Allow fasttail calls for engregesterable structs

src/jit/compiler.h

@@ -8186,13 +8186,20 @@ XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
         bool compPublishStubParam : 1; // EAX captured in prolog will be available through an instrinsic
         bool compRetBuffDefStack : 1;  // The ret buff argument definitely points into the stack.
 
-        var_types compRetType;       // Return type of the method as declared in IL
-        var_types compRetNativeType; // Normalized return type as per target arch ABI
-        unsigned  compILargsCount;   // Number of arguments (incl. implicit but not hidden)
-        unsigned  compArgsCount;     // Number of arguments (incl. implicit and     hidden)
-        unsigned  compRetBuffArg;    // position of hidden return param var (0, 1) (BAD_VAR_NUM means not present);
-        int compTypeCtxtArg; // position of hidden param for type context for generic code (CORINFO_CALLCONV_PARAMTYPE)
-        unsigned       compThisArg; // position of implicit this pointer param (not to be confused with lvaArg0Var)
+        var_types compRetType;          // Return type of the method as declared in IL
+        var_types compRetNativeType;    // Normalized return type as per target arch ABI
+        unsigned  compILargsCount;      // Number of arguments (incl. implicit but not hidden)
+        unsigned  compArgsCount;        // Number of arguments (incl. implicit and     hidden)
+
+#ifdef FEATURE_MULTIREG_ARGS
+        unsigned  compArgRegCount;      // Number of caller integer args
+        unsigned  compFloatArgRegCount; // Number of caller floating point args
+        unsigned  compStackSize;        // Caller stack size
+#endif // FEATURE_MULTIREG_ARGS
+
+        unsigned  compRetBuffArg;       // position of hidden return param var (0, 1) (BAD_VAR_NUM means not present);
+        int compTypeCtxtArg;            // position of hidden param for type context for generic code (CORINFO_CALLCONV_PARAMTYPE)
+        unsigned       compThisArg;     // position of implicit this pointer param (not to be confused with lvaArg0Var)
         unsigned       compILlocalsCount; // Number of vars : args + locals (incl. implicit but not hidden)
         unsigned       compLocalsCount;   // Number of vars : args + locals (incl. implicit and     hidden)
         unsigned       compMaxStack;

src/jit/lclvars.cpp

@@ -108,12 +108,19 @@ void Compiler::lvaInitTypeRef()
     /* Set compArgsCount and compLocalsCount */
 
     info.compArgsCount = info.compMethodInfo->args.numArgs;
+    unsigned   argRegCount      = 0;
+
+    auto incrementArgCount = [this, &argRegCount]()
+    {
+        ++info.compArgsCount;
+        ++argRegCount;
+    };
 
     // Is there a 'this' pointer
 
     if (!info.compIsStatic)
     {
-        info.compArgsCount++;
+        incrementArgCount();
     }
     else
     {
@@ -167,7 +174,7 @@ void Compiler::lvaInitTypeRef()
 
     if (hasRetBuffArg)
     {
-        info.compArgsCount++;
+        incrementArgCount();
     }
     else
     {
@@ -179,14 +186,14 @@ void Compiler::lvaInitTypeRef()
 
     if (info.compIsVarArgs)
     {
-        info.compArgsCount++;
+        incrementArgCount();
     }
 
     // Is there an extra parameter used to pass instantiation info to
     // shared generic methods and shared generic struct instance methods?
     if (info.compMethodInfo->args.callConv & CORINFO_CALLCONV_PARAMTYPE)
     {
-        info.compArgsCount++;
+        incrementArgCount();
     }
     else
     {
@@ -233,15 +240,54 @@ void Compiler::lvaInitTypeRef()
     InitVarDscInfo varDscInfo;
     varDscInfo.Init(lvaTable, hasRetBuffArg);
 
+    // Maintain a pointer to the start of the arguments.
+    LclVarDsc* varDscTrailingPointer = varDscInfo.varDsc;
     lvaInitArgs(&varDscInfo);
 
+    //-------------------------------------------------------------------------
+    // Calculate the argument register usage.
+    //
+    // This will later be used for fastTailCall determination
+    //-------------------------------------------------------------------------
+
+    unsigned   floatingRegCount = 0;
+    unsigned   stackArgCount    = 0;
+    unsigned   stackSize        = 0;
+
+    auto incrementRegCount = [&floatingRegCount, &argRegCount](LclVarDsc* varDsc)
+    {
+        varDsc->IsFloatRegType() ? ++floatingRegCount : ++argRegCount;
+    };
+
+    for (unsigned argNum = 0; argNum < info.compMethodInfo->args.numArgs; argNum++, varDscTrailingPointer++)
+    {
+        if (varDscTrailingPointer->lvRegNum != REG_STK)
+        {
+            incrementRegCount(varDscTrailingPointer);
+
+#ifdef FEATURE_MULTIREG_ARGS
+            if (varDscTrailingPointer->lvOtherArgReg != REG_NA)
+            {
+                incrementRegCount(varDscTrailingPointer);
+            }
+#endif // FEATURE_MULTIREG_ARGS
+        }
+        else
+        {
+            stackSize += varDscTrailingPointer->lvSize();
+        }
+    }
+
     //-------------------------------------------------------------------------
     // Finally the local variables
     //-------------------------------------------------------------------------
 
-    unsigned                varNum    = varDscInfo.varNum;
-    LclVarDsc*              varDsc    = varDscInfo.varDsc;
-    CORINFO_ARG_LIST_HANDLE localsSig = info.compMethodInfo->locals.args;
+    unsigned                varNum          = varDscInfo.varNum;
+    LclVarDsc*              varDsc          = varDscInfo.varDsc;
+    CORINFO_ARG_LIST_HANDLE localsSig       = info.compMethodInfo->locals.args;
+
+    // Maintain a pointer to the start of the locals.
+    varDscTrailingPointer = varDsc;
 
     for (unsigned i = 0; i < info.compMethodInfo->locals.numArgs;
          i++, varNum++, varDsc++, localsSig = info.compCompHnd->getArgNext(localsSig))
@@ -255,13 +301,25 @@ void Compiler::lvaInitTypeRef()
         varDsc->lvPinned  = ((corInfoType & CORINFO_TYPE_MOD_PINNED) != 0);
         varDsc->lvOnFrame = true; // The final home for this local variable might be our local stack frame
 
+        CORINFO_CLASS_HANDLE clsHnd = info.compCompHnd->getArgClass(&info.compMethodInfo->locals, localsSig);
+        stackSize += roundUp(compGetTypeSize(strip(corInfoType), clsHnd), TARGET_POINTER_SIZE);
+
         if (strip(corInfoType) == CORINFO_TYPE_CLASS)
         {
-            CORINFO_CLASS_HANDLE clsHnd = info.compCompHnd->getArgClass(&info.compMethodInfo->locals, localsSig);
             lvaSetClass(varNum, clsHnd);
         }
     }
 
+    //-------------------------------------------------------------------------
+    // Save the register usage information and stack size.
+    //-------------------------------------------------------------------------
+
+    stackSize += stackArgCount * REGSIZE_BYTES;
+
+    info.compArgRegCount = argRegCount;
+    info.compFloatArgRegCount = floatingRegCount;
+    info.compStackSize = stackSize;
+
     if ( // If there already exist unsafe buffers, don't mark more structs as unsafe
         // as that will cause them to be placed along with the real unsafe buffers,
         // unnecessarily exposing them to overruns. This can affect GS tests which
@@ -1253,6 +1311,10 @@ void Compiler::lvaInitVarDsc(LclVarDsc*              varDsc,
 #ifdef DEBUG
     varDsc->lvStkOffs = BAD_STK_OFFS;
 #endif
+
+#ifdef FEATURE_MULTIREG_ARGS
+    varDsc->lvOtherArgReg = REG_NA;
+#endif // FEATURE_MULTIREG_ARGS
 }
 
 /*****************************************************************************

src/jit/lower.cpp

@@ -1714,7 +1714,6 @@ void Lowering::LowerFastTailCall(GenTreeCall* call)
         fgArgTabEntryPtr argTabEntry = comp->gtArgEntryByNode(call, putArgStkNode);
         assert(argTabEntry);
         unsigned callerArgNum = argTabEntry->argNum - calleeNonStandardArgCount;
-        noway_assert(callerArgNum < comp->info.compArgsCount);
 
         unsigned   callerArgLclNum = callerArgNum;
         LclVarDsc* callerArgDsc    = comp->lvaTable + callerArgLclNum;