Implement TypeSig generator in `Intrinsics.td`

[chapuni]

Summary

This commit doesn't replace IntrinsicEmitter::ComputeFixedEncoding(),
but compares outputs to it, to make sure implementation correct.

Depends on D145871, D145872, D145874, and D146914

[chapuni]

• Use cast<ty>
• Graft deps

[arsenm]

llvm/utils/TableGen/CodeGenIntrinsics.cpp

123

Leftover dead code?

[chapuni]

llvm/utils/TableGen/CodeGenIntrinsics.cpp

123

Two reason.

• I would like to notify to downstreams with this change.
• I would like to leave the history for someone to visit.

I could squash D145937 into this.

[chapuni]

• Rebase

[arsenm]

llvm/include/llvm/IR/Intrinsics.td

185

I don't know what Idx means here in relation to the constant 0x100

215

Should be on previous line

607

Ditto

llvm/utils/TableGen/CodeGenIntrinsics.cpp

123

I don't expect the encoding of intrinsics to be a meaningful place for downstream users. Ok as long as the OLD stuff is removed in the next patch

[chapuni]

• s/DecArgCode/ResolveArgCode/
• Update ArgCode encoders
• Reformat typelist

[chapuni]

llvm/include/llvm/IR/Intrinsics.td

185

I have rewritten more descriptive.

215

I have reindented typelist deeper since I was afraid it would cause less readable.

[chapuni]

• Fix isOverloaded (intrinsic-pointer-to-any.td was failing)

[chapuni]

• Rebase

[chapuni]

• Avoid D145872 -- dynamic subscript

[chapuni]

I have introduced ListIntAt, ListLLVMTypeAt, and ListListIntAt.
I will rewind them when D145872 (dynamic subscript) is landed.

llvm/include/llvm/IR/Intrinsics.td

def ArgKind {
   int AnyInteger = 1;
   int AnyFloat   = 2;
   int AnyVector  = 3;
   int AnyPointer = 4;
 
   int MatchType  = 7;
 }
 
+// (Idx << 3) | ArgCode
+class EncAnyType<int ArgCode=0> {
+  int ret = !or(0x100, ArgCode);
+}
+
+// (Mapping[Num] << 3) | MatchType
+class EncMatchType<int Num=0> {
+  int ret = !or(0x200, Num);
+}
+
+// (Mapping[Num] << 3) | ArgCodes[Mapping[Num]]
+class EncSameWidth<int Num=0> {
+  int ret = !or(0x300, Num);
+}
+
+// ArgNo
+class EncNextArgA<int dummy=0> {
+  int ret = !or(0x400, dummy);
+}
+
+// Mapping[Num]
+class EncNextArgN<int Num=0> {
+  int ret = !or(0x500, Num);
+}
+
+class DecArgCode<
+  list<int> Mapping,
+  list<int> ArgCodes,
+  int ACIdx,
+  int ax
+> {
+  int ah = !and(ax, 0xFF00);
+  int al = !and(ax, 0x00FF);
+  int num = Mapping[al];
+  int ret = !cond(
+    !eq(ah, EncAnyType<>.ret)   : !or(!shl(ACIdx, 3), al),
+    !eq(ah, EncMatchType<>.ret) : !or(!shl(num, 3), ArgKind.MatchType),
+    !eq(ah, EncSameWidth<>.ret) : !or(!shl(num, 3), ArgCodes[num]),
+    !eq(ah, EncNextArgA<>.ret)  : ACIdx,
+    !eq(ah, EncNextArgN<>.ret)  : num,
+    true : al);
+}
+
 //===----------------------------------------------------------------------===//
 // IIT_Info
 //===----------------------------------------------------------------------===//
 
 class IIT_Base<int num> {
   int Number = num;
   list<ValueType> VTs = ?;
 }

 def IIT_V3 : IIT_Vec<3, 53>;
 def IIT_EXTERNREF : IIT_VT<externref, 54>;
 def IIT_FUNCREF : IIT_VT<funcref, 55>;
 def IIT_ANYPTR_TO_ELT : IIT_Base<56>;
 def IIT_I2 : IIT_Int<2, 57>;
 def IIT_I4 : IIT_Int<4, 58>;
 }
 
+defvar IIT_all_FixedTypes = !filter(iit, IIT_all,
+  !or(!isa<IIT_VT>(iit), !isa<IIT_Int>(iit)));
+
+defvar IIT_all_VectorTypes = !filter(iit, IIT_all,
+  !isa<IIT_Vec>(iit));
+
+defvar IIT_RetNumbers = [
+  [IIT_Done.Number],
+  []<int>,
+  [IIT_STRUCT2.Number],
+  [IIT_STRUCT3.Number],
+  [IIT_STRUCT4.Number],
+  [IIT_STRUCT5.Number],
+  [IIT_STRUCT6.Number],
+  [IIT_STRUCT7.Number],
+  [IIT_STRUCT8.Number],
+  [IIT_STRUCT9.Number],
+];
+
 //===----------------------------------------------------------------------===//
 // Types used by intrinsics.
 //===----------------------------------------------------------------------===//
 
 class LLVMType<ValueType vt> {
   ValueType VT = vt;
   int isAny = vt.isOverloaded;
+
+  int ArgCode = ?;
+  int Number = ?;
+
+  list<IIT_Base> IITs = !filter(iit, IIT_all_FixedTypes,
+    !not(!empty(!filter(iit_vt, iit.VTs,
+      !eq(iit_vt, !if(vt.isVector, vt.ElementType, vt))))));
+  assert !le(!size(IITs), 1), "Duplicate type";
+
+  list<IIT_Base> IIT_Vecs = !if(vt.isVector,
+    !filter(iit, IIT_all_VectorTypes,
+      !not(!empty(!filter(iit_vt, iit.VTs, !and(
+        !eq(iit_vt.ElementType, vt.ElementType),
+        !eq(iit_vt.nElem, vt.nElem)))))),
+    []);
+  assert !le(!size(IIT_Vecs), 1), "Duplicate type";
+
+  list<int> Sig = !listconcat(
+    !if(vt.isScalable, [IIT_SCALABLE_VEC.Number], []),
+    !foreach(iit, IIT_Vecs, iit.Number),
+    !foreach(iit, IITs,     iit.Number));
+}
+
+class LLVMAnyType<ValueType vt> : LLVMType<vt> {
+  let ArgCode = !cond(
+    !eq(vt, Any)     : ArgKind.Any,
+    !eq(vt, iAny)    : ArgKind.AnyInteger,
+    !eq(vt, fAny)    : ArgKind.AnyFloat,
+    !eq(vt, vAny)    : ArgKind.AnyVector,
+    !eq(vt, iPTRAny) : ArgKind.AnyPointer,
+  );
+  let Sig = [
+    IIT_ARG.Number,
+    EncAnyType<ArgCode>.ret,
+  ];
+
+  assert isAny, "LLVMAnyType.VT should have isOverloaded";
 }
 
 class LLVMQualPointerType<LLVMType elty, int addrspace>
   : LLVMType<iPTR>{
   LLVMType ElTy = elty;
   int AddrSpace = addrspace;
+  assert !and(!le(0, addrspace), !le(addrspace, 255)),
+    "Address space exceeds 255";
+
+  // D63507: LLVMPointerType<llvm_any_ty>
+  let isAny = elty.isAny;
+
+  let Sig = !listconcat(
+    !if(addrspace, [
+      IIT_ANYPTR.Number,
+      addrspace,
+    ], [
+      IIT_PTR.Number,
+    ]),
+    ElTy.Sig);
 }
 
 class LLVMPointerType<LLVMType elty>
   : LLVMQualPointerType<elty, 0>;
 
 class LLVMAnyPointerType<LLVMType elty>
-  : LLVMType<iPTRAny>{
+  : LLVMAnyType<iPTRAny> {
   LLVMType ElTy = elty;
 
   assert isAny, "iPTRAny should have isOverloaded";
 }
 
 // Match the type of another intrinsic parameter.  Number is an index into the
 // list of overloaded types for the intrinsic, excluding all the fixed types.
 // The Number value must refer to a previously listed type.  For example:
 //   Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_anyfloat_ty, LLVMMatchType<0>]>
 // has two overloaded types, the 2nd and 3rd arguments.  LLVMMatchType<0>
 // refers to the first overloaded type, which is the 2nd argument.
-class LLVMMatchType<int num>
+class LLVMMatchType<int num, IIT_Base IIT_Info = IIT_ARG>
   : LLVMType<OtherVT>{
-  int Number = num;
+  let Number = num;
+  let Sig = [
+    IIT_Info.Number,
+    EncMatchType<num>.ret,
+  ];
+}
+
+class LLVMMatchTypeNextArg<int num, IIT_Base IIT_Info>
+  : LLVMMatchType<num, IIT_Info> {
+  let Sig = [
+    IIT_Info.Number,
+    EncNextArgA<>.ret,
+    EncNextArgN<num>.ret,
+  ];
 }
 
 // Match the type of another intrinsic parameter that is expected to be based on
 // an integral type (i.e. either iN or <N x iM>), but change the scalar size to
 // be twice as wide or half as wide as the other type.  This is only useful when
 // the intrinsic is overloaded, so the matched type should be declared as iAny.
-class LLVMExtendedType<int num> : LLVMMatchType<num>;
-class LLVMTruncatedType<int num> : LLVMMatchType<num>;
+class LLVMExtendedType<int num> : LLVMMatchType<num, IIT_EXTEND_ARG>;
+class LLVMTruncatedType<int num> : LLVMMatchType<num, IIT_TRUNC_ARG>;
 
 // Match the scalar/vector of another intrinsic parameter but with a different
 // element type. Either both are scalars or both are vectors with the same
 // number of elements.
 class LLVMScalarOrSameVectorWidth<int idx, LLVMType elty>
-  : LLVMMatchType<idx> {
+  : LLVMMatchType<idx, IIT_SAME_VEC_WIDTH_ARG> {
   ValueType ElTy = elty.VT;
+  let Sig = !listconcat([
+    IIT_SAME_VEC_WIDTH_ARG.Number,
+    EncSameWidth<idx>.ret,
+  ], elty.Sig);
 }
 
-class LLVMPointerTo<int num> : LLVMMatchType<num>;
-class LLVMPointerToElt<int num> : LLVMMatchType<num>;
-class LLVMAnyPointerToElt<int num> : LLVMMatchType<num>;
-class LLVMVectorOfAnyPointersToElt<int num> : LLVMMatchType<num>;
-class LLVMVectorElementType<int num> : LLVMMatchType<num>;
+class LLVMPointerTo<int num> : LLVMMatchType<num, IIT_PTR_TO_ARG>;
+class LLVMPointerToElt<int num> : LLVMMatchType<num, IIT_PTR_TO_ELT>;
+class LLVMAnyPointerToElt<int num>
+  : LLVMMatchTypeNextArg<num, IIT_ANYPTR_TO_ELT>;
+class LLVMVectorOfAnyPointersToElt<int num>
+  : LLVMMatchTypeNextArg<num, IIT_VEC_OF_ANYPTRS_TO_ELT>;
+class LLVMVectorElementType<int num> : LLVMMatchType<num, IIT_VEC_ELEMENT>;
 
 // Match the type of another intrinsic parameter that is expected to be a
 // vector type, but change the element count to be half as many.
-class LLVMHalfElementsVectorType<int num> : LLVMMatchType<num>;
+class LLVMHalfElementsVectorType<int num>
+  : LLVMMatchType<num, IIT_HALF_VEC_ARG>;
 
 // Match the type of another intrinsic parameter that is expected to be a
 // vector type (i.e. <N x iM>) but with each element subdivided to
 // form a vector with more elements that are smaller than the original.
-class LLVMSubdivide2VectorType<int num> : LLVMMatchType<num>;
-class LLVMSubdivide4VectorType<int num> : LLVMMatchType<num>;
+class LLVMSubdivide2VectorType<int num>
+  : LLVMMatchType<num, IIT_SUBDIVIDE2_ARG>;
+class LLVMSubdivide4VectorType<int num>
+  : LLVMMatchType<num, IIT_SUBDIVIDE4_ARG>;
 
 // Match the element count and bit width of another intrinsic parameter, but
 // change the element type to an integer.
-class LLVMVectorOfBitcastsToInt<int num> : LLVMMatchType<num>;
+class LLVMVectorOfBitcastsToInt<int num>
+  : LLVMMatchType<num, IIT_VEC_OF_BITCASTS_TO_INT>;
 
 def llvm_void_ty       : LLVMType<isVoid>;
 
-def llvm_any_ty        : LLVMType<Any>;
-def llvm_anyint_ty     : LLVMType<iAny>;
-def llvm_anyfloat_ty   : LLVMType<fAny>;
-def llvm_anyvector_ty  : LLVMType<vAny>;
+def llvm_any_ty        : LLVMAnyType<Any>;
+def llvm_anyint_ty     : LLVMAnyType<iAny>;
+def llvm_anyfloat_ty   : LLVMAnyType<fAny>;
+def llvm_anyvector_ty  : LLVMAnyType<vAny>;
 
 def llvm_i1_ty         : LLVMType<i1>;
 def llvm_i8_ty         : LLVMType<i8>;
 def llvm_i16_ty        : LLVMType<i16>;
 def llvm_i32_ty        : LLVMType<i32>;
 def llvm_i64_ty        : LLVMType<i64>;
 def llvm_i128_ty       : LLVMType<i128>;
 def llvm_half_ty       : LLVMType<f16>;

 def llvm_v16f64_ty     : LLVMType<v16f64>;   // 16 x double
 
 def llvm_vararg_ty     : LLVMType<isVoid>;   // this means vararg here
 
 def llvm_externref_ty  : LLVMType<externref>;
 def llvm_funcref_ty    : LLVMType<funcref>;
 
 //===----------------------------------------------------------------------===//
+
+class MakeIdx<list<int> Set> {
+  list<int> IdxsR = !foreach(i, !range(Set),
+    !if(Set[i],
+      !foldl(0, !range(0, i), m, j, !add(m, Set[j])),
+      -1));
+
+  list<int> RIdxsR = !foreach(i, !range(Set),
+    !foldl(-1, !range(Set), m, j,
+      !if(!and(Set[j], !eq(IdxsR[j], i)), j, m)));
+
+  list<int> Idxs  = !foreach(a, IdxsR,  !if(!ge(a, 0), a, ?));
+  list<int> RIdxs = !foreach(a, RIdxsR, !if(!ge(a, 0), a, ?));
+}
+
+class TypeInfoGen<
+  list<LLVMType> RetTypes,
+  list<LLVMType> ParamTypes
+> {
+  list<LLVMType> AllTypes = !listconcat(RetTypes, ParamTypes);
+
+  // ArgCodes for NextArg -- isAny or MatchTypeNextArg
+  list<int> ACIdxs = MakeIdx<
+    !foreach(ty, AllTypes,
+      !or(ty.isAny, !isa<LLVMMatchTypeNextArg>(ty)))>.Idxs;
+
+  // ArgCodes (only for isAny or MatchTypeNextArg)
+  list<LLVMType> ACTys = !filter(ty, AllTypes,
+    !or(ty.isAny, !isa<LLVMMatchTypeNextArg>(ty)));
+
+  list<int> ArgCodes = !foreach(ty, ACTys, ty.ArgCode);
+
+  // Mappings MatchTypeIdx to ACTys
+  list<int> MappingRIdxs = MakeIdx<
+    !foreach(ty, ACTys, ty.isAny)>.RIdxs;
+
+  list<LLVMType> Types = !foreach(ty, AllTypes,
+    !if(!isa<LLVMMatchType>(ty), ACTys[MappingRIdxs[ty.Number]], ty));
+
+  list<list<int>> TypeSig = !listconcat(
+    [IIT_RetNumbers[!size(RetTypes)]],
+    !foreach(i, !range(AllTypes),
+      !foreach(a, AllTypes[i].Sig,
+        DecArgCode<
+          MappingRIdxs,
+          ArgCodes,
+          ACIdxs[i],
+          a>.ret)));
+}
+
+//===----------------------------------------------------------------------===//
 // Intrinsic Definitions.
 //===----------------------------------------------------------------------===//
 
 // Intrinsic class - This is used to define one LLVM intrinsic.  The name of the
 // intrinsic definition should start with "int_", then match the LLVM intrinsic
 // name with the "llvm." prefix removed, and all "."s turned into "_"s.  For
 // example, llvm.bswap.i16 -> int_bswap_i16.
 //
class Intrinsic<list<LLVMType> ret_types,
   list<IntrinsicProperty> IntrProperties = intr_properties;
   let Properties = sd_properties;
 
   // Disable applying IntrinsicProperties that are marked default with
   // IntrinsicProperty<1>
   bit DisableDefaultAttributes = disable_default_attributes;
 
   bit isTarget = false;
+
+  TypeInfoGen TypeInfo = TypeInfoGen<RetTypes, ParamTypes>;
+  bit isOverloaded = !gt(!size(TypeInfo.MappingRIdxs), 0);
+  list<LLVMType> Types = TypeInfo.Types;
+  list<list<int>> TypeSig = TypeInfo.TypeSig;
 }
 
 // Intrinsic with default attributes (disable_default_attributes = false).
 class DefaultAttrsIntrinsic<list<LLVMType> ret_types,
                 list<LLVMType> param_types = [],
                 list<IntrinsicProperty> intr_properties = [],
                 string name = "",
                 list<SDNodeProperty> sd_properties = []>

llvm/utils/TableGen/CodeGenIntrinsics.cpp

 //===- CodeGenIntrinsics.cpp - Intrinsic Class Wrapper --------------------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
CodeGenIntrinsic::CodeGenIntrinsic(Record *R,
   if (!TargetPrefix.empty()) {
     if (Name.size() < 6 + TargetPrefix.size() ||
         Name.substr(5, 1 + TargetPrefix.size()) != (TargetPrefix + "."))
       PrintFatalError(DefLoc, "Intrinsic '" + DefName +
                                   "' does not start with 'llvm." +
                                   TargetPrefix + ".'!");
   }
 
+#define OLD 1
+
+#if OLD
   ListInit *RetTypes = R->getValueAsListInit("RetTypes");
   ListInit *ParamTypes = R->getValueAsListInit("ParamTypes");
 
   // First collate a list of overloaded types.
   std::vector<MVT::SimpleValueType> OverloadedVTs;
   for (ListInit *TypeList : {RetTypes, ParamTypes}) {
     for (unsigned i = 0, e = TypeList->size(); i != e; ++i) {
       Record *TyEl = TypeList->getElementAsRecord(i);
for (unsigned i = 0, e = TypeList->size(); i != e; ++i) {
     // Reject invalid types.
     if (VT == MVT::isVoid && i != e - 1 /*void at end means varargs*/)
       PrintFatalError(DefLoc, "Intrinsic '" + DefName +
                                   " has void in result type list!");
 
     IS.ParamVTs.push_back(VT);
     IS.ParamTypeDefs.push_back(TyEl);
   }
+#endif // OLD
+
+  if (auto *Types = R->getValue("Types")) {
+#if OLD
+    auto OrigRetVTs = std::move(IS.RetVTs);
+    auto OrigParamVTs = std::move(IS.ParamVTs);
+    auto OrigOverloaded = isOverloaded;
+    IS.RetVTs.clear();
+    IS.ParamVTs.clear();
+#endif // OLD
+
+    auto *TypeList = cast<ListInit>(Types->getValue());
+    isOverloaded = R->getValueAsBit("isOverloaded");
+
+    unsigned I = 0;
+    for (unsigned E = R->getValueAsListInit("RetTypes")->size(); I < E; ++I)
+      IS.RetVTs.push_back(
+          getValueType(TypeList->getElementAsRecord(I)->getValueAsDef("VT")));
+
+    for (unsigned E = TypeList->size(); I < E; ++I)
+      IS.ParamVTs.push_back(
+          getValueType(TypeList->getElementAsRecord(I)->getValueAsDef("VT")));
+
+#if OLD
+    assert(IS.RetVTs == OrigRetVTs);
+    assert(IS.ParamVTs == OrigParamVTs);
+    assert(OrigOverloaded == isOverloaded);
+#endif
+  }
 
   // Parse the intrinsic properties.
   ListInit *PropList = R->getValueAsListInit("IntrProperties");
   for (unsigned i = 0, e = PropList->size(); i != e; ++i) {
     Record *Property = PropList->getElementAsRecord(i);
     assert(Property->isSubClassOf("IntrinsicProperty") &&
            "Expected a property!");
 

llvm/utils/TableGen/IntrinsicEmitter.cpp

 //===- IntrinsicEmitter.cpp - Generate intrinsic information --------------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
if (Ints[i].isOverloaded)
       OS << " | (1<<" << (i+1)%8 << ')';
   }
   OS << "\n};\n\n";
   // OTable contains a true bit at the position if the intrinsic is overloaded.
   OS << "return (OTable[id/8] & (1 << (id%8))) != 0;\n";
   OS << "#endif\n\n";
 }
 
+#define OLD 1
+#if OLD
 
 // NOTE: This must be kept in synch with the copy in lib/IR/Function.cpp!
 enum IIT_Info {
   // Common values should be encoded with 0-15.
   IIT_Done = 0,
   IIT_I1 = 1,
   IIT_I8 = 2,
   IIT_I16 = 3,
case MVT::Any:
     break;
   }
 }
 
 #if defined(_MSC_VER) && !defined(__clang__)
 #pragma optimize("",on)
 #endif
 
+#endif // OLD
+
 /// ComputeFixedEncoding - If we can encode the type signature for this
 /// intrinsic into 32 bits, return it.  If not, return ~0U.
 static void ComputeFixedEncoding(const CodeGenIntrinsic &Int,
                                  std::vector<unsigned char> &TypeSig) {
+#if OLD
   std::vector<unsigned char> ArgCodes;
 
   // Add codes for any overloaded result VTs.
   unsigned int NumInserted = 0;
   SmallVector<unsigned char, 8> ArgMapping;
   for (unsigned i = 0, e = Int.IS.RetVTs.size(); i != e; ++i)
     UpdateArgCodes(Int.IS.RetTypeDefs[i], ArgCodes, NumInserted, ArgMapping);
 
else {
     for (unsigned i = 0, e = Int.IS.RetVTs.size(); i != e; ++i)
       EncodeFixedType(Int.IS.RetTypeDefs[i], ArgCodes, NextArgCode, TypeSig,
                       ArgMapping);
   }
 
   for (unsigned i = 0, e = Int.IS.ParamTypeDefs.size(); i != e; ++i)
     EncodeFixedType(Int.IS.ParamTypeDefs[i], ArgCodes, NextArgCode, TypeSig,
                     ArgMapping);
+#endif // OLD
+
+  if (auto *R = Int.TheDef->getValue("TypeSig")) {
+#if OLD
+    auto OrigTS = std::move(TypeSig);
+    TypeSig.clear();
+#endif // OLD
+    for (auto &a : cast<ListInit>(R->getValue())->getValues()) {
+      for (auto &b : cast<ListInit>(a)->getValues())
+        TypeSig.push_back(cast<IntInit>(b)->getValue());
+    }
+
+#if OLD
+    assert(TypeSig == OrigTS);
+#endif
+  }
 }
 
 static void printIITEntry(raw_ostream &OS, unsigned char X) {
   OS << (unsigned)X;
 }
 
 void IntrinsicEmitter::EmitGenerator(const CodeGenIntrinsicTable &Ints,
                                      raw_ostream &OS) {