[ARM] Introduce separate features for FP registers.

[SjoerdMeijer]

Summary

The MVE extension in Arm v8.1-M permits the use of some move, load and
store isntructions which access the FP registers, even if there's no
actual FP support in the processor (in particular, if you have the
integer-only version of MVE).

Therefore, we need separate subtarget features to condition those
instructions on, which are implied by both FP and MVE but are not part
of either.

[ostannard]

llvm/lib/Target/ARM/ARMInstrVFP.td

2357

This only applies to VMSR at the moment, should it also cover the other four?

llvm/test/MC/ARM/mve-vmov-lane.s

14

I don't think this instruction has been added at this point in the patch series, so this test should probably be moved to a later patch.

llvm/test/MC/Disassembler/ARM/mve-vmov-lane.txt

6

Again, I think this need to be in a later patch.

[simon_tatham]

llvm/lib/Target/ARM/ARMInstrVFP.td

2357

Errrr. That's rather difficult to say.

The idea of the FP registers existing without FP has never come up before: the combination of v8.1-M and MVE is the first case where it's possible, which is why we haven't had to add this separate feature name before.

But according to the v8.1-M ArmARM, none of those four registers exists anyway in v8.1-M, with or without FP. FPEXC and FPINST[2] are for support code, which the modern fashion is to avoid needing anyway; FPSID could make sense in principle, but in fact it's never existed in the M profile as far as I can see.

So I think the answer is that no current version of the Arm architecture gives evidence for or against your question!

[SjoerdMeijer]

I've removed the instruction in the tests that are not yet supported; this will be added to another/later patch.

[ostannard]

LGTM with one nit.

test/MC/ARM/mve-vmov-lane.s

3 ↗

(On Diff #202141)

This comment isn't true yet, we should probably move this whole file to the patch which adds the v8.1M syntax change.

test/MC/Disassembler/ARM/mve-vmov-lane.txt

4 ↗

(On Diff #202141)

Again, this isn't true yet, I think it'd be better to move the whole file to a later patch.

[SjoerdMeijer]

Many thanks! I will remove those tests before committing!

llvm/lib/Target/ARM/ARM.td

def ModeSoftFloat         : SubtargetFeature<"soft-float","UseSoftFloat",
                                              "point features.">;
 
 
 //===----------------------------------------------------------------------===//
 // ARM Subtarget features.
 //
 
 // Floating Point, HW Division and Neon Support
+
+// FP loads/stores/moves, shared between VFP and MVE (even in the integer-only
+// version).
+def FeatureFPRegs         : SubtargetFeature<"fpregs", "HasFPRegs", "true",
+                                             "Enable FP registers">;
+
+// 16-bit FP loads/stores/moves, shared between VFP (with the v8.2A FP16
+// extension) and MVE (even in the integer-only version).
+def FeatureFPRegs16       : SubtargetFeature<"fpregs16", "HasFPRegs16", "true",
+                                             "Enable 16-bit FP registers",
+                                             [FeatureFPRegs]>;
+
+def FeatureFPRegs64       : SubtargetFeature<"fpregs64", "HasFPRegs64", "true",
+                                             "Enable 64-bit FP registers",
+                                             [FeatureFPRegs]>;
+
 def FeatureFP64           : SubtargetFeature<"fp64", "HasFP64", "true",
                                              "Floating point unit supports "
-                                             "double precision">;
+                                             "double precision",
+                                             [FeatureFPRegs64]>;
 
 def FeatureD32            : SubtargetFeature<"d32", "HasD32", "true",
                                              "Extend FP to 32 double registers">;
 
 multiclass VFPver<string name, string query, string description,
                   list<SubtargetFeature> prev = [],
                   list<SubtargetFeature> otherimplies = []> {
   def _D16_SP: SubtargetFeature<
!foreach(v, prev, !cast<SubtargetFeature>(v # "_D16")) #
       otherimplies # [FeatureFP64, !cast<SubtargetFeature>(NAME # "_D16_SP")]>;
   def "": SubtargetFeature<
     name, query, "true", description,
     prev # otherimplies # [
         !cast<SubtargetFeature>(NAME # "_D16"),
         !cast<SubtargetFeature>(NAME # "_SP")]>;
 }
 
-defm FeatureVFP2: VFPver<"vfp2", "HasVFPv2", "Enable VFP2 instructions">;
+defm FeatureVFP2: VFPver<"vfp2", "HasVFPv2", "Enable VFP2 instructions",
+                         [], [FeatureFPRegs]>;
+
 defm FeatureVFP3: VFPver<"vfp3", "HasVFPv3", "Enable VFP3 instructions",
                          [FeatureVFP2]>;
 
 def FeatureNEON           : SubtargetFeature<"neon", "HasNEON", "true",
                                              "Enable NEON instructions",
                                              [FeatureVFP3]>;
 
 def FeatureFP16           : SubtargetFeature<"fp16", "HasFP16", "true",
                                              "Enable half-precision "
                                              "floating point">;
 
 defm FeatureVFP4: VFPver<"vfp4", "HasVFPv4", "Enable VFP4 instructions",
                          [FeatureVFP3], [FeatureFP16]>;
 
 defm FeatureFPARMv8: VFPver<"fp-armv8", "HasFPARMv8", "Enable ARMv8 FP",
                          [FeatureVFP4]>;
 
 def FeatureFullFP16       : SubtargetFeature<"fullfp16", "HasFullFP16", "true",
                                              "Enable full half-precision "
                                              "floating point",
-                                             [FeatureFPARMv8_D16_SP]>;
+                                             [FeatureFPARMv8_D16_SP, FeatureFPRegs16]>;
 
 def FeatureFP16FML        : SubtargetFeature<"fp16fml", "HasFP16FML", "true",
                                              "Enable full half-precision "
                                              "floating point fml instructions",
                                              [FeatureFullFP16]>;
 
 def FeatureHWDivThumb     : SubtargetFeature<"hwdiv",
                                              "HasHardwareDivideInThumb", "true",

llvm/lib/Target/ARM/ARMInstrNEON.td

def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
   let Inst{21} = lane{1};
   let Inst{6}  = lane{0};
 }
 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
                           (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
                           IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
                           [(set GPR:$R, (extractelt (v2i32 DPR:$V),
                                            imm:$lane))]>,
-                Requires<[HasVFP2, HasFastVGETLNi32]> {
+                Requires<[HasFPRegs, HasFastVGETLNi32]> {
   let Inst{21} = lane{0};
 }
 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
           (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
                            (DSubReg_i8_reg imm:$lane))),
                      (SubReg_i8_lane imm:$lane))>;
 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),

llvm/lib/Target/ARM/ARMInstrVFP.td

 //===----------------------------------------------------------------------===//
 //  Load / store Instructions.
 //
 
 let canFoldAsLoad = 1, isReMaterializable = 1 in {
 
 def VLDRD : ADI5<0b1101, 0b01, (outs DPR:$Dd), (ins addrmode5:$addr),
                  IIC_fpLoad64, "vldr", "\t$Dd, $addr",
-                 [(set DPR:$Dd, (f64 (alignedload32 addrmode5:$addr)))]>;
+                 [(set DPR:$Dd, (f64 (alignedload32 addrmode5:$addr)))]>,
+            Requires<[HasFPRegs]>;
 
 def VLDRS : ASI5<0b1101, 0b01, (outs SPR:$Sd), (ins addrmode5:$addr),
                  IIC_fpLoad32, "vldr", "\t$Sd, $addr",
-                 [(set SPR:$Sd, (alignedload32 addrmode5:$addr))]> {
+                 [(set SPR:$Sd, (alignedload32 addrmode5:$addr))]>,
+            Requires<[HasFPRegs]> {
   // Some single precision VFP instructions may be executed on both NEON and VFP
   // pipelines.
   let D = VFPNeonDomain;
 }
 
 let isUnpredicable = 1 in
 def VLDRH : AHI5<0b1101, 0b01, (outs HPR:$Sd), (ins addrmode5fp16:$addr),
                  IIC_fpLoad16, "vldr", ".16\t$Sd, $addr",
                  [(set HPR:$Sd, (alignedload16 addrmode5fp16:$addr))]>,
-            Requires<[HasFullFP16]>;
+            Requires<[HasFPRegs16]>;
 
 } // End of 'let canFoldAsLoad = 1, isReMaterializable = 1 in'
 
 def VSTRD : ADI5<0b1101, 0b00, (outs), (ins DPR:$Dd, addrmode5:$addr),
                  IIC_fpStore64, "vstr", "\t$Dd, $addr",
-                 [(alignedstore32 (f64 DPR:$Dd), addrmode5:$addr)]>;
+                 [(alignedstore32 (f64 DPR:$Dd), addrmode5:$addr)]>,
+            Requires<[HasFPRegs]>;
 
 def VSTRS : ASI5<0b1101, 0b00, (outs), (ins SPR:$Sd, addrmode5:$addr),
                  IIC_fpStore32, "vstr", "\t$Sd, $addr",
-                 [(alignedstore32 SPR:$Sd, addrmode5:$addr)]> {
+                 [(alignedstore32 SPR:$Sd, addrmode5:$addr)]>,
+            Requires<[HasFPRegs]> {
   // Some single precision VFP instructions may be executed on both NEON and VFP
   // pipelines.
   let D = VFPNeonDomain;
 }
 
 let isUnpredicable = 1 in
 def VSTRH : AHI5<0b1101, 0b00, (outs), (ins HPR:$Sd, addrmode5fp16:$addr),
                  IIC_fpStore16, "vstr", ".16\t$Sd, $addr",
                  [(alignedstore16 HPR:$Sd, addrmode5fp16:$addr)]>,
-            Requires<[HasFullFP16]>;
+            Requires<[HasFPRegs16]>;
 
 //===----------------------------------------------------------------------===//
 //  Load / store multiple Instructions.
 //
 
 multiclass vfp_ldst_mult<string asm, bit L_bit,
                          InstrItinClass itin, InstrItinClass itin_upd> {
+  let Predicates = [HasFPRegs] in {
   // Double Precision
   def DIA :
     AXDI4<(outs), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops),
           IndexModeNone, itin,
           !strconcat(asm, "ia${p}\t$Rn, $regs"), "", []> {
     let Inst{24-23} = 0b01;       // Increment After
     let Inst{21}    = 0;          // No writeback
     let Inst{20}    = L_bit;
def SDB_UPD :
     let Inst{24-23} = 0b10;       // Decrement Before
     let Inst{21}    = 1;          // Writeback
     let Inst{20}    = L_bit;
 
     // Some single precision VFP instructions may be executed on both NEON and
     // VFP pipelines.
     let D = VFPNeonDomain;
   }
   }
+}
 
 let hasSideEffects = 0 in {
 
 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in
 defm VLDM : vfp_ldst_mult<"vldm", 1, IIC_fpLoad_m, IIC_fpLoad_mu>;
 
 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in
 defm VSTM : vfp_ldst_mult<"vstm", 0, IIC_fpStore_m, IIC_fpStore_mu>;
defm : VFPDTAnyInstAlias<"vpop${p}", "$r",
                          (VLDMDIA_UPD SP, pred:$p, dpr_reglist:$r)>;
 
 // FLDMX, FSTMX - Load and store multiple unknown precision registers for
 // pre-armv6 cores.
 // These instruction are deprecated so we don't want them to get selected.
 // However, there is no UAL syntax for them, so we keep them around for
 // (dis)assembly only.
 multiclass vfp_ldstx_mult<string asm, bit L_bit> {
+  let Predicates = [HasFPRegs] in {
   // Unknown precision
   def XIA :
     AXXI4<(outs), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops),
           IndexModeNone, !strconcat(asm, "iax${p}\t$Rn, $regs"), "", []> {
     let Inst{24-23} = 0b01;       // Increment After
     let Inst{21}    = 0;          // No writeback
     let Inst{20}    = L_bit;
   }
   def XIA_UPD :
     AXXI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops),
           IndexModeUpd, !strconcat(asm, "iax${p}\t$Rn!, $regs"), "$Rn = $wb", []> {
     let Inst{24-23} = 0b01;         // Increment After
     let Inst{21}    = 1;            // Writeback
     let Inst{20}    = L_bit;
   }
   def XDB_UPD :
     AXXI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops),
           IndexModeUpd, !strconcat(asm, "dbx${p}\t$Rn!, $regs"), "$Rn = $wb", []> {
     let Inst{24-23} = 0b10;         // Decrement Before
     let Inst{21}    = 1;            // Writeback
     let Inst{20}    = L_bit;
   }
   }
+}
 
 defm FLDM : vfp_ldstx_mult<"fldm", 1>;
 defm FSTM : vfp_ldstx_mult<"fstm", 0>;
 
 def : VFP2MnemonicAlias<"fldmeax", "fldmdbx">;
 def : VFP2MnemonicAlias<"fldmfdx", "fldmiax">;
 
 def : VFP2MnemonicAlias<"fstmeax", "fstmiax">;
def VSQRTH : AHuI<0b11101, 0b11, 0b0001, 0b11, 0,
                   (outs SPR:$Sd), (ins SPR:$Sm),
                   IIC_fpSQRT16, "vsqrt", ".f16\t$Sd, $Sm",
                   []>;
 
 let hasSideEffects = 0 in {
 let isMoveReg = 1 in {
 def VMOVD  : ADuI<0b11101, 0b11, 0b0000, 0b01, 0,
                   (outs DPR:$Dd), (ins DPR:$Dm),
-                  IIC_fpUNA64, "vmov", ".f64\t$Dd, $Dm", []>;
+                  IIC_fpUNA64, "vmov", ".f64\t$Dd, $Dm", []>,
+             Requires<[HasFPRegs64]>;
 
 def VMOVS  : ASuI<0b11101, 0b11, 0b0000, 0b01, 0,
                   (outs SPR:$Sd), (ins SPR:$Sm),
-                  IIC_fpUNA32, "vmov", ".f32\t$Sd, $Sm", []>;
+                  IIC_fpUNA32, "vmov", ".f32\t$Sd, $Sm", []>,
+             Requires<[HasFPRegs]>;
 } // isMoveReg
 
 let PostEncoderMethod = "", DecoderNamespace = "VFPV8", isUnpredicable = 1 in {
 def VMOVH  : ASuInp<0b11101, 0b11, 0b0000, 0b01, 0,
                   (outs SPR:$Sd), (ins SPR:$Sm),
                   IIC_fpUNA16, "vmovx.f16\t$Sd, $Sm", []>,
              Requires<[HasFullFP16]>;
 
 def VINSH  : ASuInp<0b11101, 0b11, 0b0000, 0b11, 0,
                   (outs SPR:$Sd), (ins SPR:$Sm),
                   IIC_fpUNA16, "vins.f16\t$Sd, $Sm", []>,
              Requires<[HasFullFP16]>;
 } // PostEncoderMethod
 } // hasSideEffects
 
 //===----------------------------------------------------------------------===//
 // FP <-> GPR Copies.  Int <-> FP Conversions.
 //
 
 let isMoveReg = 1 in {
 def VMOVRS : AVConv2I<0b11100001, 0b1010,
                       (outs GPR:$Rt), (ins SPR:$Sn),
                       IIC_fpMOVSI, "vmov", "\t$Rt, $Sn",
                       [(set GPR:$Rt, (bitconvert SPR:$Sn))]>,
+             Requires<[HasFPRegs]>,
              Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<4> Rt;
   bits<5> Sn;
 
   // Encode instruction operands.
   let Inst{19-16} = Sn{4-1};
   let Inst{7}     = Sn{0};
   let Inst{15-12} = Rt;
 
   let Inst{6-5}   = 0b00;
   let Inst{3-0}   = 0b0000;
 
   // Some single precision VFP instructions may be executed on both NEON and VFP
   // pipelines.
   let D = VFPNeonDomain;
 }
 
 // Bitcast i32 -> f32.  NEON prefers to use VMOVDRR.
 def VMOVSR : AVConv4I<0b11100000, 0b1010,
                       (outs SPR:$Sn), (ins GPR:$Rt),
                       IIC_fpMOVIS, "vmov", "\t$Sn, $Rt",
                       [(set SPR:$Sn, (bitconvert GPR:$Rt))]>,
-             Requires<[HasVFP2, UseVMOVSR]>,
+             Requires<[HasFPRegs, UseVMOVSR]>,
              Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<5> Sn;
   bits<4> Rt;
 
   // Encode instruction operands.
   let Inst{19-16} = Sn{4-1};
   let Inst{7}     = Sn{0};

 } // isMoveReg
 def : Pat<(arm_vmovsr GPR:$Rt), (VMOVSR GPR:$Rt)>, Requires<[HasVFP2, UseVMOVSR]>;
 
 let hasSideEffects = 0 in {
 def VMOVRRD  : AVConv3I<0b11000101, 0b1011,
                         (outs GPR:$Rt, GPR:$Rt2), (ins DPR:$Dm),
                         IIC_fpMOVDI, "vmov", "\t$Rt, $Rt2, $Dm",
                  [(set GPR:$Rt, GPR:$Rt2, (arm_fmrrd DPR:$Dm))]>,
+               Requires<[HasFPRegs]>,
                Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<5> Dm;
   bits<4> Rt;
   bits<4> Rt2;
 
   // Encode instruction operands.
   let Inst{3-0}   = Dm{3-0};
def VMOVRRD  : AVConv3I<0b11000101, 0b1011,
   // $Rt2 = EXTRACT_SUBREG $Dm, ssub_1
   let isExtractSubreg = 1;
 }
 
 def VMOVRRS  : AVConv3I<0b11000101, 0b1010,
                       (outs GPR:$Rt, GPR:$Rt2), (ins SPR:$src1, SPR:$src2),
                  IIC_fpMOVDI, "vmov", "\t$Rt, $Rt2, $src1, $src2",
                  [/* For disassembly only; pattern left blank */]>,
+               Requires<[HasFPRegs]>,
                Sched<[WriteFPMOV]> {
   bits<5> src1;
   bits<4> Rt;
   bits<4> Rt2;
 
   // Encode instruction operands.
   let Inst{3-0}   = src1{4-1};
   let Inst{5}     = src1{0};

 
 // FMDHR: GPR -> SPR
 // FMDLR: GPR -> SPR
 
 def VMOVDRR : AVConv5I<0b11000100, 0b1011,
                       (outs DPR:$Dm), (ins GPR:$Rt, GPR:$Rt2),
                       IIC_fpMOVID, "vmov", "\t$Dm, $Rt, $Rt2",
                       [(set DPR:$Dm, (arm_fmdrr GPR:$Rt, GPR:$Rt2))]>,
+              Requires<[HasFPRegs]>,
               Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<5> Dm;
   bits<4> Rt;
   bits<4> Rt2;
 
   // Encode instruction operands.
   let Inst{3-0}   = Dm{3-0};
def : Pat<(fneg (arm_fmdrr GPR:$Rl, GPR:$Rh)),
           (VMOVDRR GPR:$Rl, (t2EORri GPR:$Rh, (i32 0x80000000)))>,
       Requires<[IsThumb2]>;
 
 let hasSideEffects = 0 in
 def VMOVSRR : AVConv5I<0b11000100, 0b1010,
                      (outs SPR:$dst1, SPR:$dst2), (ins GPR:$src1, GPR:$src2),
                 IIC_fpMOVID, "vmov", "\t$dst1, $dst2, $src1, $src2",
                 [/* For disassembly only; pattern left blank */]>,
+              Requires<[HasFPRegs]>,
               Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<5> dst1;
   bits<4> src1;
   bits<4> src2;
 
   // Encode instruction operands.
   let Inst{3-0}   = dst1{4-1};
def VMOVSRR : AVConv5I<0b11000100, 0b1010,
   let DecoderMethod = "DecodeVMOVSRR";
 }
 
 // Move H->R, clearing top 16 bits
 def VMOVRH : AVConv2I<0b11100001, 0b1001,
                       (outs GPR:$Rt), (ins HPR:$Sn),
                       IIC_fpMOVSI, "vmov", ".f16\t$Rt, $Sn",
                       [(set GPR:$Rt, (arm_vmovrh HPR:$Sn))]>,
-             Requires<[HasFullFP16]>,
+             Requires<[HasFPRegs16]>,
              Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<4> Rt;
   bits<5> Sn;
 
   // Encode instruction operands.
   let Inst{19-16} = Sn{4-1};
   let Inst{7}     = Sn{0};
   let Inst{15-12} = Rt;
 
   let Inst{6-5}   = 0b00;
   let Inst{3-0}   = 0b0000;
 
   let isUnpredicable = 1;
 }
 
 // Move R->H, clearing top 16 bits
 def VMOVHR : AVConv4I<0b11100000, 0b1001,
                       (outs HPR:$Sn), (ins GPR:$Rt),
                       IIC_fpMOVIS, "vmov", ".f16\t$Sn, $Rt",
                       [(set HPR:$Sn, (arm_vmovhr GPR:$Rt))]>,
-             Requires<[HasFullFP16]>,
+             Requires<[HasFPRegs16]>,
              Sched<[WriteFPMOV]> {
   // Instruction operands.
   bits<5> Sn;
   bits<4> Rt;
 
   // Encode instruction operands.
   let Inst{19-16} = Sn{4-1};
   let Inst{7}     = Sn{0};
class MovFromVFP<bits<4> opc19_16, dag oops, dag iops, string opc, string asm,
   let Inst{7}     = 0;
   let Inst{6-5}   = 0b00;
   let Inst{4}     = 1;
   let Inst{3-0}   = 0b0000;
 }
 
 // APSR is the application level alias of CPSR. This FPSCR N, Z, C, V flags
 // to APSR.
-let Defs = [CPSR], Uses = [FPSCR_NZCV], Rt = 0b1111 /* apsr_nzcv */ in
+let Defs = [CPSR], Uses = [FPSCR_NZCV], Predicates = [HasFPRegs],
+    Rt = 0b1111 /* apsr_nzcv */ in
 def FMSTAT : MovFromVFP<0b0001 /* fpscr */, (outs), (ins),
                         "vmrs", "\tAPSR_nzcv, fpscr", [(arm_fmstat)]>;
 
 let DecoderMethod = "DecodeForVMRSandVMSR" in {
  // Application level FPSCR -> GPR
- let hasSideEffects = 1, Uses = [FPSCR] in
+ let hasSideEffects = 1, Uses = [FPSCR], Predicates = [HasFPRegs] in
  def VMRS :  MovFromVFP<0b0001 /* fpscr */, (outs GPRnopc:$Rt), (ins),
                         "vmrs", "\t$Rt, fpscr",
                         [(set GPRnopc:$Rt, (int_arm_get_fpscr))]>;
 
  // System level FPEXC, FPSID -> GPR
  let Uses = [FPSCR] in {
    def VMRS_FPEXC : MovFromVFP<0b1000 /* fpexc */, (outs GPRnopc:$Rt), (ins),
                                "vmrs", "\t$Rt, fpexc", []>;
class MovToVFP<bits<4> opc19_16, dag oops, dag iops, string opc, string asm,
   let Inst{19-16} = opc19_16;
   let Inst{11-8}  = 0b1010;
   let Inst{7}     = 0;
   let Inst{4}     = 1;
 }
 
 let DecoderMethod = "DecodeForVMRSandVMSR" in {
  let Defs = [FPSCR] in {
+   let Predicates = [HasFPRegs] in
    // Application level GPR -> FPSCR
    def VMSR : MovToVFP<0b0001 /* fpscr */, (outs), (ins GPRnopc:$src),
                        "vmsr", "\tfpscr, $src",
                        [(int_arm_set_fpscr GPRnopc:$src)]>;
    // System level GPR -> FPEXC
    def VMSR_FPEXC : MovToVFP<0b1000 /* fpexc */, (outs), (ins GPRnopc:$src),
                                "vmsr", "\tfpexc, $src", []>;
    // System level GPR -> FPSID

 def : VFP2MnemonicAlias<"fmrx", "vmrs">;
 def : VFP2MnemonicAlias<"fmxr", "vmsr">;
 
 // Be friendly and accept the old form of zero-compare
 def : VFP2DPInstAlias<"fcmpzd${p} $val", (VCMPZD DPR:$val, pred:$p)>;
 def : VFP2InstAlias<"fcmpzs${p} $val", (VCMPZS SPR:$val, pred:$p)>;
 
 
-def : VFP2InstAlias<"fmstat${p}", (FMSTAT pred:$p)>;
+def : InstAlias<"fmstat${p}", (FMSTAT pred:$p), 0>, Requires<[HasFPRegs]>;
 def : VFP2InstAlias<"fadds${p} $Sd, $Sn, $Sm",
                     (VADDS SPR:$Sd, SPR:$Sn, SPR:$Sm, pred:$p)>;
 def : VFP2DPInstAlias<"faddd${p} $Dd, $Dn, $Dm",
                       (VADDD DPR:$Dd, DPR:$Dn, DPR:$Dm, pred:$p)>;
 def : VFP2InstAlias<"fsubs${p} $Sd, $Sn, $Sm",
                     (VSUBS SPR:$Sd, SPR:$Sn, SPR:$Sm, pred:$p)>;
 def : VFP2DPInstAlias<"fsubd${p} $Dd, $Dn, $Dm",
                       (VSUBD DPR:$Dd, DPR:$Dn, DPR:$Dm, pred:$p)>;

llvm/lib/Target/ARM/ARMPredicates.td

def HasV6M           : Predicate<"Subtarget->hasV6MOps()">,
                                  AssemblerPredicate<"HasV6MOps",
                                                     "armv6m or armv6t2">;
 def HasV8MBaseline   : Predicate<"Subtarget->hasV8MBaselineOps()">,
                                  AssemblerPredicate<"HasV8MBaselineOps",
                                                     "armv8m.base">;
 def HasV8MMainline   : Predicate<"Subtarget->hasV8MMainlineOps()">,
                                  AssemblerPredicate<"HasV8MMainlineOps",
                                                     "armv8m.main">;
+def HasFPRegs        : Predicate<"Subtarget->hasFPRegs()">,
+                                 AssemblerPredicate<"FeatureFPRegs",
+                                                    "fp registers">;
+def HasFPRegs16      : Predicate<"Subtarget->hasFPRegs16()">,
+                                 AssemblerPredicate<"FeatureFPRegs16",
+                                                    "16-bit fp registers">;
+def HasFPRegs64      : Predicate<"Subtarget->hasFPRegs64()">,
+                                 AssemblerPredicate<"FeatureFPRegs64",
+                                                    "64-bit fp registers">;
 def HasV6T2          : Predicate<"Subtarget->hasV6T2Ops()">,
                                  AssemblerPredicate<"HasV6T2Ops", "armv6t2">;
 def NoV6T2           : Predicate<"!Subtarget->hasV6T2Ops()">;
 def HasV6K           : Predicate<"Subtarget->hasV6KOps()">,
                                  AssemblerPredicate<"HasV6KOps", "armv6k">;
 def NoV6K            : Predicate<"!Subtarget->hasV6KOps()">;
 def HasV7            : Predicate<"Subtarget->hasV7Ops()">,
                                  AssemblerPredicate<"HasV7Ops", "armv7">;

llvm/lib/Target/ARM/ARMSubtarget.h

protected:
 
   /// HasVFPv2, HasVFPv3, HasVFPv4, HasFPARMv8, HasNEON - Specify what
   /// floating point ISAs are supported.
   bool HasVFPv2 = false;
   bool HasVFPv3 = false;
   bool HasVFPv4 = false;
   bool HasFPARMv8 = false;
   bool HasNEON = false;
+  bool HasFPRegs = false;
+  bool HasFPRegs16 = false;
+  bool HasFPRegs64 = false;
 
   /// Versions of the VFP flags restricted to single precision, or to
   /// 16 d-registers, or both.
   bool HasVFPv2SP = false;
   bool HasVFPv3SP = false;
   bool HasVFPv4SP = false;
   bool HasFPARMv8SP = false;
   bool HasVFPv2D16 = false;
public:
   bool hasV8Ops()   const { return HasV8Ops;  }
   bool hasV8_1aOps() const { return HasV8_1aOps; }
   bool hasV8_2aOps() const { return HasV8_2aOps; }
   bool hasV8_3aOps() const { return HasV8_3aOps; }
   bool hasV8_4aOps() const { return HasV8_4aOps; }
   bool hasV8_5aOps() const { return HasV8_5aOps; }
   bool hasV8MBaselineOps() const { return HasV8MBaselineOps; }
   bool hasV8MMainlineOps() const { return HasV8MMainlineOps; }
+  bool hasFPRegs() const { return HasFPRegs; }
+  bool hasFPRegs16() const { return HasFPRegs16; }
+  bool hasFPRegs64() const { return HasFPRegs64; }
 
   /// @{
   /// These functions are obsolete, please consider adding subtarget features
   /// or properties instead of calling them.
   bool isCortexA5() const { return ARMProcFamily == CortexA5; }
   bool isCortexA7() const { return ARMProcFamily == CortexA7; }
   bool isCortexA8() const { return ARMProcFamily == CortexA8; }
   bool isCortexA9() const { return ARMProcFamily == CortexA9; }

llvm/test/MC/ARM/fullfp16-neg.s

 @ CHECK: instruction requires: full half-float
 @ CHECK: instruction requires: full half-float
 
 
   vldr.16 s1, [pc, #6]
   vldr.16 s2, [pc, #510]
   vldr.16 s3, [pc, #-510]
   vldr.16 s4, [r4, #-18]
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
 
 
   vstr.16 s1, [pc, #6]
   vstr.16 s2, [pc, #510]
   vstr.16 s3, [pc, #-510]
   vstr.16 s4, [r4, #-18]
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers
 
   vmov.f16 s0, #1.0
 @ CHECK: instruction requires: full half-float
 
   vmov.f16 s1, r2
   vmov.f16 r3, s4
-@ CHECK: instruction requires: full half-float
-@ CHECK: instruction requires: full half-float
+@ CHECK: instruction requires: 16-bit fp registers
+@ CHECK: instruction requires: 16-bit fp registers

llvm/test/MC/ARM/mve-fp-registers.s

+// Some simple operations on S, D and Q registers (loads, stores and moves) are
+// also avaliable in MVE, even in the integer-only version. Some of these
+// instructions (operating on D or Q registers, or FP16 values) are only
+// available for certain targets.
+
+// Note that it's not always obvious which instructions are available, for
+// example several instructions operating on D registers are available for
+// single-precision only FPUs.
+
+// All of these instructions are rejected if no VFP or MVE features are
+// present.
+// RUN: not llvm-mc -triple=thumbv8.1m.main -show-encoding                                             2>%t < %s
+// RUN: FileCheck %s < %t --check-prefix=NOFP16 --check-prefix=NOFP32 --check-prefix=NOFP64
+
+// VFP and NEON implementations by default have FP32 and FP64, but not FP16.
+// The VFPv3 FP16 extension just added conversion instructions, which we don't
+// care about here.
+// RUN: not llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+vfp2                                2>%t < %s | FileCheck %s --check-prefix=CHECK                     --check-prefix=FP32 --check-prefix=FP64
+// RUN: FileCheck %s < %t --check-prefix=NOFP16
+// RUN: not llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+fp-armv8,+neon                      2>%t < %s | FileCheck %s --check-prefix=CHECK                     --check-prefix=FP32 --check-prefix=FP64
+// RUN: FileCheck %s < %t --check-prefix=NOFP16
+
+// The v8.2A FP16 extension added loads, stores and moves for FP16.
+// RUN:     llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+fp-armv8,+fullfp16                       < %s | FileCheck %s --check-prefix=CHECK --check-prefix=FP16 --check-prefix=FP32 --check-prefix=FP64
+
+// M-profile FPUs (e.g. Cortex-M4/M7/M33) do not have FP16 instructions, and
+// the FP64 instructions are optional. They are also limited to 16 D registers,
+// but we don't test that here.
+// RUN: not llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+vfp4d16sp               2>%t < %s | FileCheck %s --check-prefix=CHECK                     --check-prefix=FP32
+// RUN: FileCheck %s < %t --check-prefix=NOFP16 --check-prefix=NOFP64
+// RUN: not llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+vfp4,-d32                           2>%t < %s | FileCheck %s --check-prefix=CHECK                     --check-prefix=FP32 --check-prefix=FP64
+// RUN: FileCheck %s < %t --check-prefix=NOFP16
+
+// Integer-only MVE, which can be combined with different options for scalar
+// FPU (or lack thereof), and has all of the move and store instructions
+// regardless of the scalar FPU.
+// RUN:     llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+mve                                 2>%t < %s | FileCheck %s --check-prefix=CHECK --check-prefix=FP16 --check-prefix=FP32 --check-prefix=FP64
+// RUN:     llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+mve,+fp-armv8-sp,+fullfp16 2>%t < %s | FileCheck %s --check-prefix=CHECK --check-prefix=FP16 --check-prefix=FP32 --check-prefix=FP64
+// RUN:     llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+mve,+fp-armv8,+fullfp16             2>%t < %s | FileCheck %s --check-prefix=CHECK --check-prefix=FP16 --check-prefix=FP32 --check-prefix=FP64
+
+// Maximal v8.1M target: MVE with FP, and scalar FP with double-precision.
+// RUN:     llvm-mc -triple=thumbv8.1m.main -show-encoding -mattr=+mve.fp                                   < %s | FileCheck %s --check-prefix=CHECK --check-prefix=FP16 --check-prefix=FP32 --check-prefix=FP64
+
+vldmia  r0, {d0}
+# FP32: vldmia  r0, {d0}               @ encoding: [0x90,0xec,0x02,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vstmia  r0, {d0}
+# FP32: vstmia  r0, {d0}                @ encoding: [0x80,0xec,0x02,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vldmia  r0, {s0}
+# FP32: vldmia  r0, {s0}                @ encoding: [0x90,0xec,0x01,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vstmia  r0, {s0}
+# FP32: vstmia  r0, {s0}                @ encoding: [0x80,0xec,0x01,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+fldmdbx r0!, {d0}
+# FP32: fldmdbx r0!, {d0}               @ encoding: [0x30,0xed,0x03,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+fstmiax r0, {d0}
+# FP32: fstmiax r0, {d0}                @ encoding: [0x80,0xec,0x03,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vldr.16 s0, [r0]
+# FP16: vldr.16 s0, [r0]                @ encoding: [0x90,0xed,0x00,0x09]
+# NOFP16: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: 16-bit fp registers
+
+vldr s0, [r0]
+# FP32: vldr    s0, [r0]                @ encoding: [0x90,0xed,0x00,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vldr d0, [r0]
+# FP32: vldr    d0, [r0]                @ encoding: [0x90,0xed,0x00,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vstr.16 s0, [r0]
+# FP16: vstr.16 s0, [r0]                @ encoding: [0x80,0xed,0x00,0x09]
+# NOFP16: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: 16-bit fp registers
+
+vstr s0, [r0]
+# FP32: vstr    s0, [r0]                @ encoding: [0x80,0xed,0x00,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vstr d0, [r0]
+# FP32: vstr    d0, [r0]                @ encoding: [0x80,0xed,0x00,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov.f16 r0, s0
+# FP16: vmov.f16        r0, s0          @ encoding: [0x10,0xee,0x10,0x09]
+# NOFP16: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: 16-bit fp registers
+
+vmov.f16 s0, r0
+# FP16: vmov.f16        s0, r0          @ encoding: [0x00,0xee,0x10,0x09]
+# NOFP16: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: 16-bit fp registers
+
+vmov s0, r0
+# FP32: vmov    s0, r0                  @ encoding: [0x00,0xee,0x10,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov r0, s0
+# FP32: vmov    r0, s0                  @ encoding: [0x10,0xee,0x10,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov r0, r1, d0
+# FP32: vmov    r0, r1, d0              @ encoding: [0x51,0xec,0x10,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov d0, r0, r1
+# FP32: vmov    d0, r0, r1              @ encoding: [0x41,0xec,0x10,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov r0, r1, s0, s1
+# FP32: vmov    r0, r1, s0, s1          @ encoding: [0x51,0xec,0x10,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov s0, s1, r0, r1
+# FP32: vmov    s0, s1, r0, r1          @ encoding: [0x41,0xec,0x10,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov.f32 s0, s1
+# FP32: vmov.f32        s0, s1          @ encoding: [0xb0,0xee,0x60,0x0a]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers
+
+vmov.f64 d0, d1
+# FP64: vmov.f64        d0, d1          @ encoding: [0xb0,0xee,0x41,0x0b]
+# NOFP64: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: 64-bit fp registers
+
+vmov.32 r0, d1[0]
+# FP32: vmov.32 r0, d1[0]               @ encoding: [0x11,0xee,0x10,0x0b]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: error: instruction requires: fp registers
+
+vmrs apsr_nzcv, fpscr
+# FP32: vmrs    APSR_nzcv, fpscr        @ encoding: [0xf1,0xee,0x10,0xfa]
+# NOFP32: :[[@LINE-2]]:{{[0-9]+}}: {{note|error}}: instruction requires: fp registers

llvm/test/MC/ARM/mve-vmov-lane.s

+// RUN: not llvm-mc -triple=thumbv8m.main   -mattr=+fp-armv8 -show-encoding < %s 2>%t | FileCheck %s --check-prefix=V80M
+// RUN: FileCheck %s < %t --check-prefix=V80M-ERROR
+// RUN:     llvm-mc -triple=thumbv8.1m.main -mattr=+fp-armv8 -show-encoding < %s 2>%t
+// RUN:     llvm-mc -triple=thumbv8.1m.main -mattr=+mve      -show-encoding < %s 2>%t
+
+// v8.1M added the Q register syntax for this instruction. The v8.1M spec does
+// not list the D register syntax as valid, but we accept it as an extension to
+// make porting code from v8.0M to v8.1M easier.
+
+vmov.32 r0, d1[0]
+// V80M: vmov.32 r0, d1[0]               @ encoding: [0x11,0xee,0x10,0x0b]
+// V81M: vmov.32 r0, d1[0]               @ encoding: [0x11,0xee,0x10,0x0b]
+
+vmov.32 r0, q0[2]
+// V80M-ERROR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction requires: armv8.1m.main with FP or MVE
+// V81M: vmov.32 r0, q0[2]               @ encoding: [0x11,0xee,0x10,0x0b]

llvm/test/MC/ARM/single-precision-fp.s

-@ RUN: not llvm-mc < %s -triple thumbv8-unknown-unknown -show-encoding -mattr=-fp64,-neon 2> %t > %t2
+@ RUN: not llvm-mc < %s -triple thumbv8-unknown-unknown -show-encoding -mattr=-fp64,-fpregs64,-neon 2> %t > %t2
 @ RUN:     FileCheck %s < %t --check-prefix=CHECK-ERRORS
 @ RUN:     FileCheck %s < %t2
 
         vadd.f64 d0, d1, d2
         vsub.f64 d2, d3, d4
         vdiv.f64 d4, d5, d6
         vmul.f64 d6, d7, d8
         vnmul.f64 d8, d9, d10

 @ CHECK-ERRORS: error: instruction requires: double precision VFP
 @ CHECK-ERRORS-NEXT: vcmpe.f64 d5, #0
 @ CHECK-ERRORS: error: instruction requires: double precision VFP
 @ CHECK-ERRORS-NEXT: vcmp.f64 d6, #0
 
         @ FIXME: overlapping aliases and a probable TableGen indeterminacy mean
         @ that the actual reason can vary by platform.
         vmov.f64 d11, d10
-@ CHECK-ERRORS: instruction requires: NEON
+@ CHECK-ERRORS: instruction requires: 64-bit fp registers
 @ CHECK-ERRORS-NEXT: vmov.f64 d11, d10
 
         vcvt.f64.s32 d9, s8
         vcvt.f64.u32 d7, s6
         vcvt.s32.f64 s5, d4
         vcvt.u32.f64 s3, d2
         vcvtr.s32.f64 s1, d0
         vcvtr.u32.f64 s1, d2

llvm/test/MC/ARM/vmrs_vmsr.s

 // CHECK-V8M: vmrs r2, fpsid              @ encoding: [0xf0,0xee,0x10,0x2a]
 // CHECK-V8M: vmrs r3, fpsid              @ encoding: [0xf0,0xee,0x10,0x3a]
 // CHECK-V8M: vmrs r4, mvfr0              @ encoding: [0xf7,0xee,0x10,0x4a]
 // CHECK-V8M: vmrs r5, mvfr1              @ encoding: [0xf6,0xee,0x10,0x5a]
 // CHECK-V8M: vmrs r6, mvfr2              @ encoding: [0xf5,0xee,0x10,0x6a]
 // ERROR-V8M: invalid operand for instruction
 // ERROR-V8M: invalid operand for instruction
 
-// ERROR-NOVFP: instruction requires: VFP2
-// ERROR-NOVFP: instruction requires: VFP2
-// ERROR-NOVFP: instruction requires: VFP2
-// ERROR-NOVFP: instruction requires: VFP2
+// ERROR-NOVFP: instruction requires: fp registers
+// ERROR-NOVFP: instruction requires: fp registers
+// ERROR-NOVFP: instruction requires: fp registers
+// ERROR-NOVFP: instruction requires: fp registers
 // ERROR-NOVFP: instruction requires: VFP2
 // ERROR-NOVFP: instruction requires: VFP2
 // ERROR-NOVFP: instruction requires: VFP2
 // ERROR-NOVFP: instruction requires: VFP2
 // ERROR-NOVFP: instruction requires: FPARMv8
 // ERROR-NOVFP: invalid instruction
 // ERROR-NOVFP: invalid instruction
 

 // CHECK-V8M: vmsr  fpscr, r0             @ encoding: [0xe1,0xee,0x10,0x0a]
 // CHECK-V8M: vmsr  fpexc, r1             @ encoding: [0xe8,0xee,0x10,0x1a]
 // CHECK-V8M: vmsr  fpsid, r2             @ encoding: [0xe0,0xee,0x10,0x2a]
 // CHECK-V8M: vmsr  fpscr, r10            @ encoding: [0xe1,0xee,0x10,0xaa]
 // ERROR-V8M: invalid operand for instruction
 // ERROR-V8M: operand must be a register in range [r0, r14]
 
 // ERROR-NOVFP: invalid instruction
+// ERROR-NOVFP: instruction requires: fp registers
 // ERROR-NOVFP: instruction requires: VFP2
 // ERROR-NOVFP: instruction requires: VFP2
-// ERROR-NOVFP: instruction requires: VFP2
-// ERROR-NOVFP: instruction requires: VFP2
+// ERROR-NOVFP: instruction requires: fp registers
 // ERROR-NOVFP: invalid instruction
 // ERROR-NOVFP: invalid instruction

llvm/test/MC/Disassembler/ARM/mve-vmov-lane.txt

+# RUN: llvm-mc -triple=thumbv8m.main   -mattr=+fp-armv8 -disassemble < %s | FileCheck %s --check-prefix=D_REG
+# RUN: llvm-mc -triple=thumbv8a        -mattr=+fp-armv8 -disassemble < %s | FileCheck %s --check-prefix=D_REG
+# RUN: llvm-mc -triple=thumbv8.1m.main -mattr=+fp-armv8 -disassemble < %s | FileCheck %s --check-prefix=Q_REG
+# RUN: llvm-mc -triple=thumbv8.1m.main -mattr=+mve      -disassemble < %s | FileCheck %s --check-prefix=Q_REG
+
+# The disassembly for this instruction varies between v8.1M and other
+# architectures. In v8.1M (with either scalar flotaing point, MVE or both), we
+# use the Q register syntax, and for all other architectures we use the D
+# register syntax.
+
+[0x11,0xee,0x10,0x0b]
+# D_REG: vmov.32 r0, d1[0]
+# Q_REG: vmov.32 r0, q0[2]