Add separate experimental DXIL op table using high OpCode bit (#7947)

Implements:
https://github.com/microsoft/hlsl-specs/blob/main/proposals/0052-experimental-dxil-ops.md

This change adds support for an experimental DXIL operation table, which
provides an independent numeric space from the main OpCode set.

It adds support for indexing into OpCode tables using the high 16-bits
as the table index, and the low 16-bits as the index into the OpCode
table.

While the fundamental change could support more tables for individual
experimental features or extensions, it's currently limited to tables
`0` for `CoreOps` and `0x8000` for `ExperimentalOps` without additional
work to unlock more tables. This maps to the existing opcodes for
`CoreOps` and opcodes with the high bit set for `ExperimentalOps`.

Now, db_dxil_op_table manages dxil op lists, and DXIL ops are built
using table methods.

Author: tex3d

docs/DXIL.rst

@@ -2111,6 +2111,8 @@ Opcodes are defined on a dense range and will be provided as enum in a header fi
 .. <py::lines('OPCODES-RST')>hctdb_instrhelp.get_opcodes_rst()</py>
 .. OPCODES-RST:BEGIN
 
+Opcode Table CoreOps, id=0: Core DXIL operations
+
 === ===================================================== =======================================================================================================================================================================================================================
 ID  Name                                                  Description
 === ===================================================== =======================================================================================================================================================================================================================
@@ -3055,6 +3057,18 @@ Given width, offset:
         ushr dest, src2, offset
     }
 
+
+
+
+Opcode Table ExperimentalOps, id=32768: Experimental DXIL operations
+
+========== =============== ================
+ID         Name            Description
+========== =============== ================
+2147483648 ExperimentalNop nop does nothing
+========== =============== ================
+
+
 .. OPCODES-RST:END
 
 
@@ -3134,10 +3148,11 @@ INSTR.CREATEHANDLEIMMRANGEID                                  Local resource mus
 INSTR.DXILSTRUCTUSER                                          Dxil struct types should only be used by ExtractValue.
 INSTR.DXILSTRUCTUSEROUTOFBOUND                                Index out of bound when extract value from dxil struct types.
 INSTR.EVALINTERPOLATIONMODE                                   Interpolation mode on %0 used with eval_* instruction must be linear, linear_centroid, linear_noperspective, linear_noperspective_centroid, linear_sample or linear_noperspective_sample.
+INSTR.EXPDXILOPCODEREQUIRESEXPSM                              Use of experimental DXILOpCode requires an experimental shader model.
 INSTR.EXTRACTVALUE                                            ExtractValue should only be used on dxil struct types and cmpxchg.
 INSTR.FAILTORESLOVETGSMPOINTER                                TGSM pointers must originate from an unambiguous TGSM global variable.
 INSTR.HANDLENOTFROMCREATEHANDLE                               Resource handle should returned by createHandle.
-INSTR.ILLEGALDXILOPCODE                                       DXILOpCode must be [0..%0].  %1 specified.
+INSTR.ILLEGALDXILOPCODE                                       DXILOpCode must be valid or a supported experimental opcode.
 INSTR.ILLEGALDXILOPFUNCTION                                   '%0' is not a DXILOpFuncition for DXILOpcode '%1'.
 INSTR.IMMBIASFORSAMPLEB                                       bias amount for sample_b must be in the range [%0,%1], but %2 was specified as an immediate.
 INSTR.INBOUNDSACCESS                                          Access to out-of-bounds memory is disallowed.

include/dxc/DXIL/DxilConstants.h

@@ -490,10 +490,44 @@ inline bool IsFeedbackTexture(DXIL::ResourceKind ResourceKind) {
          ResourceKind == DXIL::ResourceKind::FeedbackTexture2DArray;
 }
 
+// clang-format off
+// Python lines need to be not formatted.
+/* <py::lines('OPCODETABLE-ENUM')>hctdb_instrhelp.get_enum_decl("OpCodeTableID")</py>*/
+// clang-format on
+// OPCODETABLE-ENUM:BEGIN
+// Enumeration for DXIL opcode tables
+enum class OpCodeTableID : unsigned {
+  CoreOps = 0,             // Core DXIL operations
+  ExperimentalOps = 32768, // Experimental DXIL operations
+};
+// OPCODETABLE-ENUM:END
+
+// clang-format off
+// Python lines need to be not formatted.
+/* <py::lines('EXTOPCODES-ENUM')>hctdb_instrhelp.get_extended_table_opcode_enum_decls()</py>*/
+// clang-format on
+// EXTOPCODES-ENUM:BEGIN
+namespace ExperimentalOps {
+static const OpCodeTableID TableID = OpCodeTableID::ExperimentalOps;
+// Enumeration for ExperimentalOps DXIL operations
+enum class OpCode : unsigned {
+  // No-op
+  ExperimentalNop = 0, // nop does nothing
+
+  NumOpCodes = 1, // exclusive last value of enumeration
+};
+} // namespace ExperimentalOps
+static const unsigned NumOpCodeTables = 2;
+// EXTOPCODES-ENUM:END
+
+#define EXP_OPCODE(feature, opcode)                                            \
+  opcode =                                                                     \
+      (((unsigned)feature::TableID << 16) | (unsigned)feature::OpCode::opcode)
+
 // TODO: change opcodes.
 /* <py::lines('OPCODE-ENUM')>hctdb_instrhelp.get_enum_decl("OpCode")</py>*/
 // OPCODE-ENUM:BEGIN
-// Enumeration for operations specified by DXIL
+// Enumeration for CoreOps DXIL operations
 enum class OpCode : unsigned {
   //
   Reserved0 = 226,   // reserved
@@ -1089,9 +1123,23 @@ enum class OpCode : unsigned {
   NumOpCodes_Dxil_1_8 = 258,
   NumOpCodes_Dxil_1_9 = 312,
 
-  NumOpCodes = 312 // exclusive last value of enumeration
+  NumOpCodes = 312,     // exclusive last value of enumeration
+  Invalid = 0xFFFFFFFF, // stable invalid OpCode value
+
+  // OpCodes for extended tables follow.
+
+  // OpCodeTableID = 32768
+  // ExperimentalOps
+  EXP_OPCODE(ExperimentalOps, ExperimentalNop), // nop does nothing
 };
 // OPCODE-ENUM:END
+#undef EXP_OPCODE
+
+// Create Core namespace for consistency with other opcode groups
+namespace CoreOps {
+static const OpCodeTableID TableID = OpCodeTableID::CoreOps;
+using OpCode = hlsl::DXIL::OpCode;
+} // namespace CoreOps
 
 // clang-format off
   // Python lines need to be not formatted.
@@ -1244,6 +1292,9 @@ enum class OpCodeClass : unsigned {
   StorePrimitiveOutput,
   StoreVertexOutput,
 
+  // No-op
+  Nop,
+
   // Other
   CycleCounterLegacy,
 
@@ -1414,7 +1465,7 @@ enum class OpCodeClass : unsigned {
   NodeOutputIsValid,
   OutputComplete,
 
-  NumOpClasses = 196 // exclusive last value of enumeration
+  NumOpClasses = 197, // exclusive last value of enumeration
 };
 // OPCODECLASS-ENUM:END
 

include/dxc/DXIL/DxilInstructions.h

@@ -10231,5 +10231,25 @@ struct DxilInst_FDot {
   llvm::Value *get_b() const { return Instr->getOperand(2); }
   void set_b(llvm::Value *val) { Instr->setOperand(2, val); }
 };
+
+/// This instruction nop does nothing
+struct DxilInst_ExperimentalNop {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  DxilInst_ExperimentalNop(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return hlsl::OP::IsDxilOpFuncCallInst(Instr,
+                                          hlsl::OP::OpCode::ExperimentalNop);
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+  bool isArgumentListValid() const {
+    if (1 != llvm::dyn_cast<llvm::CallInst>(Instr)->getNumArgOperands())
+      return false;
+    return true;
+  }
+  // Metadata
+  bool requiresUniformInputs() const { return false; }
+};
 // INSTR-HELPER:END
 } // namespace hlsl

include/dxc/DXIL/DxilOperations.h

@@ -37,6 +37,7 @@ class OP {
 public:
   using OpCode = DXIL::OpCode;
   using OpCodeClass = DXIL::OpCodeClass;
+  using OpCodeTableID = DXIL::OpCodeTableID;
 
 public:
   OP() = delete;
@@ -131,6 +132,7 @@ class OP {
   llvm::Constant *GetFloatConst(float v);
   llvm::Constant *GetDoubleConst(double v);
 
+  static OP::OpCode getOpCode(unsigned OpCode);
   static OP::OpCode getOpCode(const llvm::Instruction *I);
   static llvm::Type *GetOverloadType(OpCode OpCode, llvm::Function *F);
   static OpCode GetDxilOpFuncCallInst(const llvm::Instruction *I);
@@ -226,8 +228,7 @@ class OP {
   std::unordered_map<const llvm::Function *, OpCodeClass> m_FunctionToOpClass;
   void UpdateCache(OpCodeClass opClass, llvm::Type *Ty, llvm::Function *F);
 
-private:
-  // Static properties.
+public:
   struct OverloadMask {
     // mask of type slot bits as (1 << TypeSlot)
     uint16_t SlotMask;
@@ -255,7 +256,31 @@ class OP {
     // AllowedOverloads[n][TS_Vector] is true.
     OverloadMask AllowedVectorElements[DXIL::kDxilMaxOloadDims];
   };
-  static const OpCodeProperty m_OpCodeProps[(unsigned)OpCode::NumOpCodes];
+  struct OpCodeTable {
+    OpCodeTableID ID;
+    const OpCodeProperty *Table;
+    unsigned Count;
+  };
+
+  // Look up table using high 16-bits as table ID, low 16-bits as OpCode.
+  // Return true if valid.
+  // unsigned versions are for use with whatever value was in a DXIL Op
+  // instruction.
+  // OpIndex is the low 16-bits, for index lookup within the table.
+  static bool DecodeOpCode(unsigned EncodedOpCode, OpCodeTableID &TableID,
+                           unsigned &OpIndex,
+                           unsigned *OptTableIndex = nullptr);
+  static bool DecodeOpCode(OpCode EncodedOpCode, OpCodeTableID &TableID,
+                           unsigned &OpIndex,
+                           unsigned *OptTableIndex = nullptr);
+  static bool IsValidOpCode(unsigned EncodedOpCode);
+  static bool IsValidOpCode(OpCode EncodedOpCode);
+
+private:
+  // Static properties.
+  static OpCodeTable g_OpCodeTables[DXIL::NumOpCodeTables];
+  static const OpCodeProperty &GetOpCodeProps(unsigned opCode);
+  static const OpCodeProperty &GetOpCodeProps(OpCode opCode);
 
   static const char *m_OverloadTypeName[TS_BasicCount];
   static const char *m_NamePrefix;

include/dxc/DXIL/DxilShaderModel.h

@@ -66,6 +66,9 @@ class ShaderModel {
   bool IsSMAtLeast(unsigned Major, unsigned Minor) const {
     return m_Major > Major || (m_Major == Major && m_Minor >= Minor);
   }
+  bool IsPreReleaseShaderModel() const {
+    return IsPreReleaseShaderModel(m_Major, m_Minor);
+  }
   bool IsSM50Plus() const { return IsSMAtLeast(5, 0); }
   bool IsSM51Plus() const { return IsSMAtLeast(5, 1); }
   bool AllowDerivatives(DXIL::ShaderKind sk) const;