[GISEL] Add G_VSCALE instruction (#84542)

llvm/docs/GlobalISel/GenericOpcode.rst

@@ -607,6 +607,17 @@ See the LLVM LangRef entry on '``llvm.lround.*'`` for details on behaviour.
 Vector Specific Operations
 --------------------------
 
+G_VSCALE
+^^^^^^^^
+
+Puts the value of the runtime ``vscale`` multiplied by the value in the source
+operand into the destination register. This can be useful in determining the
+actual runtime number of elements in a vector.
+
+.. code-block::
+
+  %0:_(s32) = G_VSCALE 4
+
 G_INSERT_SUBVECTOR
 ^^^^^^^^^^^^^^^^^^
 

llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h

@@ -1143,6 +1143,28 @@ public:
   MachineInstrBuilder buildInsert(const DstOp &Res, const SrcOp &Src,
                                   const SrcOp &Op, unsigned Index);
 
+  /// Build and insert \p Res = G_VSCALE \p MinElts
+  ///
+  /// G_VSCALE puts the value of the runtime vscale multiplied by \p MinElts
+  /// into \p Res.
+  ///
+  /// \pre setBasicBlock or setMI must have been called.
+  /// \pre \p Res must be a generic virtual register with scalar type.
+  ///
+  /// \return a MachineInstrBuilder for the newly created instruction.
+  MachineInstrBuilder buildVScale(const DstOp &Res, unsigned MinElts);
+
+  /// Build and insert \p Res = G_VSCALE \p MinElts
+  ///
+  /// G_VSCALE puts the value of the runtime vscale multiplied by \p MinElts
+  /// into \p Res.
+  ///
+  /// \pre setBasicBlock or setMI must have been called.
+  /// \pre \p Res must be a generic virtual register with scalar type.
+  ///
+  /// \return a MachineInstrBuilder for the newly created instruction.
+  MachineInstrBuilder buildVScale(const DstOp &Res, const ConstantInt &MinElts);
+
   /// Build and insert a G_INTRINSIC instruction.
   ///
   /// There are four different opcodes based on combinations of whether the

llvm/include/llvm/Support/TargetOpcodes.def

@@ -727,6 +727,9 @@ HANDLE_TARGET_OPCODE(G_BR)
 /// Generic branch to jump table entry.
 HANDLE_TARGET_OPCODE(G_BRJT)
 
+/// Generic vscale.
+HANDLE_TARGET_OPCODE(G_VSCALE)
+
 /// Generic insert subvector.
 HANDLE_TARGET_OPCODE(G_INSERT_SUBVECTOR)
 

llvm/include/llvm/Target/GenericOpcodes.td

@@ -1289,6 +1289,15 @@ def G_MERGE_VALUES : GenericInstruction {
   let variadicOpsType = type1;
 }
 
+// Generic vscale.
+// Puts the value of the runtime vscale multiplied by the value in the source
+// operand into the destination register.
+def G_VSCALE : GenericInstruction {
+  let OutOperandList = (outs type0:$dst);
+  let InOperandList = (ins unknown:$src);
+  let hasSideEffects = false;
+}
+
 /// Create a vector from multiple scalar registers. No implicit
 /// conversion is performed (i.e. the result element type must be the
 /// same as all source operands)

llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

@@ -793,6 +793,24 @@ MachineInstrBuilder MachineIRBuilder::buildInsert(const DstOp &Res,
   return buildInstr(TargetOpcode::G_INSERT, Res, {Src, Op, uint64_t(Index)});
 }
 
+MachineInstrBuilder MachineIRBuilder::buildVScale(const DstOp &Res,
+                                                  unsigned MinElts) {
+
+  auto IntN = IntegerType::get(getMF().getFunction().getContext(),
+                               Res.getLLTTy(*getMRI()).getScalarSizeInBits());
+  ConstantInt *CI = ConstantInt::get(IntN, MinElts);
+  return buildVScale(Res, *CI);
+}
+
+MachineInstrBuilder MachineIRBuilder::buildVScale(const DstOp &Res,
+                                                  const ConstantInt &MinElts) {
+  auto VScale = buildInstr(TargetOpcode::G_VSCALE);
+  VScale->setDebugLoc(DebugLoc());
+  Res.addDefToMIB(*getMRI(), VScale);
+  VScale.addCImm(&MinElts);
+  return VScale;
+}
+
 static unsigned getIntrinsicOpcode(bool HasSideEffects, bool IsConvergent) {
   if (HasSideEffects && IsConvergent)
     return TargetOpcode::G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS;

llvm/lib/CodeGen/MachineVerifier.cpp

@@ -1613,6 +1613,17 @@ void MachineVerifier::verifyPreISelGenericInstruction(const MachineInstr *MI) {
       report("G_BSWAP size must be a multiple of 16 bits", MI);
     break;
   }
+  case TargetOpcode::G_VSCALE: {
+    if (!MI->getOperand(1).isCImm()) {
+      report("G_VSCALE operand must be cimm", MI);
+      break;
+    }
+    if (MI->getOperand(1).getCImm()->isZero()) {
+      report("G_VSCALE immediate cannot be zero", MI);
+      break;
+    }
+    break;
+  }
   case TargetOpcode::G_INSERT_SUBVECTOR: {
     const MachineOperand &Src0Op = MI->getOperand(1);
     if (!Src0Op.isReg()) {

llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir

@@ -616,6 +616,9 @@
 # DEBUG-NEXT: G_BRJT (opcode {{[0-9]+}}): 2 type indices
 # DEBUG-NEXT: .. the first uncovered type index: 2, OK
 # DEBUG-NEXT: .. the first uncovered imm index: 0, OK
+# DEBUG-NEXT: G_VSCALE (opcode {{[0-9]+}}): 1 type index, 0 imm indices
+# DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: G_INSERT_SUBVECTOR (opcode {{[0-9]+}}): 2 type indices, 1 imm index
 # DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined

llvm/test/MachineVerifier/test_g_vscale.mir

@@ -0,0 +1,15 @@
+# RUN: not --crash llc -verify-machineinstrs -run-pass none -o /dev/null %s 2>&1 | FileCheck %s
+
+---
+name:            g_vscale
+body: |
+  bb.0:
+
+  %1:_(s32) = G_CONSTANT i32 4
+
+  ; CHECK: G_VSCALE operand must be cimm
+  %2:_(s32) = G_VSCALE %1
+
+  ; CHECK: G_VSCALE immediate cannot be zero
+  %3:_(s32) = G_VSCALE i32 0
+...