AMDGPU: Split extload/zextload local load patterns

[arsenm]

Summary

This will help removing the custom load predicates, allowing the
global isel emitter to handle them.

[rampitec]

LGTM

[arsenm]

r365398

lib/Target/AMDGPU/AMDGPUInstructions.td

 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 load_local : LocalLoad <load>;
-def az_extloadi8_local : LocalLoad <az_extloadi8>;
+def extloadi8_local : LocalLoad <extloadi8>;
+def zextloadi8_local : LocalLoad <zextloadi8>;
 def sextloadi8_local : LocalLoad <sextloadi8>;
-def az_extloadi16_local : LocalLoad <az_extloadi16>;
+def extloadi16_local : LocalLoad <extloadi16>;
+def zextloadi16_local : LocalLoad <zextloadi16>;
 def sextloadi16_local : LocalLoad <sextloadi16>;
 def atomic_load_32_local : LocalLoad<atomic_load_32>;
 def atomic_load_64_local : LocalLoad<atomic_load_64>;
 
 def store_local : LocalStore <store>;
 def truncstorei8_local : LocalStore <truncstorei8>;
 def truncstorei16_local : LocalStore <truncstorei16>;
 def store_local_hi16 : StoreHi16 <truncstorei16>, LocalAddress;

lib/Target/AMDGPU/DSInstructions.td

 }
 
 class DSReadPat_D16 <DS_Pseudo inst, PatFrag frag, ValueType vt> : GCNPat <
   (frag (DS1Addr1Offset i32:$ptr, i32:$offset), vt:$in),
   (inst $ptr, (as_i16imm $offset), (i1 0), $in)
 >;
 
 defm : DSReadPat_mc <DS_READ_I8, i32, "sextloadi8_local">;
-defm : DSReadPat_mc <DS_READ_U8,  i32, "az_extloadi8_local">;
 defm : DSReadPat_mc <DS_READ_I8,  i16, "sextloadi8_local">;
-defm : DSReadPat_mc <DS_READ_U8,  i16, "az_extloadi8_local">;
+defm : DSReadPat_mc <DS_READ_U8,  i32, "extloadi8_local">;
+defm : DSReadPat_mc <DS_READ_U8,  i32, "zextloadi8_local">;
+defm : DSReadPat_mc <DS_READ_U8,  i16, "extloadi8_local">;
+defm : DSReadPat_mc <DS_READ_U8,  i16, "zextloadi8_local">;
 defm : DSReadPat_mc <DS_READ_I16, i32, "sextloadi16_local">;
 defm : DSReadPat_mc <DS_READ_I16, i32, "sextloadi16_local">;
-defm : DSReadPat_mc <DS_READ_U16, i32, "az_extloadi16_local">;
+defm : DSReadPat_mc <DS_READ_U16, i32, "extloadi16_local">;
+defm : DSReadPat_mc <DS_READ_U16, i32, "zextloadi16_local">;
 defm : DSReadPat_mc <DS_READ_U16, i16, "load_local">;
 defm : DSReadPat_mc <DS_READ_B32, i32, "load_local">;
 defm : DSReadPat_mc <DS_READ_B32, i32, "atomic_load_32_local">;
 defm : DSReadPat_mc <DS_READ_B64, i64, "atomic_load_64_local">;
 
 let AddedComplexity = 100 in {
 
 defm : DSReadPat_mc <DS_READ_B64, v2i32, "load_align8_local">;

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/SIInstrInfo.td

 def AMDGPUld_glue : SDNode <"ISD::LOAD", SDTLoad,
   [SDNPHasChain, SDNPMayLoad, SDNPMemOperand, SDNPInGlue]
 >;
 
 def AMDGPUatomic_ld_glue : SDNode <"ISD::ATOMIC_LOAD", SDTAtomicLoad,
   [SDNPHasChain, SDNPMayLoad, SDNPMemOperand, SDNPInGlue]
 >;
 
-def unindexedload_glue : PatFrag <(ops node:$ptr), (AMDGPUld_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getAddressingMode() == ISD::UNINDEXED;
-}]>;
+def unindexedload_glue : PatFrag <(ops node:$ptr), (AMDGPUld_glue node:$ptr)> {
+  let IsUnindexed = 1;
+}
 
-def load_glue : PatFrag <(ops node:$ptr), (unindexedload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getExtensionType() == ISD::NON_EXTLOAD;
-}]>;
+def load_glue : PatFrag <(ops node:$ptr), (unindexedload_glue node:$ptr)> {
+  let IsNonExtLoad = 1;
+}
 
 def atomic_load_32_glue : PatFrag<(ops node:$ptr),
   (AMDGPUatomic_ld_glue node:$ptr)> {
   let IsAtomic = 1;
   let MemoryVT = i32;
 }
 
 def atomic_load_64_glue : PatFrag<(ops node:$ptr),
   (AMDGPUatomic_ld_glue node:$ptr)> {
   let IsAtomic = 1;
   let MemoryVT = i64;
 }
 
-def extload_glue : PatFrag<(ops node:$ptr), (load_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getExtensionType() == ISD::EXTLOAD;
-}]>;
+def extload_glue : PatFrag<(ops node:$ptr), (load_glue node:$ptr)> {
+  let IsLoad = 1;
+  let IsAnyExtLoad = 1;
+}
 
 def sextload_glue : PatFrag<(ops node:$ptr), (unindexedload_glue node:$ptr), [{
   return cast<LoadSDNode>(N)->getExtensionType() == ISD::SEXTLOAD;
 }]>;
 
-def zextload_glue : PatFrag<(ops node:$ptr), (unindexedload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getExtensionType() == ISD::ZEXTLOAD;
-}]>;
+def zextload_glue : PatFrag<(ops node:$ptr), (unindexedload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let IsZeroExtLoad = 1;
+}
 
-def az_extload_glue : AZExtLoadBase <unindexedload_glue>;
+def extloadi8_glue : PatFrag<(ops node:$ptr), (extload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i8;
+}
 
-def az_extloadi8_glue : PatFrag<(ops node:$ptr), (az_extload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
-}]>;
+def zextloadi8_glue : PatFrag<(ops node:$ptr), (zextload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i8;
+}
 
-def az_extloadi16_glue : PatFrag<(ops node:$ptr), (az_extload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
-}]>;
+def extloadi16_glue : PatFrag<(ops node:$ptr), (extload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i16;
+}
 
-def sextloadi8_glue : PatFrag<(ops node:$ptr), (sextload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
-}]>;
+def zextloadi16_glue : PatFrag<(ops node:$ptr), (zextload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i16;
+}
 
-def sextloadi16_glue : PatFrag<(ops node:$ptr), (sextload_glue node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
-}]>;
+def sextloadi8_glue : PatFrag<(ops node:$ptr), (sextload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i8;
+}
+
+def sextloadi16_glue : PatFrag<(ops node:$ptr), (sextload_glue node:$ptr)> {
+  let IsLoad = 1;
+  let MemoryVT = i16;
+}
 
 def load_glue_align8 : Aligned8Bytes <
   (ops node:$ptr), (load_glue node:$ptr)
 >;
 def load_glue_align16 : Aligned16Bytes <
   (ops node:$ptr), (load_glue node:$ptr)
 >;
 
 
 def load_local_m0 : LoadFrag<load_glue>, LocalAddress;
 def sextloadi8_local_m0 : LoadFrag<sextloadi8_glue>, LocalAddress;
 def sextloadi16_local_m0 : LoadFrag<sextloadi16_glue>, LocalAddress;
-def az_extloadi8_local_m0 : LoadFrag<az_extloadi8_glue>, LocalAddress;
-def az_extloadi16_local_m0 : LoadFrag<az_extloadi16_glue>, LocalAddress;
+def extloadi8_local_m0 : LoadFrag<extloadi8_glue>, LocalAddress;
+def zextloadi8_local_m0 : LoadFrag<zextloadi8_glue>, LocalAddress;
+def extloadi16_local_m0 : LoadFrag<extloadi16_glue>, LocalAddress;
+def zextloadi16_local_m0 : LoadFrag<zextloadi16_glue>, LocalAddress;
 def load_align8_local_m0 : LoadFrag <load_glue_align8>, LocalAddress;
 def load_align16_local_m0 : LoadFrag <load_glue_align16>, LocalAddress;
 def atomic_load_32_local_m0 : LoadFrag<atomic_load_32_glue>, LocalAddress;
 def atomic_load_64_local_m0 : LoadFrag<atomic_load_64_glue>, LocalAddress;
 
 
 def AMDGPUst_glue : SDNode <"ISD::STORE", SDTStore,
   [SDNPHasChain, SDNPMayStore, SDNPMemOperand, SDNPInGlue]