-mc %s -arch=mips -mcpu=mips32 -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=NORMAL -check-prefix=O32
+
+# RUN: llvm-mc %s -arch=mips -mcpu=mips64 -target-abi n32 -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=NORMAL -check-prefix=N32
+
+# RUN: llvm-mc %s -arch=mips64 -mcpu=mips64 -target-abi n64 -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=NORMAL -check-prefix=N64
+
+# RUN: llvm-mc %s -arch=mips -mcpu=mips32 -mattr=micromips -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=MICROMIPS -check-prefix=O32-MICROMIPS
+
+# RUN: llvm-mc %s -arch=mips -mcpu=mips64 -target-abi n32 -mattr=micromips -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=MICROMIPS -check-prefix=N32-MICROMIPS
+
+# RUN: llvm-mc %s -arch=mips64 -mcpu=mips64 -target-abi n64 -mattr=micromips -show-encoding |\
+# RUN:   FileCheck %s -check-prefix=ALL -check-prefix=MICROMIPS -check-prefix=N64-MICROMIPS
+
+  .weak weak_label
+
+  .text
+  .option pic2
+
+  .ent local_label
+local_label:
+  .frame  $sp, 0, $ra
+  .set noreorder
+
+  jal local_label
+  nop
+
+  jal weak_label
+  nop
+
+  jal global_label
+  nop
+
+  jal .text
+  nop
+
+  # local labels ($tmp symbols)
+  jal 1f
+  nop
+
+  .end local_label
+
+1:
+  nop
+  add $8, $8, $8
+  nop
+
+# Expanding "jal local_label":
+# O32: lw     $25, %got(local_label)($gp)   # encoding: [0x8f,0x99,A,A]
+# O32:                                      #   fixup A - offset: 0, value: local_label@GOT, kind:   fixup_Mips_GOT_Local
+# O32: addiu  $25, $25, %lo(local_label)    # encoding: [0x27,0x39,A,A]
+# O32:                                      #   fixup A - offset: 0, value: local_label@ABS_LO, kind:   fixup_Mips_LO16
+
+# N32: lw  $25, %got_disp(local_label)($gp) # encoding: [0x8f,0x99,A,A]
+# N32:                                      #   fixup A - offset: 0, value: local_label@GOT_DISP, kind:   fixup_Mips_GOT_DISP
+
+# N64: ld  $25, %got_disp(local_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64:                                      #   fixup A - offset: 0, value: local_label@GOT_DISP, kind:   fixup_Mips_GOT_DISP
+
+# O32-MICROMIPS: lw    $25, %got(local_label)($gp)      # encoding: [0xff,0x3c,A,A]
+# O32-MICROMIPS:                                        #   fixup A - offset: 0, value: local_label@GOT, kind:   fixup_MICROMIPS_GOT16
+# O32-MICROMIPS: addiu $25, $25, %lo(local_label)       # encoding: [0x33,0x39,A,A]
+# O32-MICROMIPS:                                        #   fixup A - offset: 0, value: local_label@ABS_LO, kind:   fixup_MICROMIPS_LO16
+
+# N32-MICROMIPS: lw    $25, %got_disp(local_label)($gp) # encoding: [0xff,0x3c,A,A]
+# N32-MICROMIPS:                                        #   fixup A - offset: 0, value: local_label@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# N64-MICROMIPS: ld    $25, %got_disp(local_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64-MICROMIPS:                                        #   fixup A - offset: 0, value: local_label@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# NORMAL:    jalr $25      # encoding: [0x03,0x20,0xf8,0x09]
+# MICROMIPS: jalr $ra, $25 # encoding: [0x03,0xf9,0x0f,0x3c]
+# ALL:       nop           # encoding: [0x00,0x00,0x00,0x00]
+
+
+# Expanding "jal weak_label":
+# O32: lw  $25, %call16(weak_label)($gp) # encoding: [0x8f,0x99,A,A]
+# O32:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# N32: lw  $25, %call16(weak_label)($gp) # encoding: [0x8f,0x99,A,A]
+# N32:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# N64: ld  $25, %call16(weak_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# O32-MICROMIPS: lw  $25, %call16(weak_label)($gp) # encoding: [0xff,0x3c,A,A]
+# O32-MICROMIPS:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind:   fixup_MICROMIPS_CALL16
+
+# N32-MICROMIPS: lw  $25, %call16(weak_label)($gp) # encoding: [0xff,0x3c,A,A]
+# N32-MICROMIPS:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind: fixup_MICROMIPS_CALL16
+
+# N64-MICROMIPS: ld  $25, %call16(weak_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64-MICROMIPS:                                   #   fixup A - offset: 0, value: weak_label@GOT_CALL, kind: fixup_MICROMIPS_CALL16
+
+# NORMAL:    jalr $25      # encoding: [0x03,0x20,0xf8,0x09]
+# MICROMIPS: jalr $ra, $25 # encoding: [0x03,0xf9,0x0f,0x3c]
+# ALL:       nop           # encoding: [0x00,0x00,0x00,0x00]
+
+
+# Expanding "jal global_label":
+# O32: lw  $25, %call16(global_label)($gp) # encoding: [0x8f,0x99,A,A]
+# O32:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# N32: lw  $25, %call16(global_label)($gp) # encoding: [0x8f,0x99,A,A]
+# N32:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# N64: ld  $25, %call16(global_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind:   fixup_Mips_CALL16
+
+# O32-MICROMIPS: lw  $25, %call16(global_label)($gp) # encoding: [0xff,0x3c,A,A]
+# O32-MICROMIPS:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind: fixup_MICROMIPS_CALL16
+
+# N32-MICROMIPS: lw  $25, %call16(global_label)($gp) # encoding: [0xff,0x3c,A,A]
+# N32-MICROMIPS:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind: fixup_MICROMIPS_CALL16
+
+# N64-MICROMIPS: ld  $25, %call16(global_label)($gp) # encoding: [0xdf,0x99,A,A]
+# N64-MICROMIPS:                                     #   fixup A - offset: 0, value: global_label@GOT_CALL, kind: fixup_MICROMIPS_CALL16
+
+# NORMAL:    jalr $25      # encoding: [0x03,0x20,0xf8,0x09]
+# MICROMIPS: jalr $ra, $25 # encoding: [0x03,0xf9,0x0f,0x3c]
+# ALL:       nop           # encoding: [0x00,0x00,0x00,0x00]
+
+
+# FIXME: The .text section MCSymbol isn't created when printing assembly. However,
+# it is created when generating an ELF object file.
+# Expanding "jal .text":
+# O32-FIXME: lw    $25, %got(.text)($gp)           # encoding: [0x8f,0x99,A,A]
+# O32-FIXME:                                       #   fixup A - offset: 0, value: .text@GOT, kind: fixup_Mips_GOT_Local
+# O32-FIXME: addiu $25, $25, %lo(.text)            # encoding: [0x27,0x39,A,A]
+# O32-FIXME:                                       #   fixup A - offset: 0, value: .text@ABS_LO, kind: fixup_Mips_LO16
+
+# N32-FIXME: lw  $25, %got_disp(.text)($gp)        # encoding: [0x8f,0x99,A,A]
+# N32-FIXME:                                       #   fixup A - offset: 0, value: .text@GOT_DISP, kind: fixup_Mips_GOT_DISP
+
+# N64-FIXME: ld  $25, %got_disp(.text)($gp)        # encoding: [0xdf,0x99,A,A]
+# N64-FIXME:                                       #   fixup A - offset: 0, value: .text@GOT_DISP, kind: fixup_Mips_GOT_DISP
+
+# O32-MICROMIPS-FIXME: lw    $25, %got(.text)($gp)      # encoding: [0xff,0x3c,A,A]
+# O32-MICROMIPS-FIXME:                                  #   fixup A - offset: 0, value: .text@GOT, kind: fixup_MICROMIPS_GOT16
+# O32-MICROMIPS-FIXME: addiu $25, $25, %lo(.text)       # encoding: [0x33,0x39,A,A]
+# O32-MICROMIPS-FIXME:                                  #   fixup A - offset: 0, value: .text@ABS_LO, kind: fixup_MICROMIPS_LO16
+
+# N32-MICROMIPS-FIXME: lw    $25, %got_disp(.text)($gp) # encoding: [0xff,0x3c,A,A]
+# N32-MICROMIPS-FIXME:                                  #   fixup A - offset: 0, value: .text@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# N64-MICROMIPS-FIXME: ld    $25, %got_disp(.text)($gp) # encoding: [0xdf,0x99,A,A]
+# N64-MICROMIPS-FIXME:                                  #   fixup A - offset: 0, value: .text@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# NORMAL:    jalr $25      # encoding: [0x03,0x20,0xf8,0x09]
+# MICROMIPS: jalr $ra, $25 # encoding: [0x03,0xf9,0x0f,0x3c]
+# ALL:       nop           # encoding: [0x00,0x00,0x00,0x00]
+
+
+# Expanding "jal 1f":
+# O32: lw     $25, %got($tmp0)($gp)   # encoding: [0x8f,0x99,A,A]
+# O32:                                #   fixup A - offset: 0, value: ($tmp0)@GOT, kind:   fixup_Mips_GOT_Local
+# O32: addiu  $25, $25, %lo($tmp0)    # encoding: [0x27,0x39,A,A]
+# O32:                                #   fixup A - offset: 0, value: ($tmp0)@ABS_LO, kind:   fixup_Mips_LO16
+
+# N32: lw  $25, %got_disp($tmp0)($gp) # encoding: [0x8f,0x99,A,A]
+# N32:                                #   fixup A - offset: 0, value: ($tmp0)@GOT_DISP, kind:   fixup_Mips_GOT_DISP
+
+# N64: ld  $25, %got_disp($tmp0)($gp) # encoding: [0xdf,0x99,A,A]
+# N64:                                #   fixup A - offset: 0, value: ($tmp0)@GOT_DISP, kind:   fixup_Mips_GOT_DISP
+
+# O32-MICROMIPS: lw    $25, %got($tmp0)($gp)    # encoding: [0xff,0x3c,A,A]
+# O32-MICROMIPS:                                #   fixup A - offset: 0, value: ($tmp0)@GOT, kind: fixup_MICROMIPS_GOT16
+# O32-MICROMIPS: addiu $25, $25, %lo($tmp0)     # encoding: [0x33,0x39,A,A]
+# O32-MICROMIPS:                                #   fixup A - offset: 0, value: ($tmp0)@ABS_LO, kind: fixup_MICROMIPS_LO16
+
+# N32-MICROMIPS: lw  $25, %got_disp($tmp0)($gp) # encoding: [0xff,0x3c,A,A]
+# N32-MICROMIPS:                                #   fixup A - offset: 0, value: ($tmp0)@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# N64-MICROMIPS: ld  $25, %got_disp($tmp0)($gp) # encoding: [0xdf,0x99,A,A]
+# N64-MICROMIPS:                                #   fixup A - offset: 0, value: ($tmp0)@GOT_DISP, kind: fixup_MICROMIPS_GOT_DISP
+
+# NORMAL:    jalr $25      # encoding: [0x03,0x20,0xf8,0x09]
+# MICROMIPS: jalr $ra, $25 # encoding: [0x03,0xf9,0x0f,0x3c]
+# ALL:       nop           # encoding: [0x00,0x00,0x00,0x00]

test/MC/Mips/set-nomacro.s

# CHECK-NOT: warning: macro instruction expanded into multiple instructions
   bgt $0, $8, local_label
   bgt $0, $0, local_label
 
   bgtu $7, $8, local_label
   bgtu $7, $0, local_label
   bgtu $0, $8, local_label
   bgtu $0, $0, local_label
 
+  jal foo
+  .option pic2
+  jal foo
+  .option pic0
+
   add $4, $5, $6
 
   .set noreorder
   .set nomacro
   li  $8, -16
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
   li  $8, 16
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions

 # CHECK: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
   bgtu $7, $0, local_label
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
   bgtu $0, $8, local_label
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
   bgtu $0, $0, local_label
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
 
+  jal foo
+# CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
+  .option pic2
+  jal foo
+# CHECK: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions
+  .option pic0
+
   add $4, $5, $6
 # CHECK-NOT: [[@LINE-1]]:3: warning: macro instruction expanded into multiple instructions