[x86] Model the direction flag (DF) separately from the rest of EFLAGS.

[chandlerc]

Summary

This cleans up a number of operations that only claimed te use EFLAGS
due to using DF. But no instructions which we think of us setting EFLAGS
actually modify DF (other than things like popf) and so this needlessly
creates uses of EFLAGS that aren't really there.

In fact, DF is so restrictive it is pretty easy to model. Only STD, CLD,
and the whole-flags writes (WRFLAGS and POPF) need to model this.

Sadly, this runs into issues because some part of the x86 backend
I can't find is marking uses of EFLAGS prior to a def as 'undef' very
helpfully. I need that to happen for DF as well. So far, no joy. If
anyone has suggestions, that'd be awesome and then we could land this
patch.

[efriedma]

I need that to happen for DF as well.

You do...? DF isn't actually undefined at any point; the x86 ABI says DF is clear on entry to/exit from a function.

[chandlerc]

In D45154#1054633, @efriedma wrote:

I need that to happen for DF as well.

You do...? DF isn't actually undefined at any point; the x86 ABI says DF is clear on entry to/exit from a function.

Sorry for being imprecise...

I need *some* magic for making the LLVM "use of undefined physreg" complaints go away. It could be marking it as defined on entry. It could be something else.

[craig.topper]

llvm/lib/Target/X86/X86InstrInfo.td

2113

Don't these also use EFLAGS to pass through the other bits?

llvm/lib/Target/X86/X86InstrSystem.td

703

This isn't a VMX instruction. It was introduced in the 286. Virtualization has to be aware of it to properly virtualize it.

[chandlerc]

llvm/lib/Target/X86/X86InstrSystem.td

703

Sorry, I should have finished reading that sentence in the Intel manual rather than just seeing VMX and running with it. =/ Tried to provide more useful comment.

[chandlerc]

Update fixing the remaining issues and rebasing on top of the EFLAGS patch.

With this, all my test failures are gone and I think this patch is actually ready-to-go.

[chandlerc]

Rebase now that the core flags change has landed.

[chandlerc]

Fix a remaining place where we need to mark EFLAGS as used as we pass through
some of the flags.

Also teach clang-format to be happy w/ some of the macro code in the disassembler.

[craig.topper]

llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp

272

Is this even needed? Was someone trying to protect against a trailing comma in the macro expansion? Which we now have outside...

llvm/lib/Target/X86/X86RegisterInfo.td

523

I feel like CCR is a bad name here since that should mean "condition code register" and DF is not that.

[chandlerc]

llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp

272

I assumed this was trying to create a null terminator.

I thought we made trailing commas a no-op for these things so that stuff like this was OK?

llvm/lib/Target/X86/X86RegisterInfo.td

523

Any better naming ideas?

[craig.topper]

llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp

272

Yeah trailing commas are fine for this reason, except for in enums pre-C++11. I've personally fixed a number of tablegen emitters to not pad arrays with a 0 purely to prevent trailing commas even though it should never have been an issue. So I can believe that this is there for that reason. Maybe remove it and check with an assert against array_lengthof.

Should that array be declared "static const"? Does that make much difference to the code we output?

[chandlerc]

Clean up disassembler array a bit.

[chandlerc]

llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp

272

Hey, it works!

Removed all the cruft, directly expanded the macro in the curly braces.

I switched to static constexpr. Technically speaking with const it remains ambiguous whether the initializer will be run dynamically (which would be unfortunate, once-init barrier and everything). But with constexpr we get what we really want: an actual constant blob of data with nothing else going on.

[craig.topper]

I don't have a better name for DFCCR. So I'm fine leaving it as is.

LGTM

llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp

 
 /// translateRegister - Translates an internal register to the appropriate LLVM
 ///   register, and appends it as an operand to an MCInst.
 ///
 /// @param mcInst     - The MCInst to append to.
 /// @param reg        - The Reg to append.
 static void translateRegister(MCInst &mcInst, Reg reg) {
 #define ENTRY(x) X86::x,
-  uint8_t llvmRegnums[] = {
-    ALL_REGS
-    0
-  };
+  static constexpr MCPhysReg llvmRegnums[] = {ALL_REGS};
 #undef ENTRY
 
-  uint8_t llvmRegnum = llvmRegnums[reg];
+  MCPhysReg llvmRegnum = llvmRegnums[reg];
   mcInst.addOperand(MCOperand::createReg(llvmRegnum));
 }
 
 /// tryAddingSymbolicOperand - trys to add a symbolic operand in place of the
 /// immediate Value in the MCInst.
 ///
 /// @param Value      - The immediate Value, has had any PC adjustment made by
 ///                     the caller.
 /// @param isBranch   - If the instruction is a branch instruction

llvm/lib/Target/X86/X86ISelLowering.cpp

case X86::CMOV_V64I1:
     return EmitLoweredSelect(MI, BB);
 
   case X86::RDFLAGS32:
   case X86::RDFLAGS64: {
     unsigned PushF =
         MI.getOpcode() == X86::RDFLAGS32 ? X86::PUSHF32 : X86::PUSHF64;
     unsigned Pop = MI.getOpcode() == X86::RDFLAGS32 ? X86::POP32r : X86::POP64r;
     MachineInstr *Push = BuildMI(*BB, MI, DL, TII->get(PushF));
-    // Permit reads of the FLAGS register without it being defined.
+    // Permit reads of the EFLAGS and DF registers without them being defined.
     // This intrinsic exists to read external processor state in flags, such as
     // the trap flag, interrupt flag, and direction flag, none of which are
     // modeled by the backend.
+    assert(Push->getOperand(2).getReg() == X86::EFLAGS &&
+           "Unexpected register in operand!");
     Push->getOperand(2).setIsUndef();
+    assert(Push->getOperand(3).getReg() == X86::DF &&
+           "Unexpected register in operand!");
+    Push->getOperand(3).setIsUndef();
     BuildMI(*BB, MI, DL, TII->get(Pop), MI.getOperand(0).getReg());
 
     MI.eraseFromParent(); // The pseudo is gone now.
     return BB;
   }
 
   case X86::WRFLAGS32:
   case X86::WRFLAGS64: {

llvm/lib/Target/X86/X86InstrCompiler.td

 // ELF TLS Support
 // All calls clobber the non-callee saved registers. ESP is marked as
 // a use to prevent stack-pointer assignments that appear immediately
 // before calls from potentially appearing dead.
 let Defs = [EAX, ECX, EDX, FP0, FP1, FP2, FP3, FP4, FP5, FP6, FP7,
             ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
             MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
             XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
-            XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
+            XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS, DF],
     usesCustomInserter = 1, Uses = [ESP, SSP] in {
 def TLS_addr32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
                   "# TLS_addr32",
                   [(X86tlsaddr tls32addr:$sym)]>,
                   Requires<[Not64BitMode]>;
 def TLS_base_addr32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
                   "# TLS_base_addr32",
                   [(X86tlsbaseaddr tls32baseaddr:$sym)]>,
                   Requires<[Not64BitMode]>;
 }
 
 // All calls clobber the non-callee saved registers. RSP is marked as
 // a use to prevent stack-pointer assignments that appear immediately
 // before calls from potentially appearing dead.
 let Defs = [RAX, RCX, RDX, RSI, RDI, R8, R9, R10, R11,
             FP0, FP1, FP2, FP3, FP4, FP5, FP6, FP7,
             ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7,
             MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
             XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
-            XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
+            XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS, DF],
     usesCustomInserter = 1, Uses = [RSP, SSP] in {
 def TLS_addr64 : I<0, Pseudo, (outs), (ins i64mem:$sym),
                    "# TLS_addr64",
                   [(X86tlsaddr tls64addr:$sym)]>,
                   Requires<[In64BitMode]>;
 def TLS_base_addr64 : I<0, Pseudo, (outs), (ins i64mem:$sym),
                    "# TLS_base_addr64",
                   [(X86tlsbaseaddr tls64baseaddr:$sym)]>,
                   Requires<[In64BitMode]>;
 }
 
 // Darwin TLS Support
 // For i386, the address of the thunk is passed on the stack, on return the
 // address of the variable is in %eax.  %ecx is trashed during the function
 // call.  All other registers are preserved.
-let Defs = [EAX, ECX, EFLAGS],
+let Defs = [EAX, ECX, EFLAGS, DF],
     Uses = [ESP, SSP],
     usesCustomInserter = 1 in
 def TLSCall_32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
                 "# TLSCall_32",
                 [(X86TLSCall addr:$sym)]>,
                 Requires<[Not64BitMode]>;
 
 // For x86_64, the address of the thunk is passed in %rdi, but the
 // pseudo directly use the symbol, so do not add an implicit use of
 // %rdi. The lowering will do the right thing with RDI.
 // On return the address of the variable is in %rax.  All other
 // registers are preserved.
-let Defs = [RAX, EFLAGS],
+let Defs = [RAX, EFLAGS, DF],
     Uses = [RSP, SSP],
     usesCustomInserter = 1 in
 def TLSCall_64 : I<0, Pseudo, (outs), (ins i64mem:$sym),
                   "# TLSCall_64",
                   [(X86TLSCall addr:$sym)]>,
                   Requires<[In64BitMode]>;
 } // SchedRW
 

llvm/lib/Target/X86/X86InstrInfo.cpp

reverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
   Cond[0].setImm(GetOppositeBranchCondition(CC));
   return false;
 }
 
 bool X86InstrInfo::
 isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const {
   // FIXME: Return false for x87 stack register classes for now. We can't
   // allow any loads of these registers before FpGet_ST0_80.
-  return !(RC == &X86::CCRRegClass || RC == &X86::RFP32RegClass ||
-           RC == &X86::RFP64RegClass || RC == &X86::RFP80RegClass);
+  return !(RC == &X86::CCRRegClass || RC == &X86::DFCCRRegClass ||
+           RC == &X86::RFP32RegClass || RC == &X86::RFP64RegClass ||
+           RC == &X86::RFP80RegClass);
 }
 
 /// Return a virtual register initialized with the
 /// the global base register value. Output instructions required to
 /// initialize the register in the function entry block, if necessary.
 ///
 /// TODO: Eliminate this and move the code to X86MachineFunctionInfo.
 ///

llvm/lib/Target/X86/X86InstrInfo.td

let mayLoad = 1, mayStore = 1, usesCustomInserter = 1,
   let Uses = [RSP] in
   def RDFLAGS64 : PseudoI<(outs GR64:$dst), (ins),
                    [(set GR64:$dst, (int_x86_flags_read_u64))]>,
                 Requires<[In64BitMode]>;
 }
 
 let mayLoad = 1, mayStore = 1, usesCustomInserter = 1,
     SchedRW = [WriteRMW] in {
-  let Defs = [ESP, EFLAGS], Uses = [ESP] in
+  let Defs = [ESP, EFLAGS, DF], Uses = [ESP] in
   def WRFLAGS32 : PseudoI<(outs), (ins GR32:$src),
                    [(int_x86_flags_write_u32 GR32:$src)]>,
                 Requires<[Not64BitMode]>;
 
-  let Defs = [RSP, EFLAGS], Uses = [RSP] in
+  let Defs = [RSP, EFLAGS, DF], Uses = [RSP] in
   def WRFLAGS64 : PseudoI<(outs), (ins GR64:$src),
                    [(int_x86_flags_write_u64 GR64:$src)]>,
                 Requires<[In64BitMode]>;
 }
 
-let Defs = [ESP, EFLAGS], Uses = [ESP], mayLoad = 1, hasSideEffects=0,
+let Defs = [ESP, EFLAGS, DF], Uses = [ESP], mayLoad = 1, hasSideEffects=0,
     SchedRW = [WriteLoad] in {
 def POPF16   : I<0x9D, RawFrm, (outs), (ins), "popf{w}", [], IIC_POP_F>,
                 OpSize16;
 def POPF32   : I<0x9D, RawFrm, (outs), (ins), "popf{l|d}", [], IIC_POP_FD>,
                 OpSize32, Requires<[Not64BitMode]>;
 }
 
-let Defs = [ESP], Uses = [ESP, EFLAGS], mayStore = 1, hasSideEffects=0,
+let Defs = [ESP], Uses = [ESP, EFLAGS, DF], mayStore = 1, hasSideEffects=0,
     SchedRW = [WriteStore] in {
 def PUSHF16  : I<0x9C, RawFrm, (outs), (ins), "pushf{w}", [], IIC_PUSH_F>,
                  OpSize16;
 def PUSHF32  : I<0x9C, RawFrm, (outs), (ins), "pushf{l|d}", [], IIC_PUSH_F>,
                OpSize32, Requires<[Not64BitMode]>;
 }
 
 let Defs = [RSP], Uses = [RSP], hasSideEffects=0 in {

 def PUSH64i8   : Ii8<0x6a, RawFrm, (outs), (ins i64i8imm:$imm),
                     "push{q}\t$imm", [], IIC_PUSH_IMM>, OpSize32,
                     Requires<[In64BitMode]>;
 def PUSH64i32  : Ii32S<0x68, RawFrm, (outs), (ins i64i32imm:$imm),
                     "push{q}\t$imm", [], IIC_PUSH_IMM>, OpSize32,
                     Requires<[In64BitMode]>;
 }
 
-let Defs = [RSP, EFLAGS], Uses = [RSP], mayLoad = 1, hasSideEffects=0 in
+let Defs = [RSP, EFLAGS, DF], Uses = [RSP], mayLoad = 1, hasSideEffects=0 in
 def POPF64   : I<0x9D, RawFrm, (outs), (ins), "popfq", [], IIC_POP_FD>,
                OpSize32, Requires<[In64BitMode]>, Sched<[WriteLoad]>;
-let Defs = [RSP], Uses = [RSP, EFLAGS], mayStore = 1, hasSideEffects=0 in
+let Defs = [RSP], Uses = [RSP, EFLAGS, DF], mayStore = 1, hasSideEffects=0 in
 def PUSHF64    : I<0x9C, RawFrm, (outs), (ins), "pushfq", [], IIC_PUSH_F>,
                  OpSize32, Requires<[In64BitMode]>, Sched<[WriteStore]>;
 
 let Defs = [EDI, ESI, EBP, EBX, EDX, ECX, EAX, ESP], Uses = [ESP],
     mayLoad = 1, hasSideEffects = 0, SchedRW = [WriteLoad] in {
 def POPA32   : I<0x61, RawFrm, (outs), (ins), "popal", [], IIC_POP_A>,
                OpSize32, Requires<[Not64BitMode]>;
 def POPA16   : I<0x61, RawFrm, (outs), (ins), "popaw", [], IIC_POP_A>,
def BSR64rr  : RI<0xBD, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                   IIC_BIT_SCAN_REG>, PS, Sched<[WriteBitScan]>;
 def BSR64rm  : RI<0xBD, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                   "bsr{q}\t{$src, $dst|$dst, $src}",
                   [(set GR64:$dst, EFLAGS, (X86bsr (loadi64 addr:$src)))],
                   IIC_BIT_SCAN_MEM>, PS, Sched<[WriteBitScanLd]>;
 } // Defs = [EFLAGS]
 
 let SchedRW = [WriteMicrocoded] in {
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [EDI,ESI], Uses = [EDI,ESI,EFLAGS] in {
+let Defs = [EDI,ESI], Uses = [EDI,ESI,DF] in {
 def MOVSB : I<0xA4, RawFrmDstSrc, (outs), (ins dstidx8:$dst, srcidx8:$src),
               "movsb\t{$src, $dst|$dst, $src}", [], IIC_MOVS>;
 def MOVSW : I<0xA5, RawFrmDstSrc, (outs), (ins dstidx16:$dst, srcidx16:$src),
               "movsw\t{$src, $dst|$dst, $src}", [], IIC_MOVS>, OpSize16;
 def MOVSL : I<0xA5, RawFrmDstSrc, (outs), (ins dstidx32:$dst, srcidx32:$src),
               "movs{l|d}\t{$src, $dst|$dst, $src}", [], IIC_MOVS>, OpSize32;
 def MOVSQ : RI<0xA5, RawFrmDstSrc, (outs), (ins dstidx64:$dst, srcidx64:$src),
                "movsq\t{$src, $dst|$dst, $src}", [], IIC_MOVS>,
                Requires<[In64BitMode]>;
 }
 
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [EDI], Uses = [AL,EDI,EFLAGS] in
+let Defs = [EDI], Uses = [AL,EDI,DF] in
 def STOSB : I<0xAA, RawFrmDst, (outs), (ins dstidx8:$dst),
               "stosb\t{%al, $dst|$dst, al}", [], IIC_STOS>;
-let Defs = [EDI], Uses = [AX,EDI,EFLAGS] in
+let Defs = [EDI], Uses = [AX,EDI,DF] in
 def STOSW : I<0xAB, RawFrmDst, (outs), (ins dstidx16:$dst),
               "stosw\t{%ax, $dst|$dst, ax}", [], IIC_STOS>, OpSize16;
-let Defs = [EDI], Uses = [EAX,EDI,EFLAGS] in
+let Defs = [EDI], Uses = [EAX,EDI,DF] in
 def STOSL : I<0xAB, RawFrmDst, (outs), (ins dstidx32:$dst),
               "stos{l|d}\t{%eax, $dst|$dst, eax}", [], IIC_STOS>, OpSize32;
-let Defs = [RDI], Uses = [RAX,RDI,EFLAGS] in
+let Defs = [RDI], Uses = [RAX,RDI,DF] in
 def STOSQ : RI<0xAB, RawFrmDst, (outs), (ins dstidx64:$dst),
                "stosq\t{%rax, $dst|$dst, rax}", [], IIC_STOS>,
                Requires<[In64BitMode]>;
 
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [EDI,EFLAGS], Uses = [AL,EDI,EFLAGS] in
+let Defs = [EDI,EFLAGS], Uses = [AL,EDI,DF] in
 def SCASB : I<0xAE, RawFrmDst, (outs), (ins dstidx8:$dst),
               "scasb\t{$dst, %al|al, $dst}", [], IIC_SCAS>;
-let Defs = [EDI,EFLAGS], Uses = [AX,EDI,EFLAGS] in
+let Defs = [EDI,EFLAGS], Uses = [AX,EDI,DF] in
 def SCASW : I<0xAF, RawFrmDst, (outs), (ins dstidx16:$dst),
               "scasw\t{$dst, %ax|ax, $dst}", [], IIC_SCAS>, OpSize16;
-let Defs = [EDI,EFLAGS], Uses = [EAX,EDI,EFLAGS] in
+let Defs = [EDI,EFLAGS], Uses = [EAX,EDI,DF] in
 def SCASL : I<0xAF, RawFrmDst, (outs), (ins dstidx32:$dst),
               "scas{l|d}\t{$dst, %eax|eax, $dst}", [], IIC_SCAS>, OpSize32;
-let Defs = [EDI,EFLAGS], Uses = [RAX,EDI,EFLAGS] in
+let Defs = [EDI,EFLAGS], Uses = [RAX,EDI,DF] in
 def SCASQ : RI<0xAF, RawFrmDst, (outs), (ins dstidx64:$dst),
                "scasq\t{$dst, %rax|rax, $dst}", [], IIC_SCAS>,
                Requires<[In64BitMode]>;
 
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [EDI,ESI,EFLAGS], Uses = [EDI,ESI,EFLAGS] in {
+let Defs = [EDI,ESI,EFLAGS], Uses = [EDI,ESI,DF] in {
 def CMPSB : I<0xA6, RawFrmDstSrc, (outs), (ins dstidx8:$dst, srcidx8:$src),
               "cmpsb\t{$dst, $src|$src, $dst}", [], IIC_CMPS>;
 def CMPSW : I<0xA7, RawFrmDstSrc, (outs), (ins dstidx16:$dst, srcidx16:$src),
               "cmpsw\t{$dst, $src|$src, $dst}", [], IIC_CMPS>, OpSize16;
 def CMPSL : I<0xA7, RawFrmDstSrc, (outs), (ins dstidx32:$dst, srcidx32:$src),
               "cmps{l|d}\t{$dst, $src|$src, $dst}", [], IIC_CMPS>, OpSize32;
 def CMPSQ : RI<0xA7, RawFrmDstSrc, (outs), (ins dstidx64:$dst, srcidx64:$src),
                "cmpsq\t{$dst, $src|$src, $dst}", [], IIC_CMPS>,
def DATA16_PREFIX : I<0x66, RawFrm, (outs),  (ins), "data16", [], IIC_NOP>,
                      Requires<[Not16BitMode]>;
 
 // Data instruction prefix
 def DATA32_PREFIX : I<0x66, RawFrm, (outs),  (ins), "data32", [], IIC_NOP>,
                      Requires<[In16BitMode]>;
 } // SchedRW
 
 // Repeat string operation instruction prefixes
-// These use the DF flag in the EFLAGS register to inc or dec ECX
-let Defs = [ECX], Uses = [ECX,EFLAGS], SchedRW = [WriteMicrocoded] in {
+let Defs = [ECX], Uses = [ECX,DF], SchedRW = [WriteMicrocoded] in {
 // Repeat (used with INS, OUTS, MOVS, LODS and STOS)
 def REP_PREFIX : I<0xF3, RawFrm, (outs),  (ins), "rep", []>;
 // Repeat while not equal (used with CMPS and SCAS)
 def REPNE_PREFIX : I<0xF2, RawFrm, (outs),  (ins), "repne", []>;
 }
 
 // String manipulation instructions
 let SchedRW = [WriteMicrocoded] in {
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [AL,ESI], Uses = [ESI,EFLAGS] in
+let Defs = [AL,ESI], Uses = [ESI,DF] in
 def LODSB : I<0xAC, RawFrmSrc, (outs), (ins srcidx8:$src),
               "lodsb\t{$src, %al|al, $src}", [], IIC_LODS>;
-let Defs = [AX,ESI], Uses = [ESI,EFLAGS] in
+let Defs = [AX,ESI], Uses = [ESI,DF] in
 def LODSW : I<0xAD, RawFrmSrc, (outs), (ins srcidx16:$src),
               "lodsw\t{$src, %ax|ax, $src}", [], IIC_LODS>, OpSize16;
-let Defs = [EAX,ESI], Uses = [ESI,EFLAGS] in
+let Defs = [EAX,ESI], Uses = [ESI,DF] in
 def LODSL : I<0xAD, RawFrmSrc, (outs), (ins srcidx32:$src),
               "lods{l|d}\t{$src, %eax|eax, $src}", [], IIC_LODS>, OpSize32;
-let Defs = [RAX,ESI], Uses = [ESI,EFLAGS] in
+let Defs = [RAX,ESI], Uses = [ESI,DF] in
 def LODSQ : RI<0xAD, RawFrmSrc, (outs), (ins srcidx64:$src),
                "lodsq\t{$src, %rax|rax, $src}", [], IIC_LODS>,
                Requires<[In64BitMode]>;
 }
 
 let SchedRW = [WriteSystem] in {
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [ESI], Uses = [DX,ESI,EFLAGS] in {
+let Defs = [ESI], Uses = [DX,ESI,DF] in {
 def OUTSB : I<0x6E, RawFrmSrc, (outs), (ins srcidx8:$src),
              "outsb\t{$src, %dx|dx, $src}", [], IIC_OUTS>;
 def OUTSW : I<0x6F, RawFrmSrc, (outs), (ins srcidx16:$src),
               "outsw\t{$src, %dx|dx, $src}", [], IIC_OUTS>, OpSize16;
 def OUTSL : I<0x6F, RawFrmSrc, (outs), (ins srcidx32:$src),
               "outs{l|d}\t{$src, %dx|dx, $src}", [], IIC_OUTS>, OpSize32;
 }
 
-// These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
-let Defs = [EDI], Uses = [DX,EDI,EFLAGS] in {
+let Defs = [EDI], Uses = [DX,EDI,DF] in {
 def INSB : I<0x6C, RawFrmDst, (outs), (ins dstidx8:$dst),
              "insb\t{%dx, $dst|$dst, dx}", [], IIC_INS>;
 def INSW : I<0x6D, RawFrmDst, (outs), (ins dstidx16:$dst),
              "insw\t{%dx, $dst|$dst, dx}", [], IIC_INS>,  OpSize16;
 def INSL : I<0x6D, RawFrmDst, (outs), (ins dstidx32:$dst),
              "ins{l|d}\t{%dx, $dst|$dst, dx}", [], IIC_INS>, OpSize32;
 }
 }
 
-// Flag instructions
-let SchedRW = [WriteALU] in {
+// EFLAGS management instructions.
+let SchedRW = [WriteALU], Defs = [EFLAGS], Uses = [EFLAGS] in {
 def CLC : I<0xF8, RawFrm, (outs), (ins), "clc", [], IIC_CLC_CMC_STC>;
 def STC : I<0xF9, RawFrm, (outs), (ins), "stc", [], IIC_CLC_CMC_STC>;
-def CLI : I<0xFA, RawFrm, (outs), (ins), "cli", [], IIC_CLI>;
-def STI : I<0xFB, RawFrm, (outs), (ins), "sti", [], IIC_STI>;
-def CLD : I<0xFC, RawFrm, (outs), (ins), "cld", [], IIC_CLD>;
-def STD : I<0xFD, RawFrm, (outs), (ins), "std", [], IIC_STD>;
 def CMC : I<0xF5, RawFrm, (outs), (ins), "cmc", [], IIC_CLC_CMC_STC>;
+}
 
-def CLTS : I<0x06, RawFrm, (outs), (ins), "clts", [], IIC_CLTS>, TB;
+// DF management instructions.
+// FIXME: These are a bit more expensive than CLC and STC. We should consider
+// adjusting their schedule bucket.
+let SchedRW = [WriteALU], Defs = [DF] in {
+def CLD : I<0xFC, RawFrm, (outs), (ins), "cld", [], IIC_CLD>;
+def STD : I<0xFD, RawFrm, (outs), (ins), "std", [], IIC_STD>;
 }
 
+
 // Table lookup instructions
 let Uses = [AL,EBX], Defs = [AL], hasSideEffects = 0, mayLoad = 1 in
 def XLAT : I<0xD7, RawFrm, (outs), (ins), "xlatb", [], IIC_XLAT>,
            Sched<[WriteLoad]>;
 
 let SchedRW = [WriteMicrocoded] in {
 // ASCII Adjust After Addition
 let Uses = [AL,EFLAGS], Defs = [AX,EFLAGS], hasSideEffects = 0 in

llvm/lib/Target/X86/X86InstrSystem.td

 // SMX Instruction
 let SchedRW = [WriteSystem] in {
 let Uses = [RAX, RBX, RCX, RDX], Defs = [RAX, RBX, RCX] in {
   def GETSEC : I<0x37, RawFrm, (outs), (ins), "getsec", [], IIC_SMX>, TB;
 } // Uses, Defs
 } // SchedRW
 
 //===----------------------------------------------------------------------===//
+// TS flag control instruction.
+let SchedRW = [WriteSystem] in {
+def CLTS : I<0x06, RawFrm, (outs), (ins), "clts", [], IIC_CLTS>, TB;
+}
+
+//===----------------------------------------------------------------------===//
+// IF (inside EFLAGS) management instructions.
+let SchedRW = [WriteSystem], Uses = [EFLAGS], Defs = [EFLAGS] in {
+def CLI : I<0xFA, RawFrm, (outs), (ins), "cli", [], IIC_CLI>;
+def STI : I<0xFB, RawFrm, (outs), (ins), "sti", [], IIC_STI>;
+}
+
+//===----------------------------------------------------------------------===//
 // RDPID Instruction
 let SchedRW = [WriteSystem] in {
 def RDPID32 : I<0xC7, MRM7r, (outs GR32:$dst), (ins),
               "rdpid\t$dst", [(set GR32:$dst, (int_x86_rdpid))], IIC_RDPID>, XS,
               Requires<[Not64BitMode, HasRDPID]>;
 def RDPID64 : I<0xC7, MRM7r, (outs GR64:$dst), (ins),
               "rdpid\t$dst", [], IIC_RDPID>, XS,
               Requires<[In64BitMode, HasRDPID]>;

llvm/lib/Target/X86/X86RegisterInfo.td

 def ST4 : X86Reg<"st(4)", 4>, DwarfRegNum<[37, 16, 15]>;
 def ST5 : X86Reg<"st(5)", 5>, DwarfRegNum<[38, 17, 16]>;
 def ST6 : X86Reg<"st(6)", 6>, DwarfRegNum<[39, 18, 17]>;
 def ST7 : X86Reg<"st(7)", 7>, DwarfRegNum<[40, 19, 18]>;
 
 // Floating-point status word
 def FPSW : X86Reg<"fpsw", 0>;
 
-// Status flags register
+// Status flags register.
+//
+// Note that some flags that are commonly thought of as part of the status
+// flags register are modeled separately. Typically this is due to instructions
+// reading and updating those flags independently of all the others. We don't
+// want to create false dependencies between these instructions and so we use
+// a separate register to model them.
 def EFLAGS : X86Reg<"flags", 0>;
 
+// The direction flag.
+def DF : X86Reg<"DF", 0>;
+
+
 // Segment registers
 def CS : X86Reg<"cs", 1>;
 def DS : X86Reg<"ds", 3>;
 def SS : X86Reg<"ss", 2>;
 def ES : X86Reg<"es", 0>;
 def FS : X86Reg<"fs", 4>;
 def GS : X86Reg<"gs", 5>;
 

 def CCR : RegisterClass<"X86", [i32], 32, (add EFLAGS)> {
   let CopyCost = -1;  // Don't allow copying of status registers.
   let isAllocatable = 0;
 }
 def FPCCR : RegisterClass<"X86", [i16], 16, (add FPSW)> {
   let CopyCost = -1;  // Don't allow copying of status registers.
   let isAllocatable = 0;
 }
+def DFCCR : RegisterClass<"X86", [i32], 32, (add DF)> {
+  let CopyCost = -1;  // Don't allow copying of status registers.
+  let isAllocatable = 0;
+}
 
 // AVX-512 vector/mask registers.
 def VR512 : RegisterClass<"X86", [v16f32, v8f64, v64i8, v32i16, v16i32, v8i64],
                           512, (sequence "ZMM%u", 0, 31)>;
 
 // Scalar AVX-512 floating point registers.
 def FR32X : RegisterClass<"X86", [f32], 32, (sequence "XMM%u", 0, 31)>;
 

llvm/test/CodeGen/X86/ipra-reg-usage.ll

 ; RUN: llc -enable-ipra -print-regusage -o /dev/null 2>&1 < %s | FileCheck %s
 
 target triple = "x86_64-unknown-unknown"
 declare void @bar1()
 define preserve_allcc void @foo()#0 {
-; CHECK: foo Clobbered Registers: $cs $ds $eflags $eip $eiz $es $fpsw $fs $gs $hip $ip $rip $riz $ss $ssp $bnd0 $bnd1 $bnd2 $bnd3 $cr0 $cr1 $cr2 $cr3 $cr4 $cr5 $cr6 $cr7 $cr8 $cr9 $cr10 $cr11 $cr12 $cr13 $cr14 $cr15 $dr0 $dr1 $dr2 $dr3 $dr4 $dr5 $dr6 $dr7 $dr8 $dr9 $dr10 $dr11 $dr12 $dr13 $dr14 $dr15 $fp0 $fp1 $fp2 $fp3 $fp4 $fp5 $fp6 $fp7 $k0 $k1 $k2 $k3 $k4 $k5 $k6 $k7 $mm0 $mm1 $mm2 $mm3 $mm4 $mm5 $mm6 $mm7 $r11 $st0 $st1 $st2 $st3 $st4 $st5 $st6 $st7 $xmm16 $xmm17 $xmm18 $xmm19 $xmm20 $xmm21 $xmm22 $xmm23 $xmm24 $xmm25 $xmm26 $xmm27 $xmm28 $xmm29 $xmm30 $xmm31 $ymm0 $ymm1 $ymm2 $ymm3 $ymm4 $ymm5 $ymm6 $ymm7 $ymm8 $ymm9 $ymm10 $ymm11 $ymm12 $ymm13 $ymm14 $ymm15 $ymm16 $ymm17 $ymm18 $ymm19 $ymm20 $ymm21 $ymm22 $ymm23 $ymm24 $ymm25 $ymm26 $ymm27 $ymm28 $ymm29 $ymm30 $ymm31 $zmm0 $zmm1 $zmm2 $zmm3 $zmm4 $zmm5 $zmm6 $zmm7 $zmm8 $zmm9 $zmm10 $zmm11 $zmm12 $zmm13 $zmm14 $zmm15 $zmm16 $zmm17 $zmm18 $zmm19 $zmm20 $zmm21 $zmm22 $zmm23 $zmm24 $zmm25 $zmm26 $zmm27 $zmm28 $zmm29 $zmm30 $zmm31 $r11b $r11d $r11w
+; CHECK: foo Clobbered Registers: $cs $df $ds $eflags $eip $eiz $es $fpsw $fs $gs $hip $ip $rip $riz $ss $ssp $bnd0 $bnd1 $bnd2 $bnd3 $cr0 $cr1 $cr2 $cr3 $cr4 $cr5 $cr6 $cr7 $cr8 $cr9 $cr10 $cr11 $cr12 $cr13 $cr14 $cr15 $dr0 $dr1 $dr2 $dr3 $dr4 $dr5 $dr6 $dr7 $dr8 $dr9 $dr10 $dr11 $dr12 $dr13 $dr14 $dr15 $fp0 $fp1 $fp2 $fp3 $fp4 $fp5 $fp6 $fp7 $k0 $k1 $k2 $k3 $k4 $k5 $k6 $k7 $mm0 $mm1 $mm2 $mm3 $mm4 $mm5 $mm6 $mm7 $r11 $st0 $st1 $st2 $st3 $st4 $st5 $st6 $st7 $xmm16 $xmm17 $xmm18 $xmm19 $xmm20 $xmm21 $xmm22 $xmm23 $xmm24 $xmm25 $xmm26 $xmm27 $xmm28 $xmm29 $xmm30 $xmm31 $ymm0 $ymm1 $ymm2 $ymm3 $ymm4 $ymm5 $ymm6 $ymm7 $ymm8 $ymm9 $ymm10 $ymm11 $ymm12 $ymm13 $ymm14 $ymm15 $ymm16 $ymm17 $ymm18 $ymm19 $ymm20 $ymm21 $ymm22 $ymm23 $ymm24 $ymm25 $ymm26 $ymm27 $ymm28 $ymm29 $ymm30 $ymm31 $zmm0 $zmm1 $zmm2 $zmm3 $zmm4 $zmm5 $zmm6 $zmm7 $zmm8 $zmm9 $zmm10 $zmm11 $zmm12 $zmm13 $zmm14 $zmm15 $zmm16 $zmm17 $zmm18 $zmm19 $zmm20 $zmm21 $zmm22 $zmm23 $zmm24 $zmm25 $zmm26 $zmm27 $zmm28 $zmm29 $zmm30 $zmm31 $r11b $r11d $r11w
   call void @bar1()
   call void @bar2()
   ret void
 }
 declare void @bar2()
 attributes #0 = {nounwind}