[ARM] [Assembler] Extend immediate conversions for A32, T32 and T16

[sanwou01]

Summary

Immediate converions change an instruction that cannot be encoded to an
equivalent but different instruction. For example, "ADD r0, r1, #FFFFFFFF"
cannot be encoded as an ADD instruction. However, "SUB r0, r1, #1" is
equivalent.

These conversions are different from instruction aliases. An alias maps
several assembler instructions onto one encoding. An immediate
conversion, however, maps an *invalid* instruction--e.g. with an
immediate that cannot be represented in the encoding--to a different
(but equivalent) instruction.

Several instruction immediates were being converted already, but this
was not systematically tested, nor did it cover the full set of possible
conversions.

This patch implements all possible substitutions for ARM, Thumb1 and
Thumb2 assembler and adds tests. It also adds a feature flag
(-mattr=+no-imm-conversions) to turn these substitutions off. This is
helpful for users who want their code to assemble to exactly what they
wrote.

[javed.absar]

Hi Sanne:
Thanks for the patch. Please find some comments inline.
Best Regards
Javed

lib/Target/ARM/ARM.td

264 ↗

(On Diff #90466)

If the substitution is only for instructions with immediates, would a narrower name e.g. FeatureNoImplicitImmSubstitution be more appropriate?

lib/Target/ARM/ARMInstrThumb.td

1244 ↗

(On Diff #90466)

should this be imm3 instead?

test/MC/ARM/implicit-substitutions.s

116 ↗

(On Diff #90466)

You may consider shortening this test by reducing 'CHECK-DISABLED: error: instruction requires ...' specially if there are duplication. Related question - is the feature flag just for testing?

[sanwou01]

Update in response to reviewer comments.

[sanwou01]

Hi Javed,

Thanks for the review. I have updated the naming to use "immediate conversions" throughout. I think this is a better name, thanks for pointing it out.

See further comments inline.
Thanks,
Sanne

lib/Target/ARM/ARMInstrThumb.td

1244 ↗

(On Diff #90466)

I don't think so? $imm is just a name here: the allowed range that is is controlled by the mod_imm1_7_neg below.

test/MC/ARM/implicit-substitutions.s

116 ↗

(On Diff #90466)

The repetition of "error: instruction requires: ImmediateConversions" is necessary to test that -mattr=+no-imm-conversions properly disables all conversions.

No, the feature flag is not just for testing. The flag is helpful for users who want their code to assemble to exactly what they wrote. I have added this justification to the commit message.

[rengolin]

Hi Sanne,

We already had some ADD/SUB substitutions in the assembler, so if this is a more general approach, it should be encouraged, but we need to make sure we remove the old hack in favour of this patch.

Related patches:
http://llvm.org/viewvc/llvm-project?view=revision&revision=212722
http://llvm.org/viewvc/llvm-project?view=revision&revision=241166

cheers,
--renato

[sanwou01]

Hi Renato,

Thank you for the references. I hadn't looked at Aarch64 yet, but looking through those patches quickly, it looks like all conversions for negative immediates are already supported. It looks like it is using the same aliasing mechanism that I'm using. Changing Aarch64 to use InstSubst instead of InstAlias should make sure that -mattr=+no-imm-conversions works for Aarch64 too.

That said, maybe NegativeImmediates/no-neg-immediates is an even better name for this feature!

Thanks,
Sanne

[rengolin]

Right, I just compared this patch to the two I linked earlier and it seems this patch is a lot cleaner than the ones we had before for AArch64.

If @javed.absar is happy with it, so am I.

cheers,
--renato

[javed.absar]

Thanks Sanne and Renato.
LGTM to me as well.

llvm/trunk/lib/Target/ARM/ARM.td

 def FeatureReserveR9 : SubtargetFeature<"reserve-r9", "ReserveR9", "true",
                                         "Reserve R9, making it unavailable as "
                                         "GPR">;
 
 def FeatureNoMovt : SubtargetFeature<"no-movt", "NoMovt", "true",
                                      "Don't use movt/movw pairs for 32-bit "
                                      "imms">;
 
+def FeatureNoNegativeImmediates : SubtargetFeature<"no-neg-immediates",
+                                        "NegativeImmediates", "false",
+                                        "Convert immediates and instructions "
+                                        "to their negated or complemented "
+                                        "equivalent when the immediate does "
+                                        "not fit in the encoding.">;
 
 //===----------------------------------------------------------------------===//
 // ARM ISAa.
 //
 
 def HasV4TOps   : SubtargetFeature<"v4t", "HasV4TOps", "true",
                                    "Support ARM v4T instructions">;
 def HasV5TOps   : SubtargetFeature<"v5t", "HasV5TOps", "true",

llvm/trunk/lib/Target/ARM/ARMInstrFormats.td

 
 
 //===----------------------------------------------------------------------===//
 // ARM Assembler alias templates.
 //
 // Note: When EmitPriority == 1, the alias will be used for printing
 class ARMInstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[IsARM]>;
+class ARMInstSubst<string Asm, dag Result, bit EmitPriority = 0>
+      : InstAlias<Asm, Result, EmitPriority>,
+        Requires<[IsARM,UseNegativeImmediates]>;
 class  tInstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[IsThumb]>;
+class  tInstSubst<string Asm, dag Result, bit EmitPriority = 0>
+      : InstAlias<Asm, Result, EmitPriority>,
+        Requires<[IsThumb,UseNegativeImmediates]>;
 class t2InstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[IsThumb2]>;
+class t2InstSubst<string Asm, dag Result, bit EmitPriority = 0>
+      : InstAlias<Asm, Result, EmitPriority>,
+        Requires<[IsThumb2,UseNegativeImmediates]>;
 class VFP2InstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[HasVFP2]>;
 class VFP2DPInstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[HasVFP2,HasDPVFP]>;
 class VFP3InstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[HasVFP3]>;
 class NEONInstAlias<string Asm, dag Result, bit EmitPriority = 0>
       : InstAlias<Asm, Result, EmitPriority>, Requires<[HasNEON]>;

llvm/trunk/lib/Target/ARM/ARMInstrInfo.td

 def IsNotMachO       : Predicate<"!Subtarget->isTargetMachO()">;
 def IsNaCl           : Predicate<"Subtarget->isTargetNaCl()">;
 def IsWindows        : Predicate<"Subtarget->isTargetWindows()">;
 def IsNotWindows     : Predicate<"!Subtarget->isTargetWindows()">;
 def UseNaClTrap      : Predicate<"Subtarget->useNaClTrap()">,
                                  AssemblerPredicate<"FeatureNaClTrap", "NaCl">;
 def DontUseNaClTrap  : Predicate<"!Subtarget->useNaClTrap()">;
 
+def UseNegativeImmediates :
+  Predicate<"false">,
+            AssemblerPredicate<"!FeatureNoNegativeImmediates",
+                               "NegativeImmediates">;
+
 // FIXME: Eventually this will be just "hasV6T2Ops".
 def UseMovt          : Predicate<"Subtarget->useMovt(*MF)">;
 def DontUseMovt      : Predicate<"!Subtarget->useMovt(*MF)">;
 def UseFPVMLx        : Predicate<"Subtarget->useFPVMLx()">;
 def UseMulOps        : Predicate<"Subtarget->useMulOps()">;
 
 // Prefer fused MAC for fp mul + add over fp VMLA / VMLS if they are available.
 // But only select them if more precision in FP computation is allowed.

 def : MnemonicAlias<"uqaddsubx", "uqasx">;
 // UQSAX == UQSUBADDX
 def : MnemonicAlias<"uqsubaddx", "uqsax">;
 // USAX == USUBADDX
 def : MnemonicAlias<"usubaddx", "usax">;
 
 // "mov Rd, mod_imm_not" can be handled via "mvn" in assembly, just like
 // for isel.
-def : ARMInstAlias<"mov${s}${p} $Rd, $imm",
+def : ARMInstSubst<"mov${s}${p} $Rd, $imm",
                    (MVNi rGPR:$Rd, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
-def : ARMInstAlias<"mvn${s}${p} $Rd, $imm",
+def : ARMInstSubst<"mvn${s}${p} $Rd, $imm",
                    (MOVi rGPR:$Rd, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
 // Same for AND <--> BIC
-def : ARMInstAlias<"bic${s}${p} $Rd, $Rn, $imm",
+def : ARMInstSubst<"bic${s}${p} $Rd, $Rn, $imm",
                    (ANDri GPR:$Rd, GPR:$Rn, mod_imm_not:$imm,
                           pred:$p, cc_out:$s)>;
-def : ARMInstAlias<"bic${s}${p} $Rdn, $imm",
+def : ARMInstSubst<"bic${s}${p} $Rdn, $imm",
                    (ANDri GPR:$Rdn, GPR:$Rdn, mod_imm_not:$imm,
                           pred:$p, cc_out:$s)>;
-def : ARMInstAlias<"and${s}${p} $Rd, $Rn, $imm",
+def : ARMInstSubst<"and${s}${p} $Rd, $Rn, $imm",
                    (BICri GPR:$Rd, GPR:$Rn, mod_imm_not:$imm,
                           pred:$p, cc_out:$s)>;
-def : ARMInstAlias<"and${s}${p} $Rdn, $imm",
+def : ARMInstSubst<"and${s}${p} $Rdn, $imm",
                    (BICri GPR:$Rdn, GPR:$Rdn, mod_imm_not:$imm,
                           pred:$p, cc_out:$s)>;
 
 // Likewise, "add Rd, mod_imm_neg" -> sub
-def : ARMInstAlias<"add${s}${p} $Rd, $Rn, $imm",
+def : ARMInstSubst<"add${s}${p} $Rd, $Rn, $imm",
                  (SUBri GPR:$Rd, GPR:$Rn, mod_imm_neg:$imm, pred:$p, cc_out:$s)>;
-def : ARMInstAlias<"add${s}${p} $Rd, $imm",
+def : ARMInstSubst<"add${s}${p} $Rd, $imm",
                  (SUBri GPR:$Rd, GPR:$Rd, mod_imm_neg:$imm, pred:$p, cc_out:$s)>;
+// Likewise, "sub Rd, mod_imm_neg" -> add
+def : ARMInstSubst<"sub${s}${p} $Rd, $Rn, $imm",
+                 (ADDri GPR:$Rd, GPR:$Rn, mod_imm_neg:$imm, pred:$p, cc_out:$s)>;
+def : ARMInstSubst<"sub${s}${p} $Rd, $imm",
+                 (ADDri GPR:$Rd, GPR:$Rd, mod_imm_neg:$imm, pred:$p, cc_out:$s)>;
+
+
+def : ARMInstSubst<"adc${s}${p} $Rd, $Rn, $imm",
+                 (SBCri GPR:$Rd, GPR:$Rn, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
+def : ARMInstSubst<"adc${s}${p} $Rdn, $imm",
+                 (SBCri GPR:$Rdn, GPR:$Rdn, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
+def : ARMInstSubst<"sbc${s}${p} $Rd, $Rn, $imm",
+                 (ADCri GPR:$Rd, GPR:$Rn, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
+def : ARMInstSubst<"sbc${s}${p} $Rdn, $imm",
+                 (ADCri GPR:$Rdn, GPR:$Rdn, mod_imm_not:$imm, pred:$p, cc_out:$s)>;
+
 // Same for CMP <--> CMN via mod_imm_neg
-def : ARMInstAlias<"cmp${p} $Rd, $imm",
+def : ARMInstSubst<"cmp${p} $Rd, $imm",
                    (CMNri rGPR:$Rd, mod_imm_neg:$imm, pred:$p)>;
-def : ARMInstAlias<"cmn${p} $Rd, $imm",
+def : ARMInstSubst<"cmn${p} $Rd, $imm",
                    (CMPri rGPR:$Rd, mod_imm_neg:$imm, pred:$p)>;
 
 // The shifter forms of the MOV instruction are aliased to the ASR, LSL,
 // LSR, ROR, and RRX instructions.
 // FIXME: We need C++ parser hooks to map the alias to the MOV
 //        encoding. It seems we should be able to do that sort of thing
 //        in tblgen, but it could get ugly.
 let TwoOperandAliasConstraint = "$Rm = $Rd" in {

llvm/trunk/lib/Target/ARM/ARMInstrThumb.td

def imm_sr : Operand<i32>, PatLeaf<(imm), [{
   let PrintMethod = "printThumbSRImm";
   let ParserMatchClass = ThumbSRImmAsmOperand;
 }
 
 def imm0_7_neg : PatLeaf<(i32 imm), [{
   return (uint32_t)-N->getZExtValue() < 8;
 }], imm_neg_XFORM>;
 
+def ThumbModImmNeg1_7AsmOperand : AsmOperandClass { let Name = "ThumbModImmNeg1_7"; }
+def mod_imm1_7_neg : Operand<i32>, PatLeaf<(imm), [{
+    unsigned Value = -(unsigned)N->getZExtValue();
+    return 0 < Value && Value < 8;
+  }], imm_neg_XFORM> {
+  let ParserMatchClass = ThumbModImmNeg1_7AsmOperand;
+}
+
+def ThumbModImmNeg8_255AsmOperand : AsmOperandClass { let Name = "ThumbModImmNeg8_255"; }
+def mod_imm8_255_neg : Operand<i32>, PatLeaf<(imm), [{
+    unsigned Value = -(unsigned)N->getZExtValue();
+    return 7 < Value && Value < 256;
+  }], imm_neg_XFORM> {
+  let ParserMatchClass = ThumbModImmNeg8_255AsmOperand;
+}
+
+
 def imm0_255_comp : PatLeaf<(i32 imm), [{
   return ~((uint32_t)N->getZExtValue()) < 256;
 }]>;
 
 def imm8_255 : ImmLeaf<i32, [{
   return Imm >= 8 && Imm < 256;
 }]>;
 def imm8_255_neg : PatLeaf<(i32 imm), [{
def tSUBspi : T1pIt<(outs GPRsp:$Rdn), (ins GPRsp:$Rn, t_imm0_508s4:$imm),
                     IIC_iALUi, "sub", "\t$Rdn, $imm", []>,
               T1Misc<{0,0,0,0,1,?,?}>, Sched<[WriteALU]> {
   // A6.2.5 & A8.6.214
   bits<7> imm;
   let Inst{6-0} = imm;
   let DecoderMethod = "DecodeThumbAddSPImm";
 }
 
-def : tInstAlias<"add${p} sp, $imm",
+def : tInstSubst<"add${p} sp, $imm",
                  (tSUBspi SP, t_imm0_508s4_neg:$imm, pred:$p)>;
-def : tInstAlias<"add${p} sp, sp, $imm",
+def : tInstSubst<"add${p} sp, sp, $imm",
                  (tSUBspi SP, t_imm0_508s4_neg:$imm, pred:$p)>;
 
 // Can optionally specify SP as a three operand instruction.
 def : tInstAlias<"add${p} sp, sp, $imm",
                  (tADDspi SP, t_imm0_508s4:$imm, pred:$p)>;
 def : tInstAlias<"sub${p} sp, sp, $imm",
                  (tSUBspi SP, t_imm0_508s4:$imm, pred:$p)>;
 
def tADDhirr : T1pIt<(outs GPR:$Rdn), (ins GPR:$Rn, GPR:$Rm), IIC_iALUr,
     bits<4> Rdn;
     bits<4> Rm;
     let Inst{7}   = Rdn{3};
     let Inst{6-3} = Rm;
     let Inst{2-0} = Rdn{2-0};
   }
 }
 
+def : tInstSubst<"sub${s}${p} $rd, $rn, $imm",
+                 (tADDi3 tGPR:$rd, s_cc_out:$s, tGPR:$rn, mod_imm1_7_neg:$imm, pred:$p)>;
+def : tInstSubst<"sub${s}${p} $rdn, $imm",
+                 (tADDi8 tGPR:$rdn, s_cc_out:$s, mod_imm8_255_neg:$imm, pred:$p)>;
+
+
 // AND register
 let isCommutable = 1 in
 def tAND :                      // A8.6.12
   T1sItDPEncode<0b0000, (outs tGPR:$Rdn), (ins tGPR:$Rn, tGPR:$Rm),
                 IIC_iBITr,
                 "and", "\t$Rdn, $Rm",
                 [(set tGPR:$Rdn, (and tGPR:$Rn, tGPR:$Rm))]>, Sched<[WriteALU]>;
 

 
 def tSUBi8 :                    // A8.6.210 T2
   T1sItGenEncodeImm<{1,1,1,?,?}, (outs tGPR:$Rdn),
                     (ins tGPR:$Rn, imm0_255:$imm8), IIC_iALUi,
                     "sub", "\t$Rdn, $imm8",
                     [(set tGPR:$Rdn, (add tGPR:$Rn, imm8_255_neg:$imm8))]>,
                     Sched<[WriteALU]>;
 
+def : tInstSubst<"add${s}${p} $rd, $rn, $imm",
+                 (tSUBi3 tGPR:$rd, s_cc_out:$s, tGPR:$rn, mod_imm1_7_neg:$imm, pred:$p)>;
+
+
+def : tInstSubst<"add${s}${p} $rdn, $imm",
+                 (tSUBi8 tGPR:$rdn, s_cc_out:$s, mod_imm8_255_neg:$imm, pred:$p)>;
+
+
 // Subtract register
 def tSUBrr :                    // A8.6.212
   T1sIGenEncode<0b01101, (outs tGPR:$Rd), (ins tGPR:$Rn, tGPR:$Rm),
                 IIC_iALUr,
                 "sub", "\t$Rd, $Rn, $Rm",
                 [(set tGPR:$Rd, (sub tGPR:$Rn, tGPR:$Rm))]>,
                 Sched<[WriteALU]>;
 

llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td

 defm t2ADDS : T2I_bin_s_irs <IIC_iALUi, IIC_iALUr, IIC_iALUsi, ARMaddc, 1>;
 defm t2SUBS : T2I_bin_s_irs <IIC_iALUi, IIC_iALUr, IIC_iALUsi, ARMsubc>;
 
 let hasPostISelHook = 1 in {
 defm t2ADC  : T2I_adde_sube_irs<0b1010, "adc", ARMadde, 1>;
 defm t2SBC  : T2I_adde_sube_irs<0b1011, "sbc", ARMsube>;
 }
 
+def : t2InstSubst<"adc${s}${p} $rd, $rn, $imm",
+                 (t2SBCri rGPR:$rd, rGPR:$rn, t2_so_imm_not:$imm, pred:$p, s_cc_out:$s)>;
+def : t2InstSubst<"sbc${s}${p} $rd, $rn, $imm",
+                 (t2ADCri rGPR:$rd, rGPR:$rn, t2_so_imm_not:$imm, pred:$p, s_cc_out:$s)>;
+
+def : t2InstSubst<"add${s}${p}.w $rd, $rn, $imm",
+                 (t2SUBri GPRnopc:$rd, GPRnopc:$rn, t2_so_imm_neg:$imm, pred:$p, s_cc_out:$s)>;
+def : t2InstSubst<"addw${p} $rd, $rn, $imm",
+                 (t2SUBri12 GPRnopc:$rd, GPR:$rn, t2_so_imm_neg:$imm, pred:$p)>;
+def : t2InstSubst<"sub${s}${p}.w $rd, $rn, $imm",
+                 (t2ADDri GPRnopc:$rd, GPRnopc:$rn, t2_so_imm_neg:$imm, pred:$p, s_cc_out:$s)>;
+def : t2InstSubst<"subw${p} $rd, $rn, $imm",
+                 (t2ADDri12 GPRnopc:$rd, GPR:$rn, t2_so_imm_neg:$imm, pred:$p)>;
 // RSB
 defm t2RSB  : T2I_rbin_irs  <0b1110, "rsb", sub>;
 
 // FIXME: Eliminate them if we can write def : Pat patterns which defines
 // CPSR and the implicit def of CPSR is not needed.
 defm t2RSBS : T2I_rbin_s_is <ARMsubc>;
 
 // (sub X, imm) gets canonicalized to (add X, -imm).  Match this form.
def : t2InstAlias<"add${p} $Rdn, $imm",
            (t2ADDri12 GPRnopc:$Rdn, GPRnopc:$Rdn, imm0_4095:$imm, pred:$p)>;
 def : t2InstAlias<"add${s}${p} $Rdn, $Rm",
             (t2ADDrr GPRnopc:$Rdn, GPRnopc:$Rdn, rGPR:$Rm, pred:$p, cc_out:$s)>;
 def : t2InstAlias<"add${s}${p} $Rdn, $ShiftedRm",
                   (t2ADDrs GPRnopc:$Rdn, GPRnopc:$Rdn, t2_so_reg:$ShiftedRm,
                            pred:$p, cc_out:$s)>;
 
 // add w/ negative immediates is just a sub.
-def : t2InstAlias<"add${s}${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"add${s}${p} $Rd, $Rn, $imm",
         (t2SUBri GPRnopc:$Rd, GPRnopc:$Rn, t2_so_imm_neg:$imm, pred:$p,
                  cc_out:$s)>;
-def : t2InstAlias<"add${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"add${p} $Rd, $Rn, $imm",
            (t2SUBri12 GPRnopc:$Rd, GPR:$Rn, imm0_4095_neg:$imm, pred:$p)>;
-def : t2InstAlias<"add${s}${p} $Rdn, $imm",
+def : t2InstSubst<"add${s}${p} $Rdn, $imm",
       (t2SUBri GPRnopc:$Rdn, GPRnopc:$Rdn, t2_so_imm_neg:$imm, pred:$p,
                cc_out:$s)>;
-def : t2InstAlias<"add${p} $Rdn, $imm",
+def : t2InstSubst<"add${p} $Rdn, $imm",
            (t2SUBri12 GPRnopc:$Rdn, GPRnopc:$Rdn, imm0_4095_neg:$imm, pred:$p)>;
 
-def : t2InstAlias<"add${s}${p}.w $Rd, $Rn, $imm",
+def : t2InstSubst<"add${s}${p}.w $Rd, $Rn, $imm",
         (t2SUBri GPRnopc:$Rd, GPRnopc:$Rn, t2_so_imm_neg:$imm, pred:$p,
                  cc_out:$s)>;
-def : t2InstAlias<"addw${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"addw${p} $Rd, $Rn, $imm",
            (t2SUBri12 GPRnopc:$Rd, GPR:$Rn, imm0_4095_neg:$imm, pred:$p)>;
-def : t2InstAlias<"add${s}${p}.w $Rdn, $imm",
+def : t2InstSubst<"add${s}${p}.w $Rdn, $imm",
       (t2SUBri GPRnopc:$Rdn, GPRnopc:$Rdn, t2_so_imm_neg:$imm, pred:$p,
                cc_out:$s)>;
-def : t2InstAlias<"addw${p} $Rdn, $imm",
+def : t2InstSubst<"addw${p} $Rdn, $imm",
            (t2SUBri12 GPRnopc:$Rdn, GPRnopc:$Rdn, imm0_4095_neg:$imm, pred:$p)>;
 
 
 // Aliases for SUB without the ".w" optional width specifier.
 def : t2InstAlias<"sub${s}${p} $Rd, $Rn, $imm",
         (t2SUBri GPRnopc:$Rd, GPRnopc:$Rn, t2_so_imm:$imm, pred:$p, cc_out:$s)>;
 def : t2InstAlias<"sub${p} $Rd, $Rn, $imm",
            (t2SUBri12 GPRnopc:$Rd, GPR:$Rn, imm0_4095:$imm, pred:$p)>;
def : InstAlias<"sxtb16${p} $Rd, $Rm$rot",
                 (t2SXTB16 rGPR:$Rd, rGPR:$Rm, rot_imm:$rot, pred:$p), 0>,
                 Requires<[HasDSP, IsThumb2]>;
 def : t2InstAlias<"sxth${p} $Rd, $Rm$rot",
                   (t2SXTH rGPR:$Rd, rGPR:$Rm, rot_imm:$rot, pred:$p)>;
 
 
 // "mov Rd, t2_so_imm_not" can be handled via "mvn" in assembly, just like
 // for isel.
-def : t2InstAlias<"mov${p} $Rd, $imm",
+def : t2InstSubst<"mov${p} $Rd, $imm",
                   (t2MVNi rGPR:$Rd, t2_so_imm_not:$imm, pred:$p, zero_reg)>;
-def : t2InstAlias<"mvn${p} $Rd, $imm",
-                  (t2MOVi rGPR:$Rd, t2_so_imm_not:$imm, pred:$p, zero_reg)>;
+def : t2InstSubst<"mvn${s}${p} $Rd, $imm",
+                  (t2MOVi rGPR:$Rd, t2_so_imm_not:$imm, pred:$p, s_cc_out:$s)>;
 // Same for AND <--> BIC
-def : t2InstAlias<"bic${s}${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"bic${s}${p} $Rd, $Rn, $imm",
                   (t2ANDri rGPR:$Rd, rGPR:$Rn, t2_so_imm_not:$imm,
                            pred:$p, cc_out:$s)>;
-def : t2InstAlias<"bic${s}${p} $Rdn, $imm",
+def : t2InstSubst<"bic${s}${p} $Rdn, $imm",
                   (t2ANDri rGPR:$Rdn, rGPR:$Rdn, t2_so_imm_not:$imm,
                            pred:$p, cc_out:$s)>;
-def : t2InstAlias<"and${s}${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"and${s}${p} $Rd, $Rn, $imm",
                   (t2BICri rGPR:$Rd, rGPR:$Rn, t2_so_imm_not:$imm,
                            pred:$p, cc_out:$s)>;
-def : t2InstAlias<"and${s}${p} $Rdn, $imm",
+def : t2InstSubst<"and${s}${p} $Rdn, $imm",
                   (t2BICri rGPR:$Rdn, rGPR:$Rdn, t2_so_imm_not:$imm,
                            pred:$p, cc_out:$s)>;
 // Likewise, "add Rd, t2_so_imm_neg" -> sub
-def : t2InstAlias<"add${s}${p} $Rd, $Rn, $imm",
+def : t2InstSubst<"add${s}${p} $Rd, $Rn, $imm",
                   (t2SUBri GPRnopc:$Rd, GPRnopc:$Rn, t2_so_imm_neg:$imm,
                            pred:$p, cc_out:$s)>;
-def : t2InstAlias<"add${s}${p} $Rd, $imm",
+def : t2InstSubst<"add${s}${p} $Rd, $imm",
                   (t2SUBri GPRnopc:$Rd, GPRnopc:$Rd, t2_so_imm_neg:$imm,
                            pred:$p, cc_out:$s)>;
 // Same for CMP <--> CMN via t2_so_imm_neg
-def : t2InstAlias<"cmp${p} $Rd, $imm",
+def : t2InstSubst<"cmp${p} $Rd, $imm",
                   (t2CMNri rGPR:$Rd, t2_so_imm_neg:$imm, pred:$p)>;
-def : t2InstAlias<"cmn${p} $Rd, $imm",
+def : t2InstSubst<"cmn${p} $Rd, $imm",
                   (t2CMPri rGPR:$Rd, t2_so_imm_neg:$imm, pred:$p)>;
 
 
 // Wide 'mul' encoding can be specified with only two operands.
 def : t2InstAlias<"mul${p} $Rn, $Rm",
                   (t2MUL rGPR:$Rn, rGPR:$Rm, rGPR:$Rn, pred:$p)>;
 
 // "neg" is and alias for "rsb rd, rn, #0"

llvm/trunk/lib/Target/ARM/ARMSubtarget.h

protected:
   bool GenExecuteOnly = false;
 
   /// Target machine allowed unsafe FP math (such as use of NEON fp)
   bool UnsafeFPMath = false;
 
   /// UseSjLjEH - If true, the target uses SjLj exception handling (e.g. iOS).
   bool UseSjLjEH = false;
 
+  /// Implicitly convert an instruction to a different one if its immediates
+  /// cannot be encoded. For example, ADD r0, r1, #FFFFFFFF -> SUB r0, r1, #1.
+  bool NegativeImmediates = true;
+
   /// stackAlignment - The minimum alignment known to hold of the stack frame on
   /// entry to the function and which must be maintained by every function.
   unsigned stackAlignment = 4;
 
   /// CPUString - String name of used CPU.
   std::string CPUString;
 
   unsigned MaxInterleaveFactor = 1;

llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp

public:
   bool isModImmNeg() const {
     if (!isImm()) return false;
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
     if (!CE) return false;
     int64_t Value = CE->getValue();
     return ARM_AM::getSOImmVal(Value) == -1 &&
       ARM_AM::getSOImmVal(-Value) != -1;
   }
+  bool isThumbModImmNeg1_7() const {
+    if (!isImm()) return false;
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+    int32_t Value = -(int32_t)CE->getValue();
+    return 0 < Value && Value < 8;
+  }
+  bool isThumbModImmNeg8_255() const {
+    if (!isImm()) return false;
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+    int32_t Value = -(int32_t)CE->getValue();
+    return 7 < Value && Value < 256;
+  }
   bool isConstantPoolImm() const { return Kind == k_ConstantPoolImmediate; }
   bool isBitfield() const { return Kind == k_BitfieldDescriptor; }
   bool isPostIdxRegShifted() const { return Kind == k_PostIndexRegister; }
   bool isPostIdxReg() const {
     return Kind == k_PostIndexRegister && PostIdxReg.ShiftTy ==ARM_AM::no_shift;
   }
   bool isMemNoOffset(bool alignOK = false, unsigned Alignment = 0) const {
     if (!isMem())

 
   void addModImmNegOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
     uint32_t Enc = ARM_AM::getSOImmVal(-CE->getValue());
     Inst.addOperand(MCOperand::createImm(Enc));
   }
 
+  void addThumbModImmNeg8_255Operands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    uint32_t Val = -CE->getValue();
+    Inst.addOperand(MCOperand::createImm(Val));
+  }
+
+  void addThumbModImmNeg1_7Operands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    uint32_t Val = -CE->getValue();
+    Inst.addOperand(MCOperand::createImm(Val));
+  }
+
   void addBitfieldOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     // Munge the lsb/width into a bitfield mask.
     unsigned lsb = Bitfield.LSB;
     unsigned width = Bitfield.Width;
     // Make a 32-bit mask w/ the referenced bits clear and all other bits set.
     uint32_t Mask = ~(((uint32_t)0xffffffff >> lsb) << (32 - width) >>
                       (32 - (lsb + width)));
void addPKHASRImmOperands(MCInst &Inst, unsigned N) const {
     Inst.addOperand(MCOperand::createImm(Val == 32 ? 0 : Val));
   }
 
   void addT2SOImmNotOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     // The operand is actually a t2_so_imm, but we have its bitwise
     // negation in the assembly source, so twiddle it here.
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
-    Inst.addOperand(MCOperand::createImm(~CE->getValue()));
+    Inst.addOperand(MCOperand::createImm(~(uint32_t)CE->getValue()));
   }
 
   void addT2SOImmNegOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     // The operand is actually a t2_so_imm, but we have its
     // negation in the assembly source, so twiddle it here.
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
-    Inst.addOperand(MCOperand::createImm(-CE->getValue()));
+    Inst.addOperand(MCOperand::createImm(-(uint32_t)CE->getValue()));
   }
 
   void addImm0_4095NegOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     // The operand is actually an imm0_4095, but we have its
     // negation in the assembly source, so twiddle it here.
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
     Inst.addOperand(MCOperand::createImm(-CE->getValue()));

llvm/trunk/test/MC/ARM/negative-immediates-fail.s

+# RUN: not llvm-mc -triple armv7 %s 2>&1| FileCheck %s
+
+.arm
+
+ADC r0, r1, #0xFFFFFEEE
+# CHECK: error: invalid operand for instruction
+ADC r0, r1, #0xABFEABFF
+# CHECK: error: invalid operand for instruction
+ADC r0, r1, #0xFFFFFE02
+# CHECK: error: invalid operand for instruction
+
+ADD.W r0, r0, #0xFF01FF01
+# CHECK: error: invalid operand for instruction

llvm/trunk/test/MC/ARM/negative-immediates-thumb1-fail.s

+# RUN: not llvm-mc -triple thumbv7 -mcpu=cortex-m0 %s 2>&1 | FileCheck %s
+
+.thumb
+
+ADDs r1, r0, #0xFFFFFFF5
+# CHECK: error: instruction requires: arm-mode
+
+ADDs r0, #0xFFFFFEFF
+# CHECK: error: invalid operand for instruction
+
+SUBs r1, r0, #0xFFFFFFF5
+# CHECK: error: instruction requires: arm-mode
+
+SUBs r0, #0xFFFFFEFF
+# CHECK: error: invalid operand for instruction

llvm/trunk/test/MC/ARM/negative-immediates-thumb1.s

+# RUN: llvm-mc -triple thumbv7 -mcpu=cortex-m0 %s -show-encoding | FileCheck %s
+# RUN: not llvm-mc -triple thumbv7 -mcpu=cortex-m0 %s -show-encoding -mattr=+no-neg-immediates 2>&1 | FileCheck %s -check-prefix=CHECK-DISABLED
+
+.thumb
+
+  ADDs r1, r0, #0xFFFFFFF9
+# CHECK: subs r1, r0, #7
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+  ADDs r0, #0xFFFFFF01
+# CHECK: subs r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+
+  SUBs r0, #0xFFFFFF01
+# CHECK: adds r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+
+  SUBs r1, r0, #0xFFFFFFF9
+# CHECK: adds r1, r0, #7
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates

llvm/trunk/test/MC/ARM/negative-immediates.s

+# RUN: llvm-mc -triple armv7 %s -show-encoding | FileCheck %s
+# RUN: not llvm-mc -triple armv7 %s -show-encoding -mattr=+no-neg-immediates 2>&1 | FileCheck %s -check-prefix=CHECK-DISABLED
+
+.arm
+
+  ADC r0, r1, #0xFFFFFF00
+# CHECK: sbc r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADC r0, r1, #0xFFFFFE03
+# CHECK: sbc r0, r1, #508
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADD r0, r1, #0xFFFFFF01
+# CHECK: sub r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADD
+  AND r0, r1, #0xFFFFFF00
+# CHECK: bic r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: AND
+  BIC r0, r1, #0xFFFFFF00
+# CHECK: and r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: BIC
+  CMP r0, #0xFFFFFF01
+# CHECK: cmn r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: CMP
+  CMN r0, #0xFFFFFF01
+# CHECK: cmp r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: CMN
+  MOV r0, #0xFFFFFF00
+# CHECK: mvn r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: MOV
+  MVN r0, #0xFFFFFF00
+# CHECK: mov r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: MVN
+  SBC r0, r1, #0xFFFFFF00
+# CHECK: adc r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: SBC
+  SUB r0, r1, #0xFFFFFF01
+# CHECK: add r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: SUB
+
+.thumb
+
+  ADC r0, r1, #0xFFFFFF00
+# CHECK: sbc r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADC r0, r1, #0xFFFF00FF
+# CHECK: sbc r0, r1, #65280
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADC r0, r1, #0xFFFEFFFE
+# CHECK: sbc r0, r1, #65537 @ encoding: [0x61,0xf1,0x01,0x10]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADC r0, r1, #0xFEFFFEFF
+# CHECK: sbc r0, r1, #16777472 @ encoding: [0x61,0xf1,0x01,0x20]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADC
+  ADD.W r0, r0, #0xFFFFFF01
+# CHECK: sub.w r0, r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADD.W
+  ADD.W r0, r0, #0xFF01FF02
+# CHECK: sub.w r0, r0, #16646398 @ encoding: [0xa0,0xf1,0xfe,0x10]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADD.W
+  ADDW r0, r1, #0xFFFFFF01
+# CHECK: subw r0, r1, #255 @ encoding: [0xa1,0xf2,0xff,0x00]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADDW
+  ADD.W r0, r1, #0xFFFFFF01
+# CHECK: sub.w r0, r1, #255 @ encoding: [0xa1,0xf1,0xff,0x00]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: ADD.W
+  AND r0, r1, #0xFFFFFF00
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: AND
+# CHECK: bic r0, r1, #255
+  AND r0, r1, #0xFEFFFEFF
+# CHECK: bic r0, r1, #16777472 @ encoding: [0x21,0xf0,0x01,0x20]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: AND
+  BIC r0, r1, #0xFFFFFF00
+# CHECK: and r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: BIC
+  BIC r0, r1, #0xFEFFFEFF
+# CHECK: and r0, r1, #16777472 @ encoding: [0x01,0xf0,0x01,0x20]
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: BIC
+  CMP r0, #0xFFFFFF01
+# CHECK: cmn.w r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: CMP
+  CMN r0, #0xFFFFFF01
+# CHECK: cmp.w r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: CMN
+  MOV r0, #0xFFFFFF00
+# CHECK: mvn r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: MOV
+  MVN r0, #0xFFFFFF00
+# CHECK: mov.w r0, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: MVN
+  SBC r0, r1, #0xFFFFFF00
+# CHECK: adc r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: SBC
+  SUBW r0, r1, #0xFFFFFF01
+# CHECK: addw r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: SUBW
+  SUB.W r0, r1, #0xFFFFFF01
+# CHECK: add.w r0, r1, #255
+# CHECK-DISABLED: error: instruction requires: NegativeImmediates
+# CHECK-DISABLED: SUB.W