[AMDGPU] predicate and feature refactoring

[rampitec]

Summary

We have done some predicate and feature refactoring lately but did not upstream it. This is to sync.

[rampitec]

Also renamed decoder namespaces.

[kzhuravl]

LGTM. Minor comment which does not affect the LGTM! Thanks for doing this.

lib/Target/AMDGPU/AMDGPU.td

714–716 ↗

(On Diff #193911)

Should this be renamed at some point too?

[rampitec]

lib/Target/AMDGPU/AMDGPU.td

714–716 ↗

(On Diff #193911)

Yes, but I want this to go from upstream and not the other way around. So a separate change.

[kzhuravl]

lib/Target/AMDGPU/AMDGPU.td

714–716 ↗

(On Diff #193911)

Sounds good, thanks.

[arsenm]

lib/Target/AMDGPU/AMDGPU.td

146–150 ↗

(On Diff #193911)

This is a compatibility breaking change, but we haven't been trying too hard to not break bitcode

[rampitec]

lib/Target/AMDGPU/AMDGPU.td

146–150 ↗

(On Diff #193911)

We managed to have two features at the same time, vi-insts and gfx8-insts, so compatibility is already compromised. This will just combine them from this point on.

[compnerd]

@arsenm - could write a bitcode upgrade path to deal with the breakage

llvm/trunk/lib/Target/AMDGPU/AMDGPU.td

 >;
 
 def FeatureCIInsts : SubtargetFeature<"ci-insts",
   "CIInsts",
   "true",
   "Additional instructions for CI+"
 >;
 
-def FeatureVIInsts : SubtargetFeature<"vi-insts",
-  "VIInsts",
+def FeatureGFX8Insts : SubtargetFeature<"gfx8-insts",
+  "GFX8Insts",
   "true",
-  "Additional instructions for VI+"
+  "Additional instructions for GFX8+"
 >;
 
 def FeatureGFX9Insts : SubtargetFeature<"gfx9-insts",
   "GFX9Insts",
   "true",
   "Additional instructions for GFX9+"
 >;
 
+def FeatureGFX7GFX8GFX9Insts : SubtargetFeature<"gfx7-gfx8-gfx9-insts",
+  "GFX7GFX8GFX9Insts",
+  "true",
+  "Instructions shared in GFX7, GFX8, GFX9"
+>;
+
 def FeatureSMemRealTime : SubtargetFeature<"s-memrealtime",
   "HasSMemRealTime",
   "true",
   "Has s_memrealtime instruction"
 >;
 
 def FeatureInv2PiInlineImm : SubtargetFeature<"inv-2pi-inline-imm",
   "HasInv2PiInlineImm",
def FeatureSouthernIslands : GCNSubtargetFeatureGeneration<"SOUTHERN_ISLANDS",
   FeatureDoesNotSupportSRAMECC]
 >;
 
 def FeatureSeaIslands : GCNSubtargetFeatureGeneration<"SEA_ISLANDS",
     "sea-islands",
   [FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128,
   FeatureWavefrontSize64, FeatureFlatAddressSpace,
   FeatureCIInsts, FeatureMovrel, FeatureTrigReducedRange,
-  FeatureDoesNotSupportSRAMECC]
+  FeatureGFX7GFX8GFX9Insts, FeatureDoesNotSupportSRAMECC]
 >;
 
 def FeatureVolcanicIslands : GCNSubtargetFeatureGeneration<"VOLCANIC_ISLANDS",
   "volcanic-islands",
   [FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128,
    FeatureWavefrontSize64, FeatureFlatAddressSpace,
-   FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
+   FeatureGCN3Encoding, FeatureCIInsts, Feature16BitInsts,
    FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel,
    FeatureScalarStores, FeatureInv2PiInlineImm,
    FeatureSDWA, FeatureSDWAOutModsVOPC, FeatureSDWAMac, FeatureDPP,
-   FeatureIntClamp, FeatureTrigReducedRange, FeatureDoesNotSupportSRAMECC
+   FeatureIntClamp, FeatureTrigReducedRange, FeatureDoesNotSupportSRAMECC,
+   FeatureGFX8Insts, FeatureGFX7GFX8GFX9Insts
   ]
 >;
 
 def FeatureGFX9 : GCNSubtargetFeatureGeneration<"GFX9",
   "gfx9",
   [FeatureFP64, FeatureLocalMemorySize65536,
    FeatureWavefrontSize64, FeatureFlatAddressSpace,
-   FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
+   FeatureGCN3Encoding, FeatureCIInsts, Feature16BitInsts,
    FeatureSMemRealTime, FeatureScalarStores, FeatureInv2PiInlineImm,
    FeatureApertureRegs, FeatureGFX9Insts, FeatureVOP3P, FeatureVGPRIndexMode,
    FeatureFastFMAF32, FeatureDPP, FeatureIntClamp,
    FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst,
    FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts,
-   FeatureAddNoCarryInsts, FeatureScalarAtomics, FeatureR128A16
+   FeatureAddNoCarryInsts, FeatureGFX8Insts, FeatureGFX7GFX8GFX9Insts,
+   FeatureScalarAtomics, FeatureR128A16
   ]
 >;
 
 class FeatureSet<list<SubtargetFeature> Features_> {
   list<SubtargetFeature> Features = Features_;
 }
 
 def FeatureISAVersion6_0_0 : FeatureSet<[FeatureSouthernIslands,

 // Dummy Instruction itineraries for pseudo instructions
 def ALU_NULL : FuncUnit;
 def NullALU : InstrItinClass;
 
 //===----------------------------------------------------------------------===//
 // Predicate helper class
 //===----------------------------------------------------------------------===//
 
-def isSICI : Predicate<
-  "Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS ||"
-  "Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS"
->, AssemblerPredicate<"!FeatureGCN3Encoding">;
+def isGFX6 :
+  Predicate<"Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS">,
+  AssemblerPredicate<"FeatureSouthernIslands">;
+
+def isGFX6GFX7 :
+  Predicate<"Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS ||"
+            "Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS">,
+  AssemblerPredicate<"!FeatureGCN3Encoding">;
+
+def isGFX7 :
+  Predicate<"Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS">,
+  AssemblerPredicate<"!FeatureGCN3Encoding,FeatureCIInsts">;
+
+def isGFX7GFX8GFX9 :
+  Predicate<"Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS ||"
+            "Subtarget->getGeneration() == AMDGPUSubtarget::VOLCANIC_ISLANDS ||"
+            "Subtarget->getGeneration() == AMDGPUSubtarget::GFX9">,
+  AssemblerPredicate<"FeatureGFX7GFX8GFX9Insts">;
+
+def isGFX7Plus :
+  Predicate<"Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS">,
+  AssemblerPredicate<"FeatureCIInsts">;
+
+def isGFX8Plus :
+  Predicate<"Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS">,
+  AssemblerPredicate<"FeatureGFX8Insts">;
+
+def isGFX9Plus :
+  Predicate<"Subtarget->getGeneration() >= AMDGPUSubtarget::GFX9">,
+  AssemblerPredicate<"FeatureGFX9Insts">;
 
 def isVI : Predicate <
   "Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS">,
   AssemblerPredicate<"FeatureGCN3Encoding">;
 
 def isGFX9 : Predicate <
-  "Subtarget->getGeneration() >= AMDGPUSubtarget::GFX9">,
-  AssemblerPredicate<"FeatureGFX9Insts">;
-
-// TODO: Either the name to be changed or we simply use IsCI!
-def isCIVI : Predicate <
-  "Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS">,
-  AssemblerPredicate<"FeatureCIInsts">;
+  "Subtarget->getGeneration() == AMDGPUSubtarget::GFX9">,
+  AssemblerPredicate<"FeatureGCN3Encoding,FeatureGFX9Insts">;
 
 def HasFlatAddressSpace : Predicate<"Subtarget->hasFlatAddressSpace()">,
   AssemblerPredicate<"FeatureFlatAddressSpace">;
 
 def HasFlatGlobalInsts : Predicate<"Subtarget->hasFlatGlobalInsts()">,
   AssemblerPredicate<"FeatureFlatGlobalInsts">;
 def HasFlatScratchInsts : Predicate<"Subtarget->hasFlatScratchInsts()">,
   AssemblerPredicate<"FeatureFlatScratchInsts">;

 def Has16BitInsts : Predicate<"Subtarget->has16BitInsts()">,
   AssemblerPredicate<"Feature16BitInsts">;
 def HasVOP3PInsts : Predicate<"Subtarget->hasVOP3PInsts()">,
   AssemblerPredicate<"FeatureVOP3P">;
 
 def HasSDWA : Predicate<"Subtarget->hasSDWA()">,
   AssemblerPredicate<"FeatureSDWA,FeatureVolcanicIslands">;
 
-def HasSDWA9 : Predicate<"Subtarget->hasSDWA()">,
-  AssemblerPredicate<"FeatureSDWA,FeatureGFX9">;
+def HasSDWA9 :
+  Predicate<"Subtarget->hasSDWA()">,
+  AssemblerPredicate<"FeatureGCN3Encoding,FeatureGFX9Insts,FeatureSDWA">;
 
 def HasDPP : Predicate<"Subtarget->hasDPP()">,
-  AssemblerPredicate<"FeatureDPP">;
+  AssemblerPredicate<"FeatureGCN3Encoding,FeatureDPP">;
 
 def HasR128A16 : Predicate<"Subtarget->hasR128A16()">,
   AssemblerPredicate<"FeatureR128A16">;
 
 def HasIntClamp : Predicate<"Subtarget->hasIntClamp()">,
   AssemblerPredicate<"FeatureIntClamp">;
 
 def HasMadMixInsts : Predicate<"Subtarget->hasMadMixInsts()">,

llvm/trunk/lib/Target/AMDGPU/AMDGPUGISel.td

multiclass GISelVop2IntrPat <
   def : GISelVop2CommutePat <node, inst, dst_vt, src_vt>;
 }
 
 def : GISelSop2Pat <or, S_OR_B32, i32>;
 def : GISelVop2Pat <or, V_OR_B32_e32, i32>;
 
 def : GISelSop2Pat <sra, S_ASHR_I32, i32>;
 let AddedComplexity = 100 in {
-let SubtargetPredicate = isSICI in {
+let SubtargetPredicate = isGFX6GFX7 in {
 def : GISelVop2Pat <sra, V_ASHR_I32_e32, i32>;
 }
 def : GISelVop2CommutePat <sra, V_ASHRREV_I32_e32, i32>;
 }
 def : GISelVop3Pat2CommutePat <sra, V_ASHRREV_I32_e64, i32>;
 
 // FIXME: We can't re-use SelectionDAG patterns here because they match
 // against a custom SDNode and we would need to create a generic machine

llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h

protected:
 
   // Subtarget statically properties set by tablegen
   bool FP64;
   bool FMA;
   bool MIMG_R128;
   bool IsGCN;
   bool GCN3Encoding;
   bool CIInsts;
-  bool VIInsts;
+  bool GFX8Insts;
   bool GFX9Insts;
+  bool GFX7GFX8GFX9Insts;
   bool SGPRInitBug;
   bool HasSMemRealTime;
   bool HasIntClamp;
   bool HasFmaMixInsts;
   bool HasMovrel;
   bool HasVGPRIndexMode;
   bool HasScalarStores;
   bool HasScalarAtomics;
bool hasScalarStores() const {
     return HasScalarStores;
   }
 
   bool hasScalarAtomics() const {
     return HasScalarAtomics;
   }
 
   bool hasLDSFPAtomics() const {
-    return VIInsts;
+    return GFX8Insts;
   }
 
   bool hasDPP() const {
     return HasDPP;
   }
 
   bool hasR128A16() const {
     return HasR128A16;
public:
   }
 
   // \returns true if the subtarget supports DWORDX3 load/store instructions.
   bool hasDwordx3LoadStores() const {
     return CIInsts;
   }
 
   bool hasSMovFedHazard() const {
-    return getGeneration() >= AMDGPUSubtarget::GFX9;
+    return getGeneration() == AMDGPUSubtarget::GFX9;
   }
 
   bool hasReadM0MovRelInterpHazard() const {
-    return getGeneration() >= AMDGPUSubtarget::GFX9;
+    return getGeneration() == AMDGPUSubtarget::GFX9;
   }
 
   bool hasReadM0SendMsgHazard() const {
-    return getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS;
+    return getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS &&
+           getGeneration() <= AMDGPUSubtarget::GFX9;
   }
 
   /// Return the maximum number of waves per SIMD for kernels using \p SGPRs
   /// SGPRs
   unsigned getOccupancyWithNumSGPRs(unsigned SGPRs) const;
 
   /// Return the maximum number of waves per SIMD for kernels using \p VGPRs
   /// VGPRs

llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.cpp

GCNSubtarget::GCNSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
     EnableSIScheduler(false),
     EnableDS128(false),
     EnablePRTStrictNull(false),
     DumpCode(false),
 
     FP64(false),
     GCN3Encoding(false),
     CIInsts(false),
-    VIInsts(false),
+    GFX8Insts(false),
     GFX9Insts(false),
+    GFX7GFX8GFX9Insts(false),
     SGPRInitBug(false),
     HasSMemRealTime(false),
     HasIntClamp(false),
     HasFmaMixInsts(false),
     HasMovrel(false),
     HasVGPRIndexMode(false),
     HasScalarStores(false),
     HasScalarAtomics(false),

llvm/trunk/lib/Target/AMDGPU/BUFInstructions.td

 defm BUFFER_ATOMIC_DEC_X2 : MUBUF_Pseudo_Atomics <
   "buffer_atomic_dec_x2", VReg_64, i64, atomic_dec_global
 >;
 
 let SubtargetPredicate = isVI in {
 def BUFFER_STORE_LDS_DWORD : MUBUF_Pseudo_Store_Lds <"buffer_store_lds_dword">;
 }
 
-let SubtargetPredicate = isSI in { // isn't on CI & VI
+let SubtargetPredicate = isGFX6 in { // isn't on CI & VI
 /*
 defm BUFFER_ATOMIC_RSUB        : MUBUF_Pseudo_Atomics <"buffer_atomic_rsub">;
 defm BUFFER_ATOMIC_FCMPSWAP    : MUBUF_Pseudo_Atomics <"buffer_atomic_fcmpswap">;
 defm BUFFER_ATOMIC_FMIN        : MUBUF_Pseudo_Atomics <"buffer_atomic_fmin">;
 defm BUFFER_ATOMIC_FMAX        : MUBUF_Pseudo_Atomics <"buffer_atomic_fmax">;
 defm BUFFER_ATOMIC_RSUB_X2     : MUBUF_Pseudo_Atomics <"buffer_atomic_rsub_x2">;
 defm BUFFER_ATOMIC_FCMPSWAP_X2 : MUBUF_Pseudo_Atomics <"buffer_atomic_fcmpswap_x2">;
 defm BUFFER_ATOMIC_FMIN_X2     : MUBUF_Pseudo_Atomics <"buffer_atomic_fmin_x2">;
let SubtargetPredicate = HasPackedD16VMem, D16Buf = 1 in {
   defm TBUFFER_LOAD_FORMAT_D16_XYZ   : MTBUF_Pseudo_Loads  <"tbuffer_load_format_d16_xyz",   VReg_64>;
   defm TBUFFER_LOAD_FORMAT_D16_XYZW  : MTBUF_Pseudo_Loads  <"tbuffer_load_format_d16_xyzw",  VReg_64>;
   defm TBUFFER_STORE_FORMAT_D16_X    : MTBUF_Pseudo_Stores <"tbuffer_store_format_d16_x",    VGPR_32>;
   defm TBUFFER_STORE_FORMAT_D16_XY   : MTBUF_Pseudo_Stores <"tbuffer_store_format_d16_xy",   VGPR_32>;
   defm TBUFFER_STORE_FORMAT_D16_XYZ  : MTBUF_Pseudo_Stores <"tbuffer_store_format_d16_xyz",  VReg_64>;
   defm TBUFFER_STORE_FORMAT_D16_XYZW : MTBUF_Pseudo_Stores <"tbuffer_store_format_d16_xyzw", VReg_64>;
 } // End HasPackedD16VMem.
 
-let SubtargetPredicate = isCIVI in {
+let SubtargetPredicate = isGFX7Plus in {
 
 //===----------------------------------------------------------------------===//
 // Instruction definitions for CI and newer.
 //===----------------------------------------------------------------------===//
 
 def BUFFER_WBINVL1_VOL : MUBUF_Invalidate <"buffer_wbinvl1_vol",
                                            int_amdgcn_buffer_wbinvl1_vol>;
 
-} // End let SubtargetPredicate = isCIVI
+} // End let SubtargetPredicate = isGFX7Plus
 
 //===----------------------------------------------------------------------===//
 // MUBUF Patterns
 //===----------------------------------------------------------------------===//
 
 def extract_glc : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getZExtValue() & 1, SDLoc(N), MVT::i8);
 }]>;
multiclass MUBUFLoad_Atomic_Pattern <MUBUF_Pseudo Instr_ADDR64, MUBUF_Pseudo Instr_OFFSET,
   >;
 
   def : GCNPat <
     (vt (atomic_ld (MUBUFOffsetNoGLC v4i32:$rsrc, i32:$soffset, i16:$offset))),
     (Instr_OFFSET $rsrc, $soffset, (as_i16imm $offset), 0, 0, 0)
   >;
 }
 
-let SubtargetPredicate = isSICI in {
+let SubtargetPredicate = isGFX6GFX7 in {
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_SBYTE_ADDR64, i32, sextloadi8_constant>;
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_UBYTE_ADDR64, i32, az_extloadi8_constant>;
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_SSHORT_ADDR64, i32, sextloadi16_constant>;
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, az_extloadi16_constant>;
 
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORD_ADDR64, BUFFER_LOAD_DWORD_OFFSET, i32, mubuf_load_atomic>;
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, BUFFER_LOAD_DWORDX2_OFFSET, i64, mubuf_load_atomic>;
-} // End SubtargetPredicate = isSICI
+} // End SubtargetPredicate = isGFX6GFX7
 
 multiclass MUBUFLoad_Pattern <MUBUF_Pseudo Instr_OFFSET, ValueType vt,
                                PatFrag ld> {
 
   def : GCNPat <
     (vt (ld (MUBUFOffset v4i32:$srsrc, i32:$soffset,
                           i16:$offset, i1:$glc, i1:$slc, i1:$tfe))),
     (Instr_OFFSET $srsrc, $soffset, $offset, $glc, $slc, $tfe)

 
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_SHORT_D16_OFFEN, BUFFER_LOAD_SHORT_D16_OFFSET, v2i16, load_d16_lo_private>;
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_UBYTE_D16_OFFEN, BUFFER_LOAD_UBYTE_D16_OFFSET, v2i16, az_extloadi8_d16_lo_private>;
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_SBYTE_D16_OFFEN, BUFFER_LOAD_SBYTE_D16_OFFSET, v2i16, sextloadi8_d16_lo_private>;
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_SHORT_D16_OFFEN, BUFFER_LOAD_SHORT_D16_OFFSET, v2f16, load_d16_lo_private>;
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_UBYTE_D16_OFFEN, BUFFER_LOAD_UBYTE_D16_OFFSET, v2f16, az_extloadi8_d16_lo_private>;
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_SBYTE_D16_OFFEN, BUFFER_LOAD_SBYTE_D16_OFFSET, v2f16, sextloadi8_d16_lo_private>;
 }
+
 multiclass MUBUFStore_Atomic_Pattern <MUBUF_Pseudo Instr_ADDR64, MUBUF_Pseudo Instr_OFFSET,
                                       ValueType vt, PatFrag atomic_st> {
   // Store follows atomic op convention so address is forst
   def : GCNPat <
      (atomic_st (MUBUFAddr64 v4i32:$srsrc, i64:$vaddr, i32:$soffset,
                                    i16:$offset, i1:$slc), vt:$val),
      (Instr_ADDR64 $val, $vaddr, $srsrc, $soffset, $offset, 0, $slc, 0)
   >;
 
   def : GCNPat <
     (atomic_st (MUBUFOffsetNoGLC v4i32:$rsrc, i32:$soffset, i16:$offset), vt:$val),
     (Instr_OFFSET $val, $rsrc, $soffset, (as_i16imm $offset), 0, 0, 0)
   >;
 }
-let SubtargetPredicate = isSICI in {
+let SubtargetPredicate = isGFX6GFX7 in {
 defm : MUBUFStore_Atomic_Pattern <BUFFER_STORE_DWORD_ADDR64, BUFFER_STORE_DWORD_OFFSET, i32, store_atomic_global>;
 defm : MUBUFStore_Atomic_Pattern <BUFFER_STORE_DWORDX2_ADDR64, BUFFER_STORE_DWORDX2_OFFSET, i64, store_atomic_global>;
-} // End Predicates = isSICI
+} // End Predicates = isGFX6GFX7
 
 
 multiclass MUBUFStore_Pattern <MUBUF_Pseudo Instr_OFFSET, ValueType vt,
                                PatFrag st> {
 
   def : GCNPat <
     (st vt:$vdata, (MUBUFOffset v4i32:$srsrc, i32:$soffset,
                                       i16:$offset, i1:$glc, i1:$slc, i1:$tfe)),

 
 let SubtargetPredicate = HasPackedD16VMem in {
   defm : MTBUF_StoreIntrinsicPat<SItbuffer_store_d16, f16,   "TBUFFER_STORE_FORMAT_D16_X">;
   defm : MTBUF_StoreIntrinsicPat<SItbuffer_store_d16, v2f16, "TBUFFER_STORE_FORMAT_D16_XY">;
   defm : MTBUF_StoreIntrinsicPat<SItbuffer_store_d16, v4f16, "TBUFFER_STORE_FORMAT_D16_XYZW">;
 } // End HasPackedD16VMem.
 
 //===----------------------------------------------------------------------===//
-// Target instructions, move to the appropriate target TD file
+// Target-specific instruction encodings.
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
-// SI
+// Base ENC_MUBUF for GFX6, GFX7.
 //===----------------------------------------------------------------------===//
 
 class MUBUF_Real_si <bits<7> op, MUBUF_Pseudo ps> :
   MUBUF_Real<op, ps>,
   Enc64,
   SIMCInstr<ps.PseudoInstr, SIEncodingFamily.SI> {
-  let AssemblerPredicate=isSICI;
-  let DecoderNamespace="SICI";
+  let AssemblerPredicate=isGFX6GFX7;
+  let DecoderNamespace="GFX6GFX7";
 
   let Inst{11-0}  = !if(ps.has_offset, offset, ?);
   let Inst{12}    = ps.offen;
   let Inst{13}    = ps.idxen;
   let Inst{14}    = !if(ps.has_glc, glc, ps.glc_value);
   let Inst{15}    = ps.addr64;
   let Inst{16}    = !if(ps.lds, 1, 0);
   let Inst{24-18} = op;

 
 def BUFFER_WBINVL1_SC_si        : MUBUF_Real_si <0x70, BUFFER_WBINVL1_SC>;
 def BUFFER_WBINVL1_si           : MUBUF_Real_si <0x71, BUFFER_WBINVL1>;
 
 class MTBUF_Real_si <bits<3> op, MTBUF_Pseudo ps> :
   MTBUF_Real<ps>,
   Enc64,
   SIMCInstr<ps.PseudoInstr, SIEncodingFamily.SI> {
-  let AssemblerPredicate=isSICI;
-  let DecoderNamespace="SICI";
+  let AssemblerPredicate=isGFX6GFX7;
+  let DecoderNamespace="GFX6GFX7";
 
   let Inst{11-0}  = !if(ps.has_offset, offset, ?);
   let Inst{12}    = ps.offen;
   let Inst{13}    = ps.idxen;
   let Inst{14}    = !if(ps.has_glc, glc, ps.glc_value);
   let Inst{15}    = ps.addr64;
   let Inst{18-16} = op;
   let Inst{22-19} = dfmt;

 //===----------------------------------------------------------------------===//
 // CI
 // MTBUF - GFX6, GFX7.
 //===----------------------------------------------------------------------===//
 
 class MUBUF_Real_ci <bits<7> op, MUBUF_Pseudo ps> :
   MUBUF_Real_si<op, ps> {
   let AssemblerPredicate=isCIOnly;
-  let DecoderNamespace="CI";
+  let DecoderNamespace="GFX7";
 }
 
 def BUFFER_WBINVL1_VOL_ci : MUBUF_Real_ci <0x70, BUFFER_WBINVL1_VOL>;
 
 
 //===----------------------------------------------------------------------===//
-// VI
+//  GFX8, GFX9 (VI).
 //===----------------------------------------------------------------------===//
 
 class MUBUF_Real_vi <bits<7> op, MUBUF_Pseudo ps> :
   MUBUF_Real<op, ps>,
   Enc64,
   SIMCInstr<ps.PseudoInstr, SIEncodingFamily.VI> {
   let AssemblerPredicate=isVI;
   let DecoderNamespace="VI";

llvm/trunk/lib/Target/AMDGPU/DSInstructions.td

 def DS_CONSUME       : DS_0A_RET<"ds_consume">;
 def DS_APPEND        : DS_0A_RET<"ds_append">;
 def DS_ORDERED_COUNT : DS_1A_RET_GDS<"ds_ordered_count">;
 
 //===----------------------------------------------------------------------===//
 // Instruction definitions for CI and newer.
 //===----------------------------------------------------------------------===//
 
-let SubtargetPredicate = isCIVI in {
+let SubtargetPredicate = isGFX7Plus in {
 
 defm DS_WRAP_RTN_B32 : DS_1A2D_RET_mc<"ds_wrap_rtn_b32", VGPR_32>;
 defm DS_CONDXCHG32_RTN_B64 : DS_1A1D_RET_mc<"ds_condxchg32_rtn_b64", VReg_64>;
 
 def DS_GWS_SEMA_RELEASE_ALL : DS_GWS_0D<"ds_gws_sema_release_all">;
 
 let mayStore = 0 in {
 defm DS_READ_B96 : DS_1A_RET_mc<"ds_read_b96", VReg_96>;
 defm DS_READ_B128: DS_1A_RET_mc<"ds_read_b128", VReg_128>;
 } // End mayStore = 0
 
 let mayLoad = 0 in {
 defm DS_WRITE_B96 : DS_1A1D_NORET_mc<"ds_write_b96", VReg_96>;
 defm DS_WRITE_B128 : DS_1A1D_NORET_mc<"ds_write_b128", VReg_128>;
 } // End mayLoad = 0
 
 def DS_NOP : DS_VOID<"ds_nop">;
 
-} // let SubtargetPredicate = isCIVI
+} // let SubtargetPredicate = isGFX7Plus
 
 //===----------------------------------------------------------------------===//
 // Instruction definitions for VI and newer.
 //===----------------------------------------------------------------------===//
 
-let SubtargetPredicate = isVI in {
+let SubtargetPredicate = isGFX8Plus in {
 
 let Uses = [EXEC] in {
 def DS_PERMUTE_B32  : DS_1A1D_PERMUTE <"ds_permute_b32",
                                        int_amdgcn_ds_permute>;
 def DS_BPERMUTE_B32 : DS_1A1D_PERMUTE <"ds_bpermute_b32",
                                        int_amdgcn_ds_bpermute>;
 }
 
 def DS_ADD_SRC2_F32 : DS_1A<"ds_add_src2_f32">;
 
-} // let SubtargetPredicate = isVI
+} // let SubtargetPredicate = isGFX8Plus
 
 //===----------------------------------------------------------------------===//
 // DS Patterns
 //===----------------------------------------------------------------------===//
 
 def : GCNPat <
   (int_amdgcn_ds_swizzle i32:$src, imm:$offset16),
   (DS_SWIZZLE_B32 $src, (as_i16imm $offset16), (i1 0))
class DS64Bit4ByteAlignedWritePat<DS_Pseudo inst, PatFrag frag> : GCNPat<
   (frag v2i32:$value, (DS64Bit4ByteAligned i32:$ptr, i8:$offset0, i8:$offset1)),
   (inst $ptr, (i32 (EXTRACT_SUBREG $value, sub0)),
               (i32 (EXTRACT_SUBREG $value, sub1)), $offset0, $offset1,
               (i1 0))
 >;
 
 // v2i32 loads are split into i32 loads on SI during lowering, due to a bug
 // related to bounds checking.
-let OtherPredicates = [LDSRequiresM0Init, isCIVI] in {
+let OtherPredicates = [LDSRequiresM0Init, isGFX7Plus] in {
 def : DS64Bit4ByteAlignedReadPat<DS_READ2_B32, load_local_m0>;
 def : DS64Bit4ByteAlignedWritePat<DS_WRITE2_B32, store_local_m0>;
 }
 
 let OtherPredicates = [NotLDSRequiresM0Init] in {
 def : DS64Bit4ByteAlignedReadPat<DS_READ2_B32_gfx9, load_local>;
 def : DS64Bit4ByteAlignedWritePat<DS_WRITE2_B32_gfx9, store_local>;
 }

 defm : DSAtomicCmpXChg_mc<DS_CMPST_RTN_B64, i64, "atomic_cmp_swap_local">;
 
 def : Pat <
   (SIds_ordered_count i32:$value, i16:$offset),
   (DS_ORDERED_COUNT $value, (as_i16imm $offset))
 >;
 
 //===----------------------------------------------------------------------===//
-// Real instructions
+// Target-specific instruction encodings.
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
-// SIInstructions.td
+// Base ENC_DS for GFX6, GFX7.
 //===----------------------------------------------------------------------===//
 
 class DS_Real_si <bits<8> op, DS_Pseudo ds> :
   DS_Real <ds>,
   SIMCInstr <ds.Mnemonic, SIEncodingFamily.SI> {
-  let AssemblerPredicates=[isSICI];
-  let DecoderNamespace="SICI";
+  let AssemblerPredicates=[isGFX6GFX7];
+  let DecoderNamespace="GFX6GFX7";
 
   // encoding
   let Inst{7-0}   = !if(ds.has_offset0, offset0, 0);
   let Inst{15-8}  = !if(ds.has_offset1, offset1, 0);
   let Inst{17}    = !if(ds.has_gds, gds, ds.gdsValue);
   let Inst{25-18} = op;
   let Inst{31-26} = 0x36; // ds prefix
   let Inst{39-32} = !if(ds.has_addr, addr, 0);

 def DS_MIN_SRC2_F64_si    : DS_Real_si<0xd2, DS_MIN_SRC2_F64>;
 def DS_MAX_SRC2_F64_si    : DS_Real_si<0xd3, DS_MAX_SRC2_F64>;
 def DS_WRITE_B96_si       : DS_Real_si<0xde, DS_WRITE_B96>;
 def DS_WRITE_B128_si      : DS_Real_si<0xdf, DS_WRITE_B128>;
 def DS_READ_B96_si        : DS_Real_si<0xfe, DS_READ_B96>;
 def DS_READ_B128_si       : DS_Real_si<0xff, DS_READ_B128>;
 
 //===----------------------------------------------------------------------===//
-// VIInstructions.td
+// GFX8, GFX9 (VI).
 //===----------------------------------------------------------------------===//
 
 class DS_Real_vi <bits<8> op, DS_Pseudo ds> :
   DS_Real <ds>,
   SIMCInstr <ds.Mnemonic, SIEncodingFamily.VI> {
   let AssemblerPredicates = [isVI];
   let DecoderNamespace="VI";
 

llvm/trunk/lib/Target/AMDGPU/FLATInstructions.td

class FLAT_Pseudo<string opName, dag outs, dag ins,
   // TODO: M0 if it could possibly access LDS (before gfx9? only)?
   let Uses = !if(is_flat_global, [EXEC], [EXEC, FLAT_SCR]);
 
   // Internally, FLAT instruction are executed as both an LDS and a
   // Buffer instruction; so, they increment both VM_CNT and LGKM_CNT
   // and are not considered done until both have been decremented.
   let VM_CNT = 1;
   let LGKM_CNT = !if(!or(is_flat_global, is_flat_scratch), 0, 1);
+
+  let IsNonFlatSeg = !if(!or(is_flat_global, is_flat_scratch), 1, 0);
 }
 
 class FLAT_Real <bits<7> op, FLAT_Pseudo ps> :
   InstSI <ps.OutOperandList, ps.InOperandList, ps.Mnemonic # ps.AsmOperands, []>,
   Enc64 {
 
   let isPseudo = 0;
   let isCodeGenOnly = 0;
defm FLAT_ATOMIC_XOR_X2     : FLAT_Atomic_Pseudo <"flat_atomic_xor_x2",
                                 VReg_64, i64, atomic_xor_flat>;
 
 defm FLAT_ATOMIC_INC_X2     : FLAT_Atomic_Pseudo <"flat_atomic_inc_x2",
                                 VReg_64, i64, atomic_inc_flat>;
 
 defm FLAT_ATOMIC_DEC_X2     : FLAT_Atomic_Pseudo <"flat_atomic_dec_x2",
                                 VReg_64, i64, atomic_dec_flat>;
 
-let SubtargetPredicate = isCI in { // CI Only flat instructions : FIXME Only?
+// GFX7-only flat instructions.
+let SubtargetPredicate = isGFX7 in {
 
 defm FLAT_ATOMIC_FCMPSWAP    : FLAT_Atomic_Pseudo <"flat_atomic_fcmpswap",
                                 VGPR_32, f32, null_frag, v2f32, VReg_64>;
 
 defm FLAT_ATOMIC_FCMPSWAP_X2 : FLAT_Atomic_Pseudo <"flat_atomic_fcmpswap_x2",
                                 VReg_64, f64, null_frag, v2f64, VReg_128>;
 
 defm FLAT_ATOMIC_FMIN        : FLAT_Atomic_Pseudo <"flat_atomic_fmin",
                                 VGPR_32, f32>;
 
 defm FLAT_ATOMIC_FMAX        : FLAT_Atomic_Pseudo <"flat_atomic_fmax",
                                 VGPR_32, f32>;
 
 defm FLAT_ATOMIC_FMIN_X2     : FLAT_Atomic_Pseudo <"flat_atomic_fmin_x2",
                                 VReg_64, f64>;
 
 defm FLAT_ATOMIC_FMAX_X2     : FLAT_Atomic_Pseudo <"flat_atomic_fmax_x2",
                                 VReg_64, f64>;
 
-} // End SubtargetPredicate = isCI
+} // End SubtargetPredicate = isGFX7
 
 let SubtargetPredicate = HasFlatGlobalInsts in {
 defm GLOBAL_LOAD_UBYTE    : FLAT_Global_Load_Pseudo <"global_load_ubyte", VGPR_32>;
 defm GLOBAL_LOAD_SBYTE    : FLAT_Global_Load_Pseudo <"global_load_sbyte", VGPR_32>;
 defm GLOBAL_LOAD_USHORT   : FLAT_Global_Load_Pseudo <"global_load_ushort", VGPR_32>;
 defm GLOBAL_LOAD_SSHORT   : FLAT_Global_Load_Pseudo <"global_load_sshort", VGPR_32>;
 defm GLOBAL_LOAD_DWORD    : FLAT_Global_Load_Pseudo <"global_load_dword", VGPR_32>;
 defm GLOBAL_LOAD_DWORDX2  : FLAT_Global_Load_Pseudo <"global_load_dwordx2", VReg_64>;

 //===----------------------------------------------------------------------===//
 // CI
 //===----------------------------------------------------------------------===//
 
 class FLAT_Real_ci <bits<7> op, FLAT_Pseudo ps> :
   FLAT_Real <op, ps>,
   SIMCInstr <ps.PseudoInstr, SIEncodingFamily.SI> {
   let AssemblerPredicate = isCIOnly;
-  let DecoderNamespace="CI";
+  let DecoderNamespace="GFX7";
 }
 
 def FLAT_LOAD_UBYTE_ci         : FLAT_Real_ci <0x8,  FLAT_LOAD_UBYTE>;
 def FLAT_LOAD_SBYTE_ci         : FLAT_Real_ci <0x9,  FLAT_LOAD_SBYTE>;
 def FLAT_LOAD_USHORT_ci        : FLAT_Real_ci <0xa,  FLAT_LOAD_USHORT>;
 def FLAT_LOAD_SSHORT_ci        : FLAT_Real_ci <0xb,  FLAT_LOAD_SSHORT>;
 def FLAT_LOAD_DWORD_ci         : FLAT_Real_ci <0xc,  FLAT_LOAD_DWORD>;
 def FLAT_LOAD_DWORDX2_ci       : FLAT_Real_ci <0xd,  FLAT_LOAD_DWORDX2>;

llvm/trunk/lib/Target/AMDGPU/MIMGInstructions.td

let InOperandList = (ins data_rc:$vdata, addr_rc:$vaddr, SReg_256:$srsrc,
                            DMask:$dmask, UNorm:$unorm, GLC:$glc, SLC:$slc,
                            R128A16:$r128, TFE:$tfe, LWE:$lwe, DA:$da);
   let AsmString = asm#" $vdst, $vaddr, $srsrc$dmask$unorm$glc$slc$r128$tfe$lwe$da";
 }
 
 multiclass MIMG_Atomic_Helper_m <mimg op, string asm, RegisterClass data_rc,
                                  RegisterClass addr_rc, bit enableDasm = 0> {
   let ssamp = 0, d16 = 0 in {
-    def _si : MIMG_Atomic_Helper<asm, data_rc, addr_rc, "SICI", enableDasm>,
+    def _si : MIMG_Atomic_Helper<asm, data_rc, addr_rc, "GFX6GFX7", enableDasm>,
               SIMCInstr<NAME, SIEncodingFamily.SI>,
               MIMGe<op.SI> {
-      let AssemblerPredicates = [isSICI];
+      let AssemblerPredicates = [isGFX6GFX7];
       let DisableDecoder = DisableSIDecoder;
     }
 
     def _vi : MIMG_Atomic_Helper<asm, data_rc, addr_rc, "VI", enableDasm>,
               SIMCInstr<NAME, SIEncodingFamily.VI>,
               MIMGe<op.VI> {
       let AssemblerPredicates = [isVI];
       let DisableDecoder = DisableVIDecoder;

llvm/trunk/lib/Target/AMDGPU/SIDefines.h

// TODO: Should this be spilt into VOP3 a and b?
   ClampHi = UINT64_C(1) << 48,
 
   // Is a packed VOP3P instruction.
   IsPacked = UINT64_C(1) << 49,
 
   // Is a D16 buffer instruction.
   D16Buf = UINT64_C(1) << 50,
 
+  // FLAT instruction accesses FLAT_GLBL or FLAT_SCRATCH segment.
+  IsNonFlatSeg = UINT64_C(1) << 51,
+
   // Uses floating point double precision rounding mode
-  FPDPRounding = UINT64_C(1) << 51
+  FPDPRounding = UINT64_C(1) << 52
 };
 
 // v_cmp_class_* etc. use a 10-bit mask for what operation is checked.
 // The result is true if any of these tests are true.
 enum ClassFlags : unsigned {
   S_NAN = 1 << 0,        // Signaling NaN
   Q_NAN = 1 << 1,        // Quiet NaN
   N_INFINITY = 1 << 2,   // Negative infinity

llvm/trunk/lib/Target/AMDGPU/SIInstrFormats.td

 //===-- SIInstrFormats.td - SI Instruction Encodings ----------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // SI Instruction format definitions.
 //
 //===----------------------------------------------------------------------===//
 
-def isSI : Predicate<"Subtarget->getGeneration() "
-                      "== AMDGPUSubtarget::SOUTHERN_ISLANDS">,
-           AssemblerPredicate<"FeatureSouthernIslands">;
-
-
 class InstSI <dag outs, dag ins, string asm = "",
               list<dag> pattern = []> :
   AMDGPUInst<outs, ins, asm, pattern>, GCNPredicateControl {
   // Low bits - basic encoding information.
   field bit SALU = 0;
   field bit VALU = 0;
 
   // SALU instruction formats.
class InstSI <dag outs, dag ins, string asm = "",
   field bit ClampHi = 0;
 
   // This bit indicates that this is a packed VOP3P instruction
   field bit IsPacked = 0;
 
   // This bit indicates that this is a D16 buffer instruction.
   field bit D16Buf = 0;
 
+  // This field indicates that FLAT instruction accesses FLAT_GLBL or
+  // FLAT_SCRATCH segment. Must be 0 for non-FLAT instructions.
+  field bit IsNonFlatSeg = 0;
+
   // This bit indicates that this uses the floating point double precision
   // rounding mode flags
   field bit FPDPRounding = 0;
 
   // These need to be kept in sync with the enum in SIInstrFlags.
   let TSFlags{0} = SALU;
   let TSFlags{1} = VALU;
 
class InstSI <dag outs, dag ins, string asm = "",
   let TSFlags{46} = IntClamp;
   let TSFlags{47} = ClampLo;
   let TSFlags{48} = ClampHi;
 
   let TSFlags{49} = IsPacked;
 
   let TSFlags{50} = D16Buf;
 
-  let TSFlags{51} = FPDPRounding;
+  let TSFlags{51} = IsNonFlatSeg;
+
+  let TSFlags{52} = FPDPRounding;
 
   let SchedRW = [Write32Bit];
 
   field bits<1> DisableSIDecoder = 0;
   field bits<1> DisableVIDecoder = 0;
   field bits<1> DisableDecoder = 0;
 
   let isAsmParserOnly = !if(!eq(DisableDecoder{0}, {0}), 0, 1);

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td

def isCIOnly : Predicate<"Subtarget->getGeneration() =="
   AssemblerPredicate <"FeatureSeaIslands">;
 def isVIOnly : Predicate<"Subtarget->getGeneration() =="
                          "AMDGPUSubtarget::VOLCANIC_ISLANDS">,
   AssemblerPredicate <"FeatureVolcanicIslands">;
 
 def DisableInst : Predicate <"false">, AssemblerPredicate<"FeatureDisable">;
 
 class GCNPredicateControl : PredicateControl {
-  Predicate SIAssemblerPredicate = isSICI;
+  Predicate SIAssemblerPredicate = isGFX6GFX7;
   Predicate VIAssemblerPredicate = isVI;
 }
 
 // Execpt for the NONE field, this must be kept in sync with the
 // SIEncodingFamily enum in AMDGPUInstrInfo.cpp
 def SIEncodingFamily {
   int NONE = -1;
   int SI = 0;
let isPseudo = 1, isCodeGenOnly = 1 in {
       def "" : EXP_Helper<done, node>,
                SIMCInstr <"exp"#!if(done, "_done", ""), SIEncodingFamily.NONE>;
     }
 
     let done = done in {
       def _si : EXP_Helper<done>,
                 SIMCInstr <"exp"#!if(done, "_done", ""), SIEncodingFamily.SI>,
                 EXPe {
-        let AssemblerPredicates = [isSICI];
-        let DecoderNamespace = "SICI";
+        let AssemblerPredicates = [isGFX6GFX7];
+        let DecoderNamespace = "GFX6GFX7";
         let DisableDecoder = DisableSIDecoder;
       }
 
       def _vi : EXP_Helper<done>,
                 SIMCInstr <"exp"#!if(done, "_done", ""), SIEncodingFamily.VI>,
                 EXPe_vi {
         let AssemblerPredicates = [isVI];
         let DecoderNamespace = "VI";

 }
 
 class VINTRP_Real_si <bits <2> op, string opName, dag outs, dag ins,
                       string asm> :
   VINTRPCommon <outs, ins, asm, []>,
   VINTRPe <op>,
   SIMCInstr<opName, SIEncodingFamily.SI> {
   let AssemblerPredicate = SIAssemblerPredicate;
-  let DecoderNamespace = "SICI";
+  let DecoderNamespace = "GFX6GFX7";
   let DisableDecoder = DisableSIDecoder;
 }
 
 class VINTRP_Real_vi <bits <2> op, string opName, dag outs, dag ins,
                       string asm> :
   VINTRPCommon <outs, ins, asm, []>,
   VINTRPe_vi <op>,
   SIMCInstr<opName, SIEncodingFamily.VI> {

llvm/trunk/lib/Target/AMDGPU/SIInstructions.td

def : GCNPat <
   (v4f16 (scalar_to_vector f16:$src0)),
   (INSERT_SUBREG (IMPLICIT_DEF), $src0, sub0)
 >;
 
 //===----------------------------------------------------------------------===//
 // Fract Patterns
 //===----------------------------------------------------------------------===//
 
-let SubtargetPredicate = isSI in {
+let SubtargetPredicate = isGFX6 in {
 
 // V_FRACT is buggy on SI, so the F32 version is never used and (x-floor(x)) is
 // used instead. However, SI doesn't have V_FLOOR_F64, so the most efficient
 // way to implement it is using V_FRACT_F64.
 // The workaround for the V_FRACT bug is:
 //    fract(x) = isnan(x) ? x : min(V_FRACT(x), 0.99999999999999999)
 
 // Convert floor(x) to (x - fract(x))
(V_ADD_F64
              SRCMODS.NONE,
              (V_MOV_B64_PSEUDO 0x3fefffffffffffff),
              DSTCLAMP.NONE, DSTOMOD.NONE),
          $x,
          (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, (i32 3 /*NaN*/))),
       DSTCLAMP.NONE, DSTOMOD.NONE)
 >;
 
-} // End SubtargetPredicates = isSI
+} // End SubtargetPredicates = isGFX6
 
 //============================================================================//
 // Miscellaneous Optimization Patterns
 //============================================================================//
 
 // Undo sub x, c -> add x, -c canonicalization since c is more likely
 // an inline immediate than -c.
 // TODO: Also do for 64-bit.
>;
   (min (max_oneuse vt:$src0, vt:$src1),
       (max_oneuse (min_oneuse vt:$src0, vt:$src1), vt:$src2)),
   (med3Inst SRCMODS.NONE, $src0, SRCMODS.NONE, $src1, SRCMODS.NONE, $src2, DSTCLAMP.NONE)
 >;
 }
 
 def : FPMed3Pat<f32, V_MED3_F32>;
 
-let OtherPredicates = [isGFX9] in {
+let OtherPredicates = [isGFX9Plus] in {
 def : FP16Med3Pat<f16, V_MED3_F16>;
 defm : Int16Med3Pat<V_MED3_I16, smin, smax, smax_oneuse, smin_oneuse>;
 defm : Int16Med3Pat<V_MED3_U16, umin, umax, umax_oneuse, umin_oneuse>;
-} // End Predicates = [isGFX9]
+} // End Predicates = [isGFX9Plus]

llvm/trunk/lib/Target/AMDGPU/SMInstructions.td

class SM_Time_Pseudo<string opName, SDPatternOperator node = null_frag> : SM_Pseudo<
   // inferring differently for llvm.readcyclecounter and the amdgcn
   // intrinsics.
   let mayStore = ?;
   let mayLoad = 1;
   let has_sbase = 0;
   let has_offset = 0;
 }
 
-class SM_Inval_Pseudo <string opName, SDPatternOperator node> : SM_Pseudo<
+class SM_Inval_Pseudo <string opName, SDPatternOperator node = null_frag> : SM_Pseudo<
   opName, (outs), (ins), "", [(node)]> {
   let hasSideEffects = 1;
   let mayStore = 1;
   let has_sdst = 0;
   let has_sbase = 0;
   let has_offset = 0;
 }
 

 defm S_BUFFER_STORE_DWORDX4 : SM_Pseudo_Stores <
   "s_buffer_store_dwordx4", SReg_128, SReg_128
 >;
 
 
 def S_MEMTIME : SM_Time_Pseudo <"s_memtime", int_amdgcn_s_memtime>;
 def S_DCACHE_INV : SM_Inval_Pseudo <"s_dcache_inv", int_amdgcn_s_dcache_inv>;
 
-let SubtargetPredicate = isCIVI in {
+let SubtargetPredicate = isGFX7GFX8GFX9 in {
 def S_DCACHE_INV_VOL : SM_Inval_Pseudo <"s_dcache_inv_vol", int_amdgcn_s_dcache_inv_vol>;
-} // let SubtargetPredicate = isCIVI
+} // let SubtargetPredicate = isGFX7GFX8GFX9
 
-let SubtargetPredicate = isVI in {
+let SubtargetPredicate = isGFX8Plus in {
 def S_DCACHE_WB     : SM_Inval_Pseudo <"s_dcache_wb", int_amdgcn_s_dcache_wb>;
 def S_DCACHE_WB_VOL : SM_Inval_Pseudo <"s_dcache_wb_vol", int_amdgcn_s_dcache_wb_vol>;
 def S_MEMREALTIME   : SM_Time_Pseudo <"s_memrealtime", int_amdgcn_s_memrealtime>;
 
 defm S_ATC_PROBE        : SM_Pseudo_Probe <"s_atc_probe", SReg_64>;
 defm S_ATC_PROBE_BUFFER : SM_Pseudo_Probe <"s_atc_probe_buffer", SReg_128>;
-} // SubtargetPredicate = isVI
+} // SubtargetPredicate = isGFX8Plus
 
 let SubtargetPredicate = HasFlatScratchInsts, Uses = [FLAT_SCR] in {
 defm S_SCRATCH_LOAD_DWORD    : SM_Pseudo_Loads <"s_scratch_load_dword",   SReg_64, SReg_32_XM0_XEXEC>;
 defm S_SCRATCH_LOAD_DWORDX2  : SM_Pseudo_Loads <"s_scratch_load_dwordx2", SReg_64, SReg_64_XEXEC>;
 defm S_SCRATCH_LOAD_DWORDX4  : SM_Pseudo_Loads <"s_scratch_load_dwordx4", SReg_64, SReg_128>;
 
 defm S_SCRATCH_STORE_DWORD   : SM_Pseudo_Stores <"s_scratch_store_dword",   SReg_64, SReg_32_XM0_XEXEC>;
 defm S_SCRATCH_STORE_DWORDX2 : SM_Pseudo_Stores <"s_scratch_store_dwordx2", SReg_64, SReg_64_XEXEC>;

 // SI
 //===----------------------------------------------------------------------===//
 
 class SMRD_Real_si <bits<5> op, SM_Pseudo ps>
   : SM_Real<ps>
   , SIMCInstr<ps.PseudoInstr, SIEncodingFamily.SI>
   , Enc32 {
 
-  let AssemblerPredicates = [isSICI];
-  let DecoderNamespace = "SICI";
+  let AssemblerPredicates = [isGFX6GFX7];
+  let DecoderNamespace = "GFX6GFX7";
 
   let Inst{7-0}   = !if(ps.has_offset, offset{7-0}, ?);
   let Inst{8}     = imm;
   let Inst{14-9}  = !if(ps.has_sbase, sbase{6-1}, ?);
   let Inst{21-15} = !if(ps.has_sdst, sdst{6-0}, ?);
   let Inst{26-22} = op;
   let Inst{31-27} = 0x18; //encoding
 }
def smrd_literal_offset : NamedOperandU32<"SMRDLiteralOffset",
                                           NamedMatchClass<"SMRDLiteralOffset">> {
   let OperandType = "OPERAND_IMMEDIATE";
 }
 
 class SMRD_Real_Load_IMM_ci <bits<5> op, SM_Load_Pseudo ps> :
   SM_Real<ps>,
   Enc64 {
 
-  let AssemblerPredicates = [isCIOnly];
-  let DecoderNamespace = "CI";
+  let AssemblerPredicates = [isGFX7];
+  let DecoderNamespace = "GFX7";
   let InOperandList = (ins ps.BaseClass:$sbase, smrd_literal_offset:$offset, GLC:$glc);
 
   let LGKM_CNT = ps.LGKM_CNT;
   let SMRD = ps.SMRD;
   let mayLoad = ps.mayLoad;
   let mayStore = ps.mayStore;
   let hasSideEffects = ps.hasSideEffects;
   let SchedRW = ps.SchedRW;

 def S_BUFFER_LOAD_DWORDX8_IMM_ci  : SMRD_Real_Load_IMM_ci <0x0b, S_BUFFER_LOAD_DWORDX8_IMM>;
 def S_BUFFER_LOAD_DWORDX16_IMM_ci : SMRD_Real_Load_IMM_ci <0x0c, S_BUFFER_LOAD_DWORDX16_IMM>;
 
 class SMRD_Real_ci <bits<5> op, SM_Pseudo ps>
   : SM_Real<ps>
   , SIMCInstr<ps.PseudoInstr, SIEncodingFamily.SI>
   , Enc32 {
 
-  let AssemblerPredicates = [isCIOnly];
-  let DecoderNamespace = "CI";
+  let AssemblerPredicates = [isGFX7];
+  let DecoderNamespace = "GFX7";
 
   let Inst{7-0}   = !if(ps.has_offset, offset{7-0}, ?);
   let Inst{8}     = imm;
   let Inst{14-9}  = !if(ps.has_sbase, sbase{6-1}, ?);
   let Inst{21-15} = !if(ps.has_sdst, sdst{6-0}, ?);
   let Inst{26-22} = op;
   let Inst{31-27} = 0x18; //encoding
 }
def : GCNPat <
     (smrd_load (SMRDImm i64:$sbase, i32:$offset)),
     (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, 0))
   >;
 
   // 2. 32-bit IMM offset on CI
   def : GCNPat <
     (smrd_load (SMRDImm32 i64:$sbase, i32:$offset)),
     (vt (!cast<InstSI>(Instr#"_IMM_ci") $sbase, $offset, 0))> {
-    let OtherPredicates = [isCIOnly];
+    let OtherPredicates = [isGFX7];
   }
 
   // 3. SGPR offset
   def : GCNPat <
     (smrd_load (SMRDSgpr i64:$sbase, i32:$offset)),
     (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, 0))
   >;
 

 
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORD",     f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX2",   v2f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX4",   v4f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX8",   v8f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX16",  v16f32>;
 } // End let AddedComplexity = 100
 
-let OtherPredicates = [isSICI] in {
+let OtherPredicates = [isGFX6GFX7] in {
 def : GCNPat <
   (i64 (readcyclecounter)),
   (S_MEMTIME)
 >;
 }
 
 let OtherPredicates = [isVI] in {
 
 def : GCNPat <
   (i64 (readcyclecounter)),
   (S_MEMREALTIME)
 >;
 
 } // let OtherPredicates = [isVI]

llvm/trunk/lib/Target/AMDGPU/SOPInstructions.td

 def S_MOVRELS_B32 : SOP1_32 <"s_movrels_b32">;
 def S_MOVRELS_B64 : SOP1_64 <"s_movrels_b64">;
 def S_MOVRELD_B32 : SOP1_32 <"s_movreld_b32">;
 def S_MOVRELD_B64 : SOP1_64 <"s_movreld_b64">;
 } // End Uses = [M0]
 
 def S_CBRANCH_JOIN : SOP1_0_32R <"s_cbranch_join">;
 def S_MOV_REGRD_B32 : SOP1_32 <"s_mov_regrd_b32">;
+
 let Defs = [SCC] in {
 def S_ABS_I32 : SOP1_32 <"s_abs_i32">;
 } // End Defs = [SCC]
 def S_MOV_FED_B32 : SOP1_32 <"s_mov_fed_b32">;
 
 let SubtargetPredicate = HasVGPRIndexMode in {
 def S_SET_GPR_IDX_IDX : SOP1_0_32<"s_set_gpr_idx_idx"> {
   let Uses = [M0];
   let Defs = [M0];
 }
 }
 
-let SubtargetPredicate = isGFX9 in {
+let SubtargetPredicate = isGFX9Plus in {
   let hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] in {
     def S_ANDN1_SAVEEXEC_B64 : SOP1_64<"s_andn1_saveexec_b64">;
     def S_ORN1_SAVEEXEC_B64  : SOP1_64<"s_orn1_saveexec_b64">;
     def S_ANDN1_WREXEC_B64   : SOP1_64<"s_andn1_wrexec_b64">;
     def S_ANDN2_WREXEC_B64   : SOP1_64<"s_andn2_wrexec_b64">;
   } // End hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC]
 
   def S_BITREPLICATE_B64_B32 : SOP1_64_32<"s_bitreplicate_b64_b32">;
-} // End SubtargetPredicate = isGFX9
+} // End SubtargetPredicate = isGFX9Plus
 
 //===----------------------------------------------------------------------===//
 // SOP2 Instructions
 //===----------------------------------------------------------------------===//
 
 class SOP2_Pseudo<string opName, dag outs, dag ins,
                   string asmOps, list<dag> pattern=[]> :
   SOP_Pseudo<opName, outs, ins, asmOps, pattern> {
def S_RFE_RESTORE_B64 : SOP2_Pseudo <
     (ins SSrc_b64:$src0, SSrc_b32:$src1),
     "$src0, $src1"
   > {
     let hasSideEffects = 1;
     let has_sdst = 0;
   }
 }
 
-let SubtargetPredicate = isGFX9 in {
+let SubtargetPredicate = isGFX9Plus in {
   def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">;
   def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">;
   def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">;
 
   let Defs = [SCC] in {
     def S_LSHL1_ADD_U32 : SOP2_32<"s_lshl1_add_u32">;
     def S_LSHL2_ADD_U32 : SOP2_32<"s_lshl2_add_u32">;
     def S_LSHL3_ADD_U32 : SOP2_32<"s_lshl3_add_u32">;
     def S_LSHL4_ADD_U32 : SOP2_32<"s_lshl4_add_u32">;
   } // End Defs = [SCC]
 
   let isCommutable = 1 in {
     def S_MUL_HI_U32 : SOP2_32<"s_mul_hi_u32",
       [(set i32:$sdst, (UniformBinFrag<mulhu> SSrc_b32:$src0, SSrc_b32:$src1))]>;
     def S_MUL_HI_I32 : SOP2_32<"s_mul_hi_i32",
       [(set i32:$sdst, (UniformBinFrag<mulhs> SSrc_b32:$src0, SSrc_b32:$src1))]>;
   }
-}
+} // End SubtargetPredicate = isGFX9Plus
 
 //===----------------------------------------------------------------------===//
 // SOPK Instructions
 //===----------------------------------------------------------------------===//
 
 class SOPK_Pseudo <string opName, dag outs, dag ins,
                    string asmOps, list<dag> pattern=[]> :
   InstSI <outs, ins, "", pattern>,
def S_SETREG_IMM32_B32 : SOPK_Pseudo <
   (outs), (ins i32imm:$imm, hwreg:$simm16),
   "$simm16, $imm"> {
   let Size = 8; // Unlike every other SOPK instruction.
   let has_sdst = 0;
 }
 
 } // End hasSideEffects = 1
 
-let SubtargetPredicate = isGFX9 in {
+let SubtargetPredicate = isGFX9Plus in {
   def S_CALL_B64 : SOPK_Pseudo<
       "s_call_b64",
       (outs SReg_64:$sdst),
       (ins s16imm:$simm16),
       "$sdst, $simm16"> {
     let isCall = 1;
   }
-}
+} // End SubtargetPredicate = isGFX9Plus
 
 //===----------------------------------------------------------------------===//
 // SOPC Instructions
 //===----------------------------------------------------------------------===//
 
 class SOPCe <bits<7> op> : Enc32 {
   bits<8> src0;
   bits<8> src1;

 def S_CMP_LE_U32 : SOPC_CMP_32 <0x0b, "s_cmp_le_u32", COND_ULE, "s_cmp_ge_u32">;
 
 def S_BITCMP0_B32 : SOPC_32 <0x0c, "s_bitcmp0_b32">;
 def S_BITCMP1_B32 : SOPC_32 <0x0d, "s_bitcmp1_b32">;
 def S_BITCMP0_B64 : SOPC_64_32 <0x0e, "s_bitcmp0_b64">;
 def S_BITCMP1_B64 : SOPC_64_32 <0x0f, "s_bitcmp1_b64">;
 def S_SETVSKIP : SOPC_32 <0x10, "s_setvskip">;
 
-let SubtargetPredicate = isVI in {
+let SubtargetPredicate = isGFX8Plus in {
 def S_CMP_EQ_U64 : SOPC_CMP_64 <0x12, "s_cmp_eq_u64", COND_EQ>;
 def S_CMP_LG_U64 : SOPC_CMP_64 <0x13, "s_cmp_lg_u64", COND_NE>;
-}
+} // End SubtargetPredicate = isGFX8Plus
 
 let SubtargetPredicate = HasVGPRIndexMode in {
 def S_SET_GPR_IDX_ON : SOPC <0x11,
   (outs),
   (ins SSrc_b32:$src0, GPRIdxMode:$src1),
   "s_set_gpr_idx_on $src0,$src1"> {
   let Defs = [M0]; // No scc def
   let Uses = [M0]; // Other bits of m0 unmodified.

 
 let isTerminator = 1 in {
 
 def S_ENDPGM : SOPP <0x00000001, (ins EndpgmImm:$simm16), "s_endpgm $simm16"> {
   let isBarrier = 1;
   let isReturn = 1;
 }
 
-let SubtargetPredicate = isVI in {
 def S_ENDPGM_SAVED : SOPP <0x0000001B, (ins), "s_endpgm_saved"> {
+  let SubtargetPredicate = isGFX8Plus;
   let simm16 = 0;
   let isBarrier = 1;
   let isReturn = 1;
 }
-}
 
-let SubtargetPredicate = isGFX9 in {
+let SubtargetPredicate = isGFX9Plus in {
   let isBarrier = 1, isReturn = 1, simm16 = 0 in {
     def S_ENDPGM_ORDERED_PS_DONE :
       SOPP<0x01e, (ins), "s_endpgm_ordered_ps_done">;
   } // End isBarrier = 1, isReturn = 1, simm16 = 0
-} // End SubtargetPredicate = isGFX9
+} // End SubtargetPredicate = isGFX9Plus
 
 let isBranch = 1, SchedRW = [WriteBranch] in {
 def S_BRANCH : SOPP <
   0x00000002, (ins sopp_brtarget:$simm16), "s_branch $simm16",
   [(br bb:$simm16)]> {
   let isBarrier = 1;
 }
 
def S_BARRIER : SOPP <0x0000000a, (ins), "s_barrier",
   [(int_amdgcn_s_barrier)]> {
   let SchedRW = [WriteBarrier];
   let simm16 = 0;
   let mayLoad = 1;
   let mayStore = 1;
   let isConvergent = 1;
 }
 
-let SubtargetPredicate = isVI in {
 def S_WAKEUP : SOPP <0x00000003, (ins), "s_wakeup"> {
+  let SubtargetPredicate = isGFX8Plus;
   let simm16 = 0;
   let mayLoad = 1;
   let mayStore = 1;
 }
-}
 
 let mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
 def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "s_waitcnt $simm16">;
 def S_SETHALT : SOPP <0x0000000d, (ins i16imm:$simm16), "s_sethalt $simm16">;
 def S_SETKILL : SOPP <0x0000000b, (ins i16imm:$simm16), "s_setkill $simm16">;
 
 // On SI the documentation says sleep for approximately 64 * low 2
 // bits, consistent with the reported maximum of 448. On VI the

 
 def : GCNPat <
   (int_amdgcn_s_waitcnt i32:$simm16),
   (S_WAITCNT (as_i16imm $simm16))
 >;
 
 
 //===----------------------------------------------------------------------===//
-// Real target instructions, move this to the appropriate subtarget TD file
+// Target-specific instruction encodings.
 //===----------------------------------------------------------------------===//
 
 class Select_si<string opName> :
   SIMCInstr<opName, SIEncodingFamily.SI> {
-  list<Predicate> AssemblerPredicates = [isSICI];
-  string DecoderNamespace = "SICI";
+  list<Predicate> AssemblerPredicates = [isGFX6GFX7];
+  string DecoderNamespace = "GFX6GFX7";
 }
 
 class SOP1_Real_si<bits<8> op, SOP1_Pseudo ps> :
   SOP1_Real<op, ps>,
   Select_si<ps.Mnemonic>;
 
 class SOP2_Real_si<bits<7> op, SOP2_Pseudo ps> :
   SOP2_Real<op, ps>,

llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td

def V_READFIRSTLANE_B32 :
 
   let Inst{8-0}   = src0;
   let Inst{16-9}  = 0x2;
   let Inst{24-17} = vdst;
   let Inst{31-25} = 0x3f; //encoding
 }
 
 let SchedRW = [WriteQuarterRate32] in {
-defm V_CVT_I32_F64 : VOP1Inst <"v_cvt_i32_f64", VOP_I32_F64, fp_to_sint>;
+defm V_CVT_I32_F64 : VOP1Inst <"v_cvt_i32_f64", VOP_I32_F64,  fp_to_sint>;
 defm V_CVT_F64_I32 : VOP1Inst <"v_cvt_f64_i32", VOP1_F64_I32, sint_to_fp>;
+defm V_CVT_F32_F64 : VOP1Inst <"v_cvt_f32_f64", VOP_F32_F64,  fpround>;
+defm V_CVT_F64_F32 : VOP1Inst <"v_cvt_f64_f32", VOP_F64_F32,  fpextend>;
+defm V_CVT_U32_F64 : VOP1Inst <"v_cvt_u32_f64", VOP_I32_F64,  fp_to_uint>;
+defm V_CVT_F64_U32 : VOP1Inst <"v_cvt_f64_u32", VOP1_F64_I32, uint_to_fp>;
 defm V_CVT_F32_I32 : VOP1Inst <"v_cvt_f32_i32", VOP1_F32_I32, sint_to_fp>;
 defm V_CVT_F32_U32 : VOP1Inst <"v_cvt_f32_u32", VOP1_F32_I32, uint_to_fp>;
 defm V_CVT_U32_F32 : VOP1Inst <"v_cvt_u32_f32", VOP_I32_F32, fp_to_uint>;
 defm V_CVT_I32_F32 : VOP1Inst <"v_cvt_i32_f32", VOP_I32_F32, fp_to_sint>;
 let FPDPRounding = 1 in {
 defm V_CVT_F16_F32 : VOP1Inst <"v_cvt_f16_f32", VOP_F16_F32, fpround>;
 } // End FPDPRounding = 1
 defm V_CVT_F32_F16 : VOP1Inst <"v_cvt_f32_f16", VOP_F32_F16, fpextend>;
 defm V_CVT_RPI_I32_F32 : VOP1Inst <"v_cvt_rpi_i32_f32", VOP_I32_F32, cvt_rpi_i32_f32>;
 defm V_CVT_FLR_I32_F32 : VOP1Inst <"v_cvt_flr_i32_f32", VOP_I32_F32, cvt_flr_i32_f32>;
 defm V_CVT_OFF_F32_I4 : VOP1Inst  <"v_cvt_off_f32_i4", VOP1_F32_I32>;
-defm V_CVT_F32_F64 : VOP1Inst <"v_cvt_f32_f64", VOP_F32_F64, fpround>;
-defm V_CVT_F64_F32 : VOP1Inst <"v_cvt_f64_f32", VOP_F64_F32, fpextend>;
-defm V_CVT_U32_F64 : VOP1Inst <"v_cvt_u32_f64", VOP_I32_F64, fp_to_uint>;
-defm V_CVT_F64_U32 : VOP1Inst <"v_cvt_f64_u32", VOP1_F64_I32, uint_to_fp>;
 } // End SchedRW = [WriteQuarterRate32]
 
 defm V_CVT_F32_UBYTE0 : VOP1Inst <"v_cvt_f32_ubyte0", VOP1_F32_I32, AMDGPUcvt_f32_ubyte0>;
 defm V_CVT_F32_UBYTE1 : VOP1Inst <"v_cvt_f32_ubyte1", VOP1_F32_I32, AMDGPUcvt_f32_ubyte1>;
 defm V_CVT_F32_UBYTE2 : VOP1Inst <"v_cvt_f32_ubyte2", VOP1_F32_I32, AMDGPUcvt_f32_ubyte2>;
 defm V_CVT_F32_UBYTE3 : VOP1Inst <"v_cvt_f32_ubyte3", VOP1_F32_I32, AMDGPUcvt_f32_ubyte3>;
 
 defm V_FRACT_F32 : VOP1Inst <"v_fract_f32", VOP_F32_F32, AMDGPUfract>;

  // so this must have an implicit def of the register added to it.
 defm V_MOVRELD_B32 : VOP1Inst <"v_movreld_b32", VOP_MOVRELD>;
 defm V_MOVRELS_B32 : VOP1Inst <"v_movrels_b32", VOP_I32_VI32_NO_EXT>;
 defm V_MOVRELSD_B32 : VOP1Inst <"v_movrelsd_b32", VOP_NO_EXT<VOP_I32_I32>>;
 } // End Uses = [M0, EXEC]
 
 defm V_MOV_FED_B32 : VOP1Inst <"v_mov_fed_b32", VOP_I32_I32>;
 
-// These instruction only exist on SI and CI
-let SubtargetPredicate = isSICI in {
-
+let SubtargetPredicate = isGFX6GFX7 in {
   let SchedRW = [WriteQuarterRate32] in {
-defm V_LOG_CLAMP_F32 : VOP1Inst <"v_log_clamp_f32", VOP_F32_F32, int_amdgcn_log_clamp>;
-defm V_RCP_CLAMP_F32 : VOP1Inst <"v_rcp_clamp_f32", VOP_F32_F32>;
-defm V_RCP_LEGACY_F32 : VOP1Inst <"v_rcp_legacy_f32", VOP_F32_F32, AMDGPUrcp_legacy>;
-defm V_RSQ_CLAMP_F32 : VOP1Inst <"v_rsq_clamp_f32", VOP_F32_F32, AMDGPUrsq_clamp>;
-defm V_RSQ_LEGACY_F32 : VOP1Inst <"v_rsq_legacy_f32", VOP_F32_F32, AMDGPUrsq_legacy>;
+    defm V_LOG_CLAMP_F32 :
+      VOP1Inst<"v_log_clamp_f32", VOP_F32_F32, int_amdgcn_log_clamp>;
+    defm V_RCP_CLAMP_F32 :
+      VOP1Inst<"v_rcp_clamp_f32", VOP_F32_F32>;
+    defm V_RCP_LEGACY_F32 :
+      VOP1Inst<"v_rcp_legacy_f32", VOP_F32_F32, AMDGPUrcp_legacy>;
+    defm V_RSQ_CLAMP_F32 :
+      VOP1Inst<"v_rsq_clamp_f32", VOP_F32_F32, AMDGPUrsq_clamp>;
+    defm V_RSQ_LEGACY_F32 :
+      VOP1Inst<"v_rsq_legacy_f32", VOP_F32_F32, AMDGPUrsq_legacy>;
   } // End SchedRW = [WriteQuarterRate32]
 
   let SchedRW = [WriteDouble] in {
-defm V_RCP_CLAMP_F64 : VOP1Inst <"v_rcp_clamp_f64", VOP_F64_F64>;
-defm V_RSQ_CLAMP_F64 : VOP1Inst <"v_rsq_clamp_f64", VOP_F64_F64, AMDGPUrsq_clamp>;
+    defm V_RCP_CLAMP_F64 :
+      VOP1Inst<"v_rcp_clamp_f64", VOP_F64_F64>;
+    defm V_RSQ_CLAMP_F64 :
+      VOP1Inst<"v_rsq_clamp_f64", VOP_F64_F64, AMDGPUrsq_clamp>;
   } // End SchedRW = [WriteDouble]
+} // End SubtargetPredicate = isGFX6GFX7
 
-} // End SubtargetPredicate = isSICI
-
-
-let SubtargetPredicate = isCIVI in {
+let SubtargetPredicate = isGFX7GFX8GFX9 in {
+  let SchedRW = [WriteQuarterRate32] in {
+    defm V_LOG_LEGACY_F32 : VOP1Inst<"v_log_legacy_f32", VOP_F32_F32>;
+    defm V_EXP_LEGACY_F32 : VOP1Inst<"v_exp_legacy_f32", VOP_F32_F32>;
+  } // End SchedRW = [WriteQuarterRate32]
+} // End SubtargetPredicate = isGFX7GFX8GFX9
 
+let SubtargetPredicate = isGFX7Plus in {
   let SchedRW = [WriteDoubleAdd] in {
-defm V_TRUNC_F64 : VOP1Inst <"v_trunc_f64", VOP_F64_F64, ftrunc>;
-defm V_CEIL_F64 : VOP1Inst <"v_ceil_f64", VOP_F64_F64, fceil>;
-defm V_FLOOR_F64 : VOP1Inst <"v_floor_f64", VOP_F64_F64, ffloor>;
-defm V_RNDNE_F64 : VOP1Inst <"v_rndne_f64", VOP_F64_F64, frint>;
+    defm V_TRUNC_F64 : VOP1Inst<"v_trunc_f64", VOP_F64_F64, ftrunc>;
+    defm V_CEIL_F64  : VOP1Inst<"v_ceil_f64", VOP_F64_F64, fceil>;
+    defm V_RNDNE_F64 : VOP1Inst<"v_rndne_f64", VOP_F64_F64, frint>;
+    defm V_FLOOR_F64 : VOP1Inst<"v_floor_f64", VOP_F64_F64, ffloor>;
   } // End SchedRW = [WriteDoubleAdd]
-
-let SchedRW = [WriteQuarterRate32] in {
-defm V_LOG_LEGACY_F32 : VOP1Inst <"v_log_legacy_f32", VOP_F32_F32>;
-defm V_EXP_LEGACY_F32 : VOP1Inst <"v_exp_legacy_f32", VOP_F32_F32>;
-} // End SchedRW = [WriteQuarterRate32]
-
-} // End SubtargetPredicate = isCIVI
-
+} // End SubtargetPredicate = isGFX7Plus
 
 let SubtargetPredicate = Has16BitInsts in {
 
 let FPDPRounding = 1 in {
 defm V_CVT_F16_U16 : VOP1Inst <"v_cvt_f16_u16", VOP1_F16_I16, uint_to_fp>;
 defm V_CVT_F16_I16 : VOP1Inst <"v_cvt_f16_i16", VOP1_F16_I16, sint_to_fp>;
 } // End FPDPRounding = 1
 defm V_CVT_U16_F16 : VOP1Inst <"v_cvt_u16_f16", VOP_I16_F16, fp_to_uint>;
def VOP_SWAP_I32 : VOPProfile<[i32, i32, i32, untyped]> {
   let Outs32 = (outs VGPR_32:$vdst, VGPR_32:$vdst1);
   let Ins32 = (ins VGPR_32:$src0, VGPR_32:$src1);
   let Outs64 = Outs32;
   let Asm32 = " $vdst, $src0";
   let Asm64 = "";
   let Ins64 = (ins);
 }
 
-let SubtargetPredicate = isGFX9 in {
-  let Constraints = "$vdst = $src1, $vdst1 = $src0",
-      DisableEncoding="$vdst1,$src1",
-      SchedRW = [Write64Bit, Write64Bit] in {
-// Never VOP3. Takes as long as 2 v_mov_b32s
-def V_SWAP_B32 : VOP1_Pseudo <"v_swap_b32", VOP_SWAP_I32, [], 1>;
+let SubtargetPredicate = isGFX9Plus in {
+  def V_SWAP_B32 : VOP1_Pseudo<"v_swap_b32", VOP_SWAP_I32, [], 1> {
+    let Constraints = "$vdst = $src1, $vdst1 = $src0";
+    let DisableEncoding = "$vdst1,$src1";
+    let SchedRW = [Write64Bit, Write64Bit];
   }
 
-defm V_SCREEN_PARTITION_4SE_B32 : VOP1Inst <"v_screen_partition_4se_b32", VOP_I32_I32>;
-
   defm V_SAT_PK_U8_I16    : VOP1Inst<"v_sat_pk_u8_i16", VOP_I32_I32>;
   defm V_CVT_NORM_I16_F16 : VOP1Inst<"v_cvt_norm_i16_f16", VOP_I16_F16>;
   defm V_CVT_NORM_U16_F16 : VOP1Inst<"v_cvt_norm_u16_f16", VOP_I16_F16>;
+} // End SubtargetPredicate = isGFX9Plus
 
+let SubtargetPredicate = isGFX9 in {
+  defm V_SCREEN_PARTITION_4SE_B32 : VOP1Inst <"v_screen_partition_4se_b32", VOP_I32_I32>;
 } // End SubtargetPredicate = isGFX9
 
 //===----------------------------------------------------------------------===//
 // Target
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 // SI
 //===----------------------------------------------------------------------===//
 
 multiclass VOP1_Real_si <bits<9> op> {
-  let AssemblerPredicates = [isSICI], DecoderNamespace = "SICI" in {
+  let AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7" in {
     def _e32_si :
       VOP1_Real<!cast<VOP1_Pseudo>(NAME#"_e32"), SIEncodingFamily.SI>,
       VOP1e<op{7-0}, !cast<VOP1_Pseudo>(NAME#"_e32").Pfl>;
     def _e64_si :
       VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
       VOP3e_si <{1, 1, op{6-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
   }
 }

 defm V_MOVRELS_B32       : VOP1_Real_si <0x43>;
 defm V_MOVRELSD_B32      : VOP1_Real_si <0x44>;
 
 //===----------------------------------------------------------------------===//
 // CI
 //===----------------------------------------------------------------------===//
 
 multiclass VOP1_Real_ci <bits<9> op> {
-  let AssemblerPredicates = [isCIOnly], DecoderNamespace = "CI" in {
+  let AssemblerPredicates = [isCIOnly], DecoderNamespace = "GFX7" in {
     def _e32_ci :
       VOP1_Real<!cast<VOP1_Pseudo>(NAME#"_e32"), SIEncodingFamily.SI>,
       VOP1e<op{7-0}, !cast<VOP1_Pseudo>(NAME#"_e32").Pfl>;
     def _e64_ci :
       VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
       VOP3e_si <{1, 1, op{6-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
   }
 }
 
 defm V_TRUNC_F64         : VOP1_Real_ci <0x17>;
 defm V_CEIL_F64          : VOP1_Real_ci <0x18>;
 defm V_FLOOR_F64         : VOP1_Real_ci <0x1A>;
 defm V_RNDNE_F64         : VOP1_Real_ci <0x19>;
 defm V_LOG_LEGACY_F32    : VOP1_Real_ci <0x45>;
 defm V_EXP_LEGACY_F32    : VOP1_Real_ci <0x46>;
 
 //===----------------------------------------------------------------------===//
-// VI
+// GFX8, GFX9 (VI).
 //===----------------------------------------------------------------------===//
 
 class VOP1_DPPe <bits<8> op, VOP1_DPP_Pseudo ps, VOPProfile P = ps.Pfl> :
   VOP_DPPe <P> {
   bits<8> vdst;
   let Inst{8-0}   = 0xfa; // dpp
   let Inst{16-9}  = op;
   let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0);

 def : GCNPat <
   (i32 (int_amdgcn_update_dpp i32:$old, i32:$src, imm:$dpp_ctrl, imm:$row_mask,
                       imm:$bank_mask, imm:$bound_ctrl)),
   (V_MOV_B32_dpp $old, $src, (as_i32imm $dpp_ctrl),
                        (as_i32imm $row_mask), (as_i32imm $bank_mask),
                        (as_i1imm $bound_ctrl))
 >;
 
+} // End OtherPredicates = [isVI]
+
+let OtherPredicates = [isGFX8Plus] in {
 def : GCNPat<
   (i32 (anyext i16:$src)),
   (COPY $src)
 >;
 
 def : GCNPat<
    (i64 (anyext i16:$src)),
    (REG_SEQUENCE VReg_64,
      (i32 (COPY $src)), sub0,
      (V_MOV_B32_e32 (i32 0)), sub1)
 >;
 
 def : GCNPat<
   (i16 (trunc i32:$src)),
   (COPY $src)
 >;
 
 def : GCNPat <
   (i16 (trunc i64:$src)),
   (EXTRACT_SUBREG $src, sub0)
 >;
 
-} // End OtherPredicates = [isVI]
+} // End OtherPredicates = [isGFX8Plus]
 
 //===----------------------------------------------------------------------===//
 // GFX9
 //===----------------------------------------------------------------------===//
 
 multiclass VOP1_Real_gfx9 <bits<10> op> {
   let AssemblerPredicates = [isGFX9], DecoderNamespace = "GFX9" in {
     defm NAME : VOP1_Real_e32e64_vi <op>;

llvm/trunk/lib/Target/AMDGPU/VOP2Instructions.td

def : GCNPat<
     (V_ADDC_U32_e64 $src0, $src1, $src2, 0)
 >;
 
 def : GCNPat<
     (AMDGPUsube i32:$src0, i32:$src1, i1:$src2),
     (V_SUBB_U32_e64 $src0, $src1, $src2, 0)
 >;
 
-// These instructions only exist on SI and CI
-let SubtargetPredicate = isSICI, Predicates = [isSICI] in {
-
+let SubtargetPredicate = isGFX6GFX7 in {
 defm V_MIN_LEGACY_F32 : VOP2Inst <"v_min_legacy_f32", VOP_F32_F32_F32, AMDGPUfmin_legacy>;
 defm V_MAX_LEGACY_F32 : VOP2Inst <"v_max_legacy_f32", VOP_F32_F32_F32, AMDGPUfmax_legacy>;
 
 let isCommutable = 1 in {
 defm V_MAC_LEGACY_F32 : VOP2Inst <"v_mac_legacy_f32", VOP_F32_F32_F32>;
 defm V_LSHR_B32 : VOP2Inst <"v_lshr_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, srl>;
 defm V_ASHR_I32 : VOP2Inst <"v_ashr_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, sra>;
 defm V_LSHL_B32 : VOP2Inst <"v_lshl_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, shl>;
 } // End isCommutable = 1
-
-} // End let SubtargetPredicate = SICI, Predicates = [isSICI]
+} // End SubtargetPredicate = isGFX6GFX7
 
 class DivergentBinOp<SDPatternOperator Op, VOP_Pseudo Inst> :
   GCNPat<
       (getDivergentFrag<Op>.ret Inst.Pfl.Src0VT:$src0, Inst.Pfl.Src1VT:$src1),
       !if(!cast<Commutable_REV>(Inst).IsOrig,
         (Inst $src0, $src1),
         (Inst $src1, $src0)
       )

 
 class ZExt_i16_i1_Pat <SDNode ext> : GCNPat <
   (i16 (ext i1:$src)),
   (V_CNDMASK_B32_e64 (i32 0/*src0mod*/), (i32 0/*src0*/),
                      (i32 0/*src1mod*/), (i32 1/*src1*/),
                      $src)
 >;
 
-let Predicates = [Has16BitInsts] in {
+let Predicates = [Has16BitInsts, isGFX7GFX8GFX9] in {
 
 defm : Arithmetic_i16_Pats<add, V_ADD_U16_e64>;
 defm : Arithmetic_i16_Pats<mul, V_MUL_LO_U16_e64>;
 defm : Arithmetic_i16_Pats<sub, V_SUB_U16_e64>;
 defm : Arithmetic_i16_Pats<smin, V_MIN_I16_e64>;
 defm : Arithmetic_i16_Pats<smax, V_MAX_I16_e64>;
 defm : Arithmetic_i16_Pats<umin, V_MIN_U16_e64>;
 defm : Arithmetic_i16_Pats<umax, V_MAX_U16_e64>;

 // Undo sub x, c -> add x, -c canonicalization since c is more likely
 // an inline immediate than -c.
 // TODO: Also do for 64-bit.
 def : GCNPat<
   (add i16:$src0, (i16 NegSubInlineConst16:$src1)),
   (V_SUB_U16_e64 $src0, NegSubInlineConst16:$src1)
 >;
 
-} // End Predicates = [Has16BitInsts]
+} // End Predicates = [Has16BitInsts, isGFX7GFX8GFX9]
 
 //===----------------------------------------------------------------------===//
 // SI
 //===----------------------------------------------------------------------===//
 
-let AssemblerPredicates = [isSICI], DecoderNamespace = "SICI" in {
+let AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7" in {
 
 multiclass VOP2_Real_si <bits<6> op> {
   def _si :
     VOP2_Real<!cast<VOP2_Pseudo>(NAME), SIEncodingFamily.SI>,
     VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME).Pfl>;
 }
 
 multiclass VOP2_Real_MADK_si <bits<6> op> {

 }
 
 multiclass VOP2be_Real_e32e64_si <bits<6> op> : VOP2_Real_e32_si<op> {
   def _e64_si :
     VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
     VOP3be_si <{1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
 }
 
-} // End AssemblerPredicates = [isSICI], DecoderNamespace = "SICI"
+} // End AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7"
 
 defm V_CNDMASK_B32        : VOP2_Real_e32e64_si <0x0>;
 defm V_ADD_F32            : VOP2_Real_e32e64_si <0x3>;
 defm V_SUB_F32            : VOP2_Real_e32e64_si <0x4>;
 defm V_SUBREV_F32         : VOP2_Real_e32e64_si <0x5>;
 defm V_MUL_LEGACY_F32     : VOP2_Real_e32e64_si <0x7>;
 defm V_MUL_F32            : VOP2_Real_e32e64_si <0x8>;
 defm V_MUL_I32_I24        : VOP2_Real_e32e64_si <0x9>;

llvm/trunk/lib/Target/AMDGPU/VOP3Instructions.td

 def V_MQSAD_PK_U16_U8 : VOP3Inst <"v_mqsad_pk_u16_u8", VOP3_Profile<VOP_I64_I64_I32_I64, VOP3_CLAMP>>;
 } // End Constraints = "@earlyclobber $vdst"
 
 def V_TRIG_PREOP_F64 : VOP3Inst <"v_trig_preop_f64", VOP3_Profile<VOP_F64_F64_I32>, AMDGPUtrig_preop> {
   let SchedRW = [WriteDouble];
 }
 
 let SchedRW = [Write64Bit] in {
-// These instructions only exist on SI and CI
-let SubtargetPredicate = isSICI, Predicates = [isSICI] in {
+let SubtargetPredicate = isGFX6GFX7, Predicates = [isGFX6GFX7] in {
 def V_LSHL_B64 : VOP3Inst <"v_lshl_b64", VOP3_Profile<VOP_PAT_GEN<VOP_I64_I64_I32>>, shl>;
 def V_LSHR_B64 : VOP3Inst <"v_lshr_b64", VOP3_Profile<VOP_PAT_GEN<VOP_I64_I64_I32>>, srl>;
 def V_ASHR_I64 : VOP3Inst <"v_ashr_i64", VOP3_Profile<VOP_PAT_GEN<VOP_I64_I64_I32>>, sra>;
 def V_MULLIT_F32 : VOP3Inst <"v_mullit_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
-} // End SubtargetPredicate = isSICI, Predicates = [isSICI]
+} // End SubtargetPredicate = isGFX6GFX7, Predicates = [isGFX6GFX7]
 
-let SubtargetPredicate = isVI in {
+let SubtargetPredicate = isGFX8Plus in {
 def V_LSHLREV_B64 : VOP3Inst <"v_lshlrev_b64", VOP3_Profile<VOP_I64_I32_I64>>;
 def V_LSHRREV_B64 : VOP3Inst <"v_lshrrev_b64", VOP3_Profile<VOP_I64_I32_I64>>;
 def V_ASHRREV_I64 : VOP3Inst <"v_ashrrev_i64", VOP3_Profile<VOP_I64_I32_I64>>;
-} // End SubtargetPredicate = isVI
+} // End SubtargetPredicate = isGFX8Plus
 } // End SchedRW = [Write64Bit]
 
 let Predicates = [isVI] in {
 def : GCNPat <
  (getDivergentFrag<shl>.ret i64:$x, i32:$y),
  (V_LSHLREV_B64 $y, $x)
 >;
 def : AMDGPUPat <
  (getDivergentFrag<srl>.ret i64:$x, i32:$y),
  (V_LSHRREV_B64 $y, $x)
 >;
 def : AMDGPUPat <
  (getDivergentFrag<sra>.ret i64:$x, i32:$y),
  (V_ASHRREV_I64 $y, $x)
 >;
 }
 
 
-let SubtargetPredicate = isCIVI in {
+let SchedRW = [Write32Bit] in {
+let SubtargetPredicate = isGFX8Plus in {
+def V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUperm>;
+} // End SubtargetPredicate = isGFX8Plus
+} // End SchedRW = [Write32Bit]
+
+let SubtargetPredicate = isGFX7Plus in {
 
 let Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32] in {
 def V_QSAD_PK_U16_U8 : VOP3Inst <"v_qsad_pk_u16_u8", VOP3_Profile<VOP_I64_I64_I32_I64, VOP3_CLAMP>>;
 def V_MQSAD_U32_U8 : VOP3Inst <"v_mqsad_u32_u8", VOP3_Profile<VOP_V4I32_I64_I32_V4I32, VOP3_CLAMP>>;
 } // End Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32]
 
 let isCommutable = 1 in {
 let SchedRW = [WriteQuarterRate32, WriteSALU] in {
 def V_MAD_U64_U32 : VOP3Inst <"v_mad_u64_u32", VOP3b_I64_I1_I32_I32_I64>;
 def V_MAD_I64_I32 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>;
 } // End SchedRW = [WriteDouble, WriteSALU]
 } // End isCommutable = 1
 
-} // End SubtargetPredicate = isCIVI
+} // End SubtargetPredicate = isGFX7Plus
 
 
 def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, AMDGPUdiv_fixup> {
   let Predicates = [Has16BitInsts, isVIOnly];
   let FPDPRounding = 1;
 }
 def V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9",
                                       VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUdiv_fixup> {
   let renamedInGFX9 = 1;
-  let Predicates = [Has16BitInsts, isGFX9];
+  let Predicates = [Has16BitInsts, isGFX9Plus];
   let FPDPRounding = 1;
 }
 
 def V_FMA_F16 : VOP3Inst <"v_fma_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, fma> {
   let Predicates = [Has16BitInsts, isVIOnly];
   let FPDPRounding = 1;
 }
 def V_FMA_F16_gfx9 : VOP3Inst <"v_fma_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, fma> {
   let renamedInGFX9 = 1;
-  let Predicates = [Has16BitInsts, isGFX9];
+  let Predicates = [Has16BitInsts, isGFX9Plus];
   let FPDPRounding = 1;
 }
 
 let SubtargetPredicate = Has16BitInsts, isCommutable = 1 in {
 
 let renamedInGFX9 = 1 in {
 def V_MAD_U16 : VOP3Inst <"v_mad_u16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_CLAMP>>;
 def V_MAD_I16 : VOP3Inst <"v_mad_i16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_CLAMP>>;

 } // End Uses = [M0, EXEC]
 } // End FPDPRounding = 1
 } // End renamedInGFX9 = 1
 
 let SubtargetPredicate = isGFX9 in {
 def V_MAD_F16_gfx9   : VOP3Inst <"v_mad_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>> {
   let FPDPRounding = 1;
 }
+} // End SubtargetPredicate = isGFX9
+
+let SubtargetPredicate = isGFX9Plus in {
 def V_MAD_U16_gfx9   : VOP3Inst <"v_mad_u16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>>;
 def V_MAD_I16_gfx9   : VOP3Inst <"v_mad_i16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>>;
 def V_INTERP_P2_F16_gfx9 : VOP3Interp <"v_interp_p2_f16_gfx9", VOP3_INTERP16<[f16, f32, i32, f32]>>;
-} // End SubtargetPredicate = isGFX9
+} // End SubtargetPredicate = isGFX9Plus
 
 let Uses = [M0, EXEC], FPDPRounding = 1 in {
 def V_INTERP_P1LL_F16 : VOP3Interp <"v_interp_p1ll_f16", VOP3_INTERP16<[f32, f32, i32, untyped]>,
        [(set f32:$vdst, (AMDGPUinterp_p1ll_f16 f32:$src0, (i32 imm:$attrchan),
                                                           (i32 imm:$attr),
                                                           (i32 imm:$src0_modifiers),
                                                           (i1 imm:$high),
                                                           (i1 imm:$clamp),

 } // End Uses = [M0, EXEC], FPDPRounding = 1
 
 } // End SubtargetPredicate = Has16BitInsts, isCommutable = 1
 
 let SubtargetPredicate = isVI in {
 def V_INTERP_P1_F32_e64  : VOP3Interp <"v_interp_p1_f32", VOP3_INTERP>;
 def V_INTERP_P2_F32_e64  : VOP3Interp <"v_interp_p2_f32", VOP3_INTERP>;
 def V_INTERP_MOV_F32_e64 : VOP3Interp <"v_interp_mov_f32", VOP3_INTERP_MOV>;
-
-def V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUperm>;
 } // End SubtargetPredicate = isVI
 
 let Predicates = [Has16BitInsts] in {
 
 multiclass Ternary_i16_Pats <SDPatternOperator op1, SDPatternOperator op2,
                              Instruction inst, SDPatternOperator op3> {
 def : GCNPat <
   (op2 (op1 i16:$src0, i16:$src1), i16:$src2),
[{
     }
 
     return true;
   }]
 > {
   let PredicateCodeUsesOperands = 1;
 }
 
-let SubtargetPredicate = isGFX9 in {
+let SubtargetPredicate = isGFX9Plus in {
 def V_PACK_B32_F16 : VOP3Inst <"v_pack_b32_f16", VOP3_Profile<VOP_B32_F16_F16, VOP3_OPSEL>>;
 def V_LSHL_ADD_U32 : VOP3Inst <"v_lshl_add_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 def V_ADD_LSHL_U32 : VOP3Inst <"v_add_lshl_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 def V_ADD3_U32 : VOP3Inst <"v_add3_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 def V_LSHL_OR_B32 : VOP3Inst <"v_lshl_or_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 def V_AND_OR_B32 : VOP3Inst <"v_and_or_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 def V_OR3_B32 : VOP3Inst <"v_or3_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
 

 def : ThreeOp_i32_Pats<shl, add, V_LSHL_ADD_U32>;
 def : ThreeOp_i32_Pats<add, shl, V_ADD_LSHL_U32>;
 def : ThreeOp_i32_Pats<add, add, V_ADD3_U32>;
 def : ThreeOp_i32_Pats<shl, or, V_LSHL_OR_B32>;
 def : ThreeOp_i32_Pats<and, or, V_AND_OR_B32>;
 def : ThreeOp_i32_Pats<or, or, V_OR3_B32>;
 def : ThreeOp_i32_Pats<xor, add, V_XAD_U32>;
 
-} // End SubtargetPredicate = isGFX9
+} // End SubtargetPredicate = isGFX9Plus
 
 //===----------------------------------------------------------------------===//
 // Integer Clamp Patterns
 //===----------------------------------------------------------------------===//
 
 class getClampPat<VOPProfile P, SDPatternOperator node> {
   dag ret3 = (P.DstVT (node P.Src0VT:$src0, P.Src1VT:$src1, P.Src2VT:$src2));
   dag ret2 = (P.DstVT (node P.Src0VT:$src0, P.Src1VT:$src1));

 def : IntClampPat<V_SAD_U16, int_amdgcn_sad_u16>;
 
 def : IntClampPat<V_MSAD_U8, int_amdgcn_msad_u8>;
 def : IntClampPat<V_MQSAD_PK_U16_U8, int_amdgcn_mqsad_pk_u16_u8>;
 
 def : IntClampPat<V_QSAD_PK_U16_U8, int_amdgcn_qsad_pk_u16_u8>;
 def : IntClampPat<V_MQSAD_U32_U8, int_amdgcn_mqsad_u32_u8>;
 
+
 //===----------------------------------------------------------------------===//
-// Target
+// Target-specific instruction encodings.
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
-// SI
+// GFX6, GFX7.
 //===----------------------------------------------------------------------===//
 
-let AssemblerPredicates = [isSICI], DecoderNamespace = "SICI" in {
+let AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7" in {
 
 multiclass VOP3_Real_si<bits<9> op> {
   def _si : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.SI>,
             VOP3e_si <op, !cast<VOP3_Pseudo>(NAME).Pfl>;
 }
 
 multiclass VOP3be_Real_si<bits<9> op> {
   def _si : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.SI>,
             VOP3be_si <op, !cast<VOP3_Pseudo>(NAME).Pfl>;
 }
 
-} // End AssemblerPredicates = [isSICI], DecoderNamespace = "SICI"
+} // End AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7"
 
 defm V_MAD_LEGACY_F32   : VOP3_Real_si <0x140>;
 defm V_MAD_F32          : VOP3_Real_si <0x141>;
 defm V_MAD_I32_I24      : VOP3_Real_si <0x142>;
 defm V_MAD_U32_U24      : VOP3_Real_si <0x143>;
 defm V_CUBEID_F32       : VOP3_Real_si <0x144>;
 defm V_CUBESC_F32       : VOP3_Real_si <0x145>;
 defm V_CUBETC_F32       : VOP3_Real_si <0x146>;

 defm V_DIV_SCALE_F64    : VOP3be_Real_si <0x16e>;
 defm V_DIV_FMAS_F32     : VOP3_Real_si <0x16f>;
 defm V_DIV_FMAS_F64     : VOP3_Real_si <0x170>;
 defm V_MSAD_U8          : VOP3_Real_si <0x171>;
 defm V_MQSAD_PK_U16_U8  : VOP3_Real_si <0x173>;
 defm V_TRIG_PREOP_F64   : VOP3_Real_si <0x174>;
 
 //===----------------------------------------------------------------------===//
-// CI
+// GFX7.
 //===----------------------------------------------------------------------===//
 
 multiclass VOP3_Real_ci<bits<9> op> {
   def _ci : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.SI>,
             VOP3e_si <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
     let AssemblerPredicates = [isCIOnly];
-    let DecoderNamespace = "CI";
+    let DecoderNamespace = "GFX7";
   }
 }
 
 multiclass VOP3be_Real_ci<bits<9> op> {
   def _ci : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.SI>,
             VOP3be_si <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
     let AssemblerPredicates = [isCIOnly];
-    let DecoderNamespace = "CI";
+    let DecoderNamespace = "GFX7";
   }
 }
 
 defm V_QSAD_PK_U16_U8   : VOP3_Real_ci <0x172>;
 defm V_MQSAD_U32_U8     : VOP3_Real_ci <0x175>;
 defm V_MAD_U64_U32      : VOP3be_Real_ci <0x176>;
 defm V_MAD_I64_I32      : VOP3be_Real_ci <0x177>;
 
 //===----------------------------------------------------------------------===//
-// VI
+// GFX8, GFX9 (VI).
 //===----------------------------------------------------------------------===//
 
 let AssemblerPredicates = [isVI], DecoderNamespace = "VI" in {
 
 multiclass VOP3_Real_vi<bits<10> op> {
   def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME), SIEncodingFamily.VI>,
             VOP3e_vi <op, !cast<VOP_Pseudo>(NAME).Pfl>;
 }

llvm/trunk/lib/Target/AMDGPU/VOPCInstructions.td

 defm V_CMPX_NGE_F64 : VOPCX_F64 <"v_cmpx_nge_f64", "v_cmpx_nle_f64">;
 defm V_CMPX_NLG_F64 : VOPCX_F64 <"v_cmpx_nlg_f64">;
 defm V_CMPX_NGT_F64 : VOPCX_F64 <"v_cmpx_ngt_f64", "v_cmpx_nlt_f64">;
 defm V_CMPX_NLE_F64 : VOPCX_F64 <"v_cmpx_nle_f64">;
 defm V_CMPX_NEQ_F64 : VOPCX_F64 <"v_cmpx_neq_f64">;
 defm V_CMPX_NLT_F64 : VOPCX_F64 <"v_cmpx_nlt_f64">;
 defm V_CMPX_TRU_F64 : VOPCX_F64 <"v_cmpx_tru_f64">;
 
-let SubtargetPredicate = isSICI in {
+let SubtargetPredicate = isGFX6GFX7 in {
 
 defm V_CMPS_F_F32 : VOPC_F32 <"v_cmps_f_f32">;
 defm V_CMPS_LT_F32 : VOPC_F32 <"v_cmps_lt_f32", COND_NULL, "v_cmps_gt_f32">;
 defm V_CMPS_EQ_F32 : VOPC_F32 <"v_cmps_eq_f32">;
 defm V_CMPS_LE_F32 : VOPC_F32 <"v_cmps_le_f32", COND_NULL, "v_cmps_ge_f32">;
 defm V_CMPS_GT_F32 : VOPC_F32 <"v_cmps_gt_f32">;
 defm V_CMPS_LG_F32 : VOPC_F32 <"v_cmps_lg_f32">;
 defm V_CMPS_GE_F32 : VOPC_F32 <"v_cmps_ge_f32">;

 defm V_CMPSX_NGE_F64 : VOPCX_F64 <"v_cmpsx_nge_f64", "v_cmpsx_nle_f64">;
 defm V_CMPSX_NLG_F64 : VOPCX_F64 <"v_cmpsx_nlg_f64">;
 defm V_CMPSX_NGT_F64 : VOPCX_F64 <"v_cmpsx_ngt_f64", "v_cmpsx_nlt_f64">;
 defm V_CMPSX_NLE_F64 : VOPCX_F64 <"v_cmpsx_nle_f64">;
 defm V_CMPSX_NEQ_F64 : VOPCX_F64 <"v_cmpsx_neq_f64">;
 defm V_CMPSX_NLT_F64 : VOPCX_F64 <"v_cmpsx_nlt_f64">;
 defm V_CMPSX_TRU_F64 : VOPCX_F64 <"v_cmpsx_tru_f64">;
 
-} // End SubtargetPredicate = isSICI
+} // End SubtargetPredicate = isGFX6GFX7
 
 let SubtargetPredicate = Has16BitInsts in {
 
 defm V_CMP_F_F16    : VOPC_F16 <"v_cmp_f_f16">;
 defm V_CMP_LT_F16   : VOPC_F16 <"v_cmp_lt_f16", COND_OLT, "v_cmp_gt_f16">;
 defm V_CMP_EQ_F16   : VOPC_F16 <"v_cmp_eq_f16", COND_OEQ>;
 defm V_CMP_LE_F16   : VOPC_F16 <"v_cmp_le_f16", COND_OLE, "v_cmp_ge_f16">;
 defm V_CMP_GT_F16   : VOPC_F16 <"v_cmp_gt_f16", COND_OGT>;

 def : FCMP_Pattern <COND_UEQ, V_CMP_NLG_F16_e64, f16>;
 def : FCMP_Pattern <COND_UNE, V_CMP_NEQ_F16_e64, f16>;
 def : FCMP_Pattern <COND_UGT, V_CMP_NLE_F16_e64, f16>;
 def : FCMP_Pattern <COND_UGE, V_CMP_NLT_F16_e64, f16>;
 def : FCMP_Pattern <COND_ULT, V_CMP_NGE_F16_e64, f16>;
 def : FCMP_Pattern <COND_ULE, V_CMP_NGT_F16_e64, f16>;
 
 //===----------------------------------------------------------------------===//
-// Target
+// Target-specific instruction encodings.
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 // SI
 //===----------------------------------------------------------------------===//
 
 multiclass VOPC_Real_si <bits<9> op> {
-  let AssemblerPredicates = [isSICI], DecoderNamespace = "SICI" in {
+  let AssemblerPredicates = [isGFX6GFX7], DecoderNamespace = "GFX6GFX7" in {
     def _e32_si :
       VOPC_Real<!cast<VOPC_Pseudo>(NAME#"_e32"), SIEncodingFamily.SI>,
       VOPCe<op{7-0}>;
 
     def _e64_si :
       VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
       VOP3a_si <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl> {
       // Encoding used for VOPC instructions encoded as VOP3
       // Differs from VOP3e by destination name (sdst) as VOPC doesn't have vector dst
       bits<8> sdst;
       let Inst{7-0} = sdst;
     }
   }
   def : VOPCInstAlias <!cast<VOP3_Pseudo>(NAME#"_e64"),
                        !cast<Instruction>(NAME#"_e32_si")> {
-    let AssemblerPredicate = isSICI;
+    let AssemblerPredicate = isGFX6GFX7;
   }
 }
 
 defm V_CMP_F_F32     : VOPC_Real_si <0x0>;
 defm V_CMP_LT_F32    : VOPC_Real_si <0x1>;
 defm V_CMP_EQ_F32    : VOPC_Real_si <0x2>;
 defm V_CMP_LE_F32    : VOPC_Real_si <0x3>;
 defm V_CMP_GT_F32    : VOPC_Real_si <0x4>;