[Autobackout][FunctionalRegression]Revert of change: df4a2a2: Fix subroutine handling for intel_reqd_sub_group_size(32)

Previously, using `intel_reqd_sub_group_size(32)` on DG2 resulted in two
    redundant SIMD32 call instructions being generated in vISA, which could
    lead to unexpected issues. This change ensures that only a single SIMD32
    call instruction is generated. All function arguments and return values
    are now correctly passed using two SIMD16 instructions, eliminating
    redundancy and improving

Author: sys-igc

IGC/Compiler/CISACodeGen/EmitVISAPass.cpp

@@ -399,8 +399,9 @@ uint EmitPass::DecideInstanceAndSlice(const llvm::BasicBlock &blk, SDAG &sdag,
   }
 
   if (CallInst *callInst = dyn_cast<CallInst>(sdag.m_root)) {
-    // Disable slicing for function calls (stackcalls and subroutines)
-    if (isUserFunctionCall(callInst)) {
+    // Disable slicing for function calls
+    Function *F = dyn_cast<Function>(IGCLLVM::getCalledValue(callInst));
+    if (!F || F->hasFnAttribute("visaStackCall")) {
       numInstance = 1;
       slicing = false;
     }
@@ -11635,7 +11636,7 @@ void EmitPass::emitCall(llvm::CallInst *inst) {
     // When both symbols are the same, then this argument passing has been
     // lifted to use a global vISA variable, just skip the copy.
     if (Dst != Src) {
-      emitCopyAllInstances(Dst, Src, Arg.getType());
+      emitCopyAll(Dst, Src, Arg.getType());
     }
   }
   m_currFuncHasSubroutine = true;
@@ -11646,7 +11647,7 @@ void EmitPass::emitCall(llvm::CallInst *inst) {
   if (!inst->use_empty()) {
     CVariable *Dst = GetSymbol(inst);
     CVariable *Src = m_currShader->getOrCreateReturnSymbol(F);
-    emitCopyAllInstances(Dst, Src, inst->getType());
+    emitCopyAll(Dst, Src, inst->getType());
   }
 }
 
@@ -11665,7 +11666,7 @@ void EmitPass::emitReturn(llvm::ReturnInst *inst) {
     if (!RetTy->isVoidTy()) {
       CVariable *Dst = m_currShader->getOrCreateReturnSymbol(F);
       CVariable *Src = GetSymbol(inst->getReturnValue());
-      emitCopyAllInstances(Dst, Src, RetTy);
+      emitCopyAll(Dst, Src, RetTy);
     }
 
     m_encoder->SubroutineRet(nullptr, F);
@@ -21034,14 +21035,6 @@ void EmitPass::emitCopyAll(CVariable *Dst, CVariable *Src, llvm::Type *Ty) {
   }
 }
 
-void EmitPass::emitCopyAllInstances(CVariable *Dst, CVariable *Src,
-                                    llvm::Type *type) {
-  for (uint instance = 0; instance < Dst->GetNumberInstance(); instance++) {
-    m_encoder->SetSecondHalf(instance == 1 ? true : false);
-    emitCopyAll(Dst, Src, type);
-  }
-}
-
 void EmitPass::emitSqrt(Instruction *inst) {
   GenIntrinsicInst *intrinCall = llvm::cast<GenIntrinsicInst>(inst);
   CVariable *src0 = GetSymbol(intrinCall->getArgOperand(0));

IGC/Compiler/CISACodeGen/EmitVISAPass.hpp

@@ -534,7 +534,6 @@ class EmitPass : public llvm::FunctionPass
         bool allowLargerSIMDSize = false, CVariable* predicate = nullptr);
     void emitConstantVector(CVariable* Dst, uint64_t value = 0);
     void emitCopyAll(CVariable* Dst, CVariable* Src, llvm::Type* Ty);
-    void emitCopyAllInstances(CVariable *Dst, CVariable *Src, llvm::Type *Ty);
 
     void emitPredicatedVectorCopy(CVariable* Dst, CVariable* Src, CVariable* pred);
     void emitPredicatedVectorSelect(CVariable* Dst, CVariable* Src0,