AMDGPU: Avoid code predicates for extload PatFrags

[arsenm]

Summary

Use the MemoryVT field. This will be necessary for tablegen to
automatically handle patterns for GlobalISel.

Doesn't handle the d16 lo/hi patterns. Those are a special case since
it involvess the custom node type.

[rampitec]

LGTM

[arsenm]

r366168

lib/Target/AMDGPU/AMDGPUInstructions.td

 }]>;
 
 class FlatStoreAddress : CodePatPred<[{
   const auto AS = cast<MemSDNode>(N)->getAddressSpace();
   return AS == AMDGPUAS::FLAT_ADDRESS ||
          AS == AMDGPUAS::GLOBAL_ADDRESS;
 }]>;
 
-class AZExtLoadBase <SDPatternOperator ld_node>: PatFrag<(ops node:$ptr),
-                                              (ld_node node:$ptr), [{
-  LoadSDNode *L = cast<LoadSDNode>(N);
-  return L->getExtensionType() == ISD::ZEXTLOAD ||
-         L->getExtensionType() == ISD::EXTLOAD;
-}]>;
-
-def az_extload : AZExtLoadBase <unindexedload>;
-
-def az_extloadi8 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
-}]>;
-
-def az_extloadi16 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
-}]>;
-
-def az_extloadi32 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i32;
-}]>;
-
 class PrivateLoad <SDPatternOperator op> : LoadFrag <op>, PrivateAddress;
 class PrivateStore <SDPatternOperator op> : StoreFrag <op>, PrivateAddress;
 
 class LocalLoad <SDPatternOperator op> : LoadFrag <op>, LocalAddress;
 class LocalStore <SDPatternOperator op> : StoreFrag <op>, LocalAddress;
 
 class RegionLoad <SDPatternOperator op> : LoadFrag <op>, RegionAddress;
 class RegionStore <SDPatternOperator op> : StoreFrag <op>, RegionAddress;
 
 class GlobalLoad <SDPatternOperator op> : LoadFrag<op>, GlobalLoadAddress;
 class GlobalStore <SDPatternOperator op> : StoreFrag<op>, GlobalAddress;
 
 class FlatLoad <SDPatternOperator op> : LoadFrag <op>, FlatLoadAddress;
 class FlatStore <SDPatternOperator op> : StoreFrag <op>, FlatStoreAddress;
 
 class ConstantLoad <SDPatternOperator op> : LoadFrag <op>, ConstantAddress;
 
 
 def load_private : PrivateLoad <load>;
-def az_extloadi8_private : PrivateLoad <az_extloadi8>;
+def extloadi8_private : PrivateLoad <extloadi8>;
+def zextloadi8_private : PrivateLoad <zextloadi8>;
 def sextloadi8_private : PrivateLoad <sextloadi8>;
-def az_extloadi16_private : PrivateLoad <az_extloadi16>;
+def extloadi16_private : PrivateLoad <extloadi16>;
+def zextloadi16_private : PrivateLoad <zextloadi16>;
 def sextloadi16_private : PrivateLoad <sextloadi16>;
 
 def store_private : PrivateStore <store>;
 def truncstorei8_private : PrivateStore<truncstorei8>;
 def truncstorei16_private : PrivateStore <truncstorei16>;
 def store_hi16_private : StoreHi16 <truncstorei16>, PrivateAddress;
 def truncstorei8_hi16_private : StoreHi16<truncstorei8>, PrivateAddress;
 
 
 def load_global : GlobalLoad <load>;
 def sextloadi8_global : GlobalLoad <sextloadi8>;
-def az_extloadi8_global : GlobalLoad <az_extloadi8>;
+def extloadi8_global : GlobalLoad <extloadi8>;
+def zextloadi8_global : GlobalLoad <zextloadi8>;
 def sextloadi16_global : GlobalLoad <sextloadi16>;
-def az_extloadi16_global : GlobalLoad <az_extloadi16>;
+def extloadi16_global : GlobalLoad <extloadi16>;
+def zextloadi16_global : GlobalLoad <zextloadi16>;
 def atomic_load_global : GlobalLoad<atomic_load>;
 
 def store_global : GlobalStore <store>;
 def truncstorei8_global : GlobalStore <truncstorei8>;
 def truncstorei16_global : GlobalStore <truncstorei16>;
 def store_atomic_global : GlobalStore<atomic_store>;
 def truncstorei8_hi16_global : StoreHi16 <truncstorei8>, GlobalAddress;
 def truncstorei16_hi16_global : StoreHi16 <truncstorei16>, GlobalAddress;
def store_align8_local : Aligned8Bytes <
   (ops node:$val, node:$ptr), (store_local node:$val, node:$ptr)
 >;
 
 def store_align16_local : Aligned16Bytes <
   (ops node:$val, node:$ptr), (store_local node:$val, node:$ptr)
 >;
 
 def load_flat          : FlatLoad <load>;
-def az_extloadi8_flat  : FlatLoad <az_extloadi8>;
+def extloadi8_flat  : FlatLoad <extloadi8>;
+def zextloadi8_flat  : FlatLoad <zextloadi8>;
 def sextloadi8_flat    : FlatLoad <sextloadi8>;
-def az_extloadi16_flat : FlatLoad <az_extloadi16>;
+def extloadi16_flat : FlatLoad <extloadi16>;
+def zextloadi16_flat : FlatLoad <zextloadi16>;
 def sextloadi16_flat   : FlatLoad <sextloadi16>;
 def atomic_load_flat   : FlatLoad<atomic_load>;
 
 def store_flat         : FlatStore <store>;
 def truncstorei8_flat  : FlatStore <truncstorei8>;
 def truncstorei16_flat : FlatStore <truncstorei16>;
 def atomic_store_flat  : FlatStore <atomic_store>;
 def truncstorei8_hi16_flat  : StoreHi16<truncstorei8>, FlatStoreAddress;
 def truncstorei16_hi16_flat : StoreHi16<truncstorei16>, FlatStoreAddress;
 
 
 def constant_load : ConstantLoad<load>;
 def sextloadi8_constant : ConstantLoad <sextloadi8>;
-def az_extloadi8_constant : ConstantLoad <az_extloadi8>;
+def extloadi8_constant : ConstantLoad <extloadi8>;
+def zextloadi8_constant : ConstantLoad <zextloadi8>;
 def sextloadi16_constant : ConstantLoad <sextloadi16>;
-def az_extloadi16_constant : ConstantLoad <az_extloadi16>;
+def extloadi16_constant : ConstantLoad <extloadi16>;
+def zextloadi16_constant : ConstantLoad <zextloadi16>;
 
 
 class local_binary_atomic_op<SDNode atomic_op> :
   PatFrag<(ops node:$ptr, node:$value),
     (atomic_op node:$ptr, node:$value), [{
   return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS;
 }]>;
 

lib/Target/AMDGPU/BUFInstructions.td

 >;
 defm BUFFER_LOAD_DWORDX3 : MUBUF_Pseudo_Loads <
   "buffer_load_dwordx3", VReg_96, v3i32
 >;
 defm BUFFER_LOAD_DWORDX4 : MUBUF_Pseudo_Loads <
   "buffer_load_dwordx4", VReg_128, v4i32
 >;
 
-defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, az_extloadi8_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, extloadi8_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, zextloadi8_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_SBYTE", i32, sextloadi8_global>;
-defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, az_extloadi16_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, extloadi16_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, zextloadi16_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_SSHORT", i32, sextloadi16_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORD", i32, load_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORDX2", v2i32, load_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORDX3", v3i32, load_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORDX4", v4i32, load_global>;
 
 // This is not described in AMD documentation,
 // but 'lds' versions of these opcodes are available
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, 0)
   >;
 }
 
 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_UBYTE_ADDR64, i32, extloadi8_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_UBYTE_ADDR64, i32, zextloadi8_constant>;
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_SSHORT_ADDR64, i32, sextloadi16_constant>;
-def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, az_extloadi16_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, extloadi16_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, zextloadi16_constant>;
 
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORD_ADDR64, BUFFER_LOAD_DWORD_OFFSET, i32, atomic_load_global>;
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, BUFFER_LOAD_DWORDX2_OFFSET, i64, atomic_load_global>;
 } // 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, i1:$dlc))),
     (Instr_OFFSET $srsrc, $soffset, $offset, $glc, $slc, $tfe, $dlc)
   >;
 }
 
 let OtherPredicates = [Has16BitInsts] in {
 
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_global>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_global>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_global>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_global>;
 
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_OFFSET, i16, load_global>;
 
 } // End OtherPredicates = [Has16BitInsts]
 
 multiclass MUBUFScratchLoadPat <MUBUF_Pseudo InstrOffen,
                                 MUBUF_Pseudo InstrOffset,
                                 ValueType vt, PatFrag ld> {
multiclass MUBUFScratchLoadPat_D16 <MUBUF_Pseudo InstrOffen,
 
   def : GCNPat <
     (ld_frag (MUBUFScratchOffset v4i32:$srsrc, i32:$soffset, u16imm:$offset), vt:$in),
     (InstrOffset $srsrc, $soffset, $offset, 0, 0, 0, 0, $in)
   >;
 }
 
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SBYTE_OFFEN, BUFFER_LOAD_SBYTE_OFFSET, i32, sextloadi8_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, az_extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, zextloadi8_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SBYTE_OFFEN, BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SSHORT_OFFEN, BUFFER_LOAD_SSHORT_OFFSET, i32, sextloadi16_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, az_extloadi16_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, extloadi16_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, zextloadi16_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i16, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORD_OFFEN, BUFFER_LOAD_DWORD_OFFSET, i32, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX2_OFFEN, BUFFER_LOAD_DWORDX2_OFFSET, v2i32, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX3_OFFEN, BUFFER_LOAD_DWORDX3_OFFSET, v3i32, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX4_OFFEN, BUFFER_LOAD_DWORDX4_OFFSET, v4i32, load_private>;
 
 let OtherPredicates = [D16PreservesUnusedBits] in {
 defm : MUBUFScratchLoadPat_D16<BUFFER_LOAD_SHORT_D16_HI_OFFEN, BUFFER_LOAD_SHORT_D16_HI_OFFSET, v2i16, load_d16_hi_private>;

lib/Target/AMDGPU/EvergreenInstructions.td

 class EGPat<dag pattern, dag result> : AMDGPUPat<pattern, result> {
   let SubtargetPredicate = isEG;
 }
 
 class EGOrCaymanPat<dag pattern, dag result> : AMDGPUPat<pattern, result> {
   let SubtargetPredicate = isEGorCayman;
 }
 
-// FIXME: These are deprecated
-def az_extloadi8_local : LocalLoad <az_extloadi8>;
-def az_extloadi16_local : LocalLoad <az_extloadi16>;
-
 //===----------------------------------------------------------------------===//
 // Evergreen / Cayman store instructions
 //===----------------------------------------------------------------------===//
 
 let SubtargetPredicate = isEGorCayman in {
 
 class CF_MEM_RAT_CACHELESS <bits<6> rat_inst, bits<4> rat_id, bits<4> mask, dag ins,
                            string name, list<dag> pattern>

lib/Target/AMDGPU/FLATInstructions.td

 class FlatSignedAtomicPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt,
                      ValueType data_vt = vt> : GCNPat <
   (vt (node (FLATSignedAtomic i64:$vaddr, i16:$offset, i1:$slc), data_vt:$data)),
   (inst $vaddr, $data, $offset, $slc)
 >;
 
 let OtherPredicates = [HasFlatAddressSpace] in {
 
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, extloadi8_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, zextloadi8_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_flat, i32>;
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_flat, i16>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, extloadi8_flat, i16>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, zextloadi8_flat, i16>;
 def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_flat, i16>;
-def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, extloadi16_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, zextloadi16_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_USHORT, load_flat, i16>;
 def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_DWORD, load_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_DWORDX2, load_flat, v2i32>;
 def : FlatLoadPat <FLAT_LOAD_DWORDX3, load_flat, v3i32>;
 def : FlatLoadPat <FLAT_LOAD_DWORDX4, load_flat, v4i32>;
 
 def : FlatLoadAtomicPat <FLAT_LOAD_DWORD, atomic_load_flat, i32>;

 def : FlatLoadPat_D16 <FLAT_LOAD_SHORT_D16, load_d16_lo_flat, v2i16>;
 def : FlatLoadPat_D16 <FLAT_LOAD_SHORT_D16, load_d16_lo_flat, v2f16>;
 }
 
 } // End OtherPredicates = [HasFlatAddressSpace]
 
 let OtherPredicates = [HasFlatGlobalInsts], AddedComplexity = 10 in {
 
-def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, az_extloadi8_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, extloadi8_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, zextloadi8_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SBYTE, sextloadi8_global, i32>;
-def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, az_extloadi8_global, i16>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, extloadi8_global, i16>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, zextloadi8_global, i16>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SBYTE, sextloadi8_global, i16>;
-def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, az_extloadi16_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, extloadi16_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, zextloadi16_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SSHORT, sextloadi16_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, load_global, i16>;
 
 def : FlatLoadSignedPat <GLOBAL_LOAD_DWORD, load_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX2, load_global, v2i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX3, load_global, v3i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX4, load_global, v4i32>;
 

lib/Target/AMDGPU/R600Instructions.td

class VTX_READ <string name, dag outs, list<dag> pattern>
 
   // VTX_WORD3 (Padding)
   //
   // Inst{127-96} = 0;
 
   let VTXInst = 1;
 }
 
+// FIXME: Deprecated.
+class AZExtLoadBase <SDPatternOperator ld_node>: PatFrag<(ops node:$ptr),
+                                              (ld_node node:$ptr), [{
+  LoadSDNode *L = cast<LoadSDNode>(N);
+  return L->getExtensionType() == ISD::ZEXTLOAD ||
+         L->getExtensionType() == ISD::EXTLOAD;
+}]>;
+
+def az_extload : AZExtLoadBase <unindexedload>;
+
+def az_extloadi8 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
+}]>;
+
+def az_extloadi16 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
+}]>;
+
+def az_extloadi32 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i32;
+}]>;
+
+// FIXME: These are deprecated
+def az_extloadi8_local : LocalLoad <az_extloadi8>;
+def az_extloadi16_local : LocalLoad <az_extloadi16>;
+
 class LoadParamFrag <PatFrag load_type> : PatFrag <
   (ops node:$ptr), (load_type node:$ptr),
   [{ return isConstantLoad(cast<LoadSDNode>(N), 0) ||
             (cast<LoadSDNode>(N)->getAddressSpace() == AMDGPUAS::PARAM_I_ADDRESS); }]
 >;
 
 def vtx_id3_az_extloadi8 : LoadParamFrag<az_extloadi8>;
 def vtx_id3_az_extloadi16 : LoadParamFrag<az_extloadi16>;