[Wasm RyuJit] Null checks for calls, overflow checks for int to int casts, resolve stack ordering, more (#124534)

Various fixes from running crossgen

* fix return type classification for structs returned as primitives
* null checks for calls
* overflow checks for int to int casts
* stack ordering and null check for delegate invoke
* stack ordering for NOT

Author: AndyAyersMS

src/coreclr/jit/codegen.h

@@ -216,6 +216,7 @@ class CodeGen final : public CodeGenInterface
     unsigned                   findTargetDepth(BasicBlock* target);
     void                       WasmProduceReg(GenTree* node);
     regNumber                  GetMultiUseOperandReg(GenTree* operand);
+    void                       genEmitNullCheck(regNumber reg);
 #endif
 
     void        genEmitStartBlock(BasicBlock* block);

src/coreclr/jit/codegenwasm.cpp

@@ -747,16 +747,19 @@ static constexpr uint32_t PackTypes(var_types toType, var_types fromType)
 //
 void CodeGen::genIntToIntCast(GenTreeCast* cast)
 {
-    if (cast->gtOverflow())
+    GenIntCastDesc desc(cast);
+
+    if (desc.CheckKind() != GenIntCastDesc::CHECK_NONE)
     {
-        NYI_WASM("Overflow checks");
+        GenTree*  castValue = cast->gtGetOp1();
+        regNumber castReg   = GetMultiUseOperandReg(castValue);
+        genIntCastOverflowCheck(cast, desc, castReg);
     }
 
-    GenIntCastDesc desc(cast);
-    var_types      toType     = genActualType(cast->CastToType());
-    var_types      fromType   = genActualType(cast->CastOp());
-    int            extendSize = desc.ExtendSrcSize();
-    instruction    ins        = INS_none;
+    var_types   toType     = genActualType(cast->CastToType());
+    var_types   fromType   = genActualType(cast->CastOp());
+    int         extendSize = desc.ExtendSrcSize();
+    instruction ins        = INS_none;
     assert(fromType == TYP_INT || fromType == TYP_LONG);
 
     genConsumeOperands(cast);
@@ -819,6 +822,98 @@ void CodeGen::genIntToIntCast(GenTreeCast* cast)
     WasmProduceReg(cast);
 }
 
+//------------------------------------------------------------------------
+// genIntCastOverflowCheck: Generate overflow checking code for an integer cast.
+//
+// Arguments:
+//    cast - The GT_CAST node
+//    desc - The cast description
+//    reg  - The register containing the value to check
+//
+void CodeGen::genIntCastOverflowCheck(GenTreeCast* cast, const GenIntCastDesc& desc, regNumber reg)
+{
+    bool const     is64BitSrc = (desc.CheckSrcSize() == 8);
+    emitAttr const srcSize    = is64BitSrc ? EA_8BYTE : EA_4BYTE;
+
+    GetEmitter()->emitIns_I(INS_local_get, srcSize, WasmRegToIndex(reg));
+
+    switch (desc.CheckKind())
+    {
+        case GenIntCastDesc::CHECK_POSITIVE:
+        {
+            // INT or LONG to ULONG
+            GetEmitter()->emitIns_I(is64BitSrc ? INS_i64_const : INS_i32_const, srcSize, 0);
+            GetEmitter()->emitIns(is64BitSrc ? INS_i64_lt_s : INS_i32_lt_s);
+            genJumpToThrowHlpBlk(SCK_OVERFLOW);
+            break;
+        }
+
+        case GenIntCastDesc::CHECK_UINT_RANGE:
+        {
+            // (U)LONG to UINT
+            assert(is64BitSrc);
+            GetEmitter()->emitIns_I(INS_i64_const, srcSize, UINT32_MAX);
+            // We can re-interpret LONG as ULONG
+            // Then negative values will be larger than UINT32_MAX
+            GetEmitter()->emitIns(INS_i64_gt_u);
+            genJumpToThrowHlpBlk(SCK_OVERFLOW);
+            break;
+        }
+
+        case GenIntCastDesc::CHECK_POSITIVE_INT_RANGE:
+        {
+            // ULONG to INT
+            GetEmitter()->emitIns_I(INS_i64_const, srcSize, INT32_MAX);
+            GetEmitter()->emitIns(INS_i64_gt_u);
+            genJumpToThrowHlpBlk(SCK_OVERFLOW);
+            break;
+        }
+
+        case GenIntCastDesc::CHECK_INT_RANGE:
+        {
+            // LONG to INT
+            GetEmitter()->emitIns(INS_i64_extend32_s);
+            GetEmitter()->emitIns_I(INS_local_get, srcSize, WasmRegToIndex(reg));
+            GetEmitter()->emitIns(INS_i64_ne);
+            genJumpToThrowHlpBlk(SCK_OVERFLOW);
+            break;
+        }
+
+        case GenIntCastDesc::CHECK_SMALL_INT_RANGE:
+        {
+            // (U)(INT|LONG) to Small INT
+            const int castMaxValue = desc.CheckSmallIntMax();
+            const int castMinValue = desc.CheckSmallIntMin();
+
+            if (castMinValue == 0)
+            {
+                // When the minimum is 0, a single unsigned upper-bound check is sufficient.
+                // For signed sources, negative values become large unsigned values and
+                // thus also trigger the overflow via the same comparison.
+                GetEmitter()->emitIns_I(is64BitSrc ? INS_i64_const : INS_i32_const, srcSize, castMaxValue);
+                GetEmitter()->emitIns(is64BitSrc ? INS_i64_gt_u : INS_i32_gt_u);
+            }
+            else
+            {
+                // We need to check a range around zero, eg [-128, 127] for 8-bit signed.
+                // Do two compares and combine the results: (src > max) | (src < min).
+                assert(!cast->IsUnsigned());
+                GetEmitter()->emitIns_I(is64BitSrc ? INS_i64_const : INS_i32_const, srcSize, castMaxValue);
+                GetEmitter()->emitIns(is64BitSrc ? INS_i64_gt_s : INS_i32_gt_s);
+                GetEmitter()->emitIns_I(INS_local_get, srcSize, WasmRegToIndex(reg));
+                GetEmitter()->emitIns_I(is64BitSrc ? INS_i64_const : INS_i32_const, srcSize, castMinValue);
+                GetEmitter()->emitIns(is64BitSrc ? INS_i64_lt_s : INS_i32_lt_s);
+                GetEmitter()->emitIns(INS_i32_or);
+            }
+            genJumpToThrowHlpBlk(SCK_OVERFLOW);
+            break;
+        }
+
+        default:
+            unreached();
+    }
+}
+
 //------------------------------------------------------------------------
 // genFloatToIntCast: Generate code for a floating point to integer cast
 //
@@ -1368,6 +1463,22 @@ void CodeGen::genJumpToThrowHlpBlk(SpecialCodeKind codeKind)
 void CodeGen::genCodeForNullCheck(GenTreeIndir* tree)
 {
     genConsumeAddress(tree->Addr());
+    genEmitNullCheck(REG_NA);
+}
+
+//---------------------------------------------------------------------
+// genEmitNullCheck - generate code for a null check
+//
+// Arguments:
+//    regNum - register to check, or REG_NA if value to check is on the stack
+//
+void CodeGen::genEmitNullCheck(regNumber reg)
+{
+    if (reg != REG_NA)
+    {
+        GetEmitter()->emitIns_I(INS_local_get, EA_PTRSIZE, WasmRegToIndex(reg));
+    }
+
     GetEmitter()->emitIns_I(INS_I_const, EA_PTRSIZE, m_compiler->compMaxUncheckedOffsetForNullObject);
     GetEmitter()->emitIns(INS_I_le_u);
     genJumpToThrowHlpBlk(SCK_NULL_CHECK);
@@ -1629,15 +1740,20 @@ void CodeGen::genCodeForStoreInd(GenTreeStoreInd* tree)
 //------------------------------------------------------------------------
 // genCall: Produce code for a GT_CALL node
 //
+// Arguments:
+//    call - the GT_CALL node
+//
 void CodeGen::genCall(GenTreeCall* call)
 {
+    regNumber thisReg = REG_NA;
+
     if (call->NeedsNullCheck())
     {
-        NYI_WASM("Insert nullchecks for calls that need it in lowering");
+        CallArg* thisArg  = call->gtArgs.GetThisArg();
+        GenTree* thisNode = thisArg->GetNode();
+        thisReg           = GetMultiUseOperandReg(thisNode);
     }
 
-    assert(!call->IsTailCall());
-
     for (CallArg& arg : call->gtArgs.EarlyArgs())
     {
         genConsumeReg(arg.GetEarlyNode());
@@ -1648,6 +1764,11 @@ void CodeGen::genCall(GenTreeCall* call)
         genConsumeReg(arg.GetLateNode());
     }
 
+    if (call->NeedsNullCheck())
+    {
+        genEmitNullCheck(thisReg);
+    }
+
     genCallInstruction(call);
     WasmProduceReg(call);
 }

src/coreclr/jit/compiler.cpp

@@ -766,7 +766,7 @@ var_types Compiler::getReturnTypeForStruct(CORINFO_CLASS_HANDLE     clsHnd,
     }
     else
     {
-        howToReturnStruct = SPK_ByValue;
+        howToReturnStruct = SPK_PrimitiveType;
         useType           = WasmClassifier::ToJitType(abiType);
     }
 

src/coreclr/jit/gentree.cpp

@@ -31035,6 +31035,11 @@ void ReturnTypeDesc::InitializeStructReturnType(Compiler*                comp,
                 m_regType[i] = comp->getJitGCType(gcPtrs[i]);
             }
 
+#elif defined(TARGET_WASM)
+
+            // For Wasm, structs are either returned by-ref or as primitives.
+            unreached();
+
 #else //  TARGET_XXX
 
             // This target needs support here!
@@ -34175,3 +34180,26 @@ ValueSize ValueSize::FromJitType(var_types type)
             return ValueSize(genTypeSize(type));
     }
 }
+
+//------------------------------------------------------------------------
+// gtFirstNodeInOperandOrder : return the first node of this tree
+//    in operand order
+//
+// Returns:
+//   If tree is a leaf, return the tree.
+//   If the tree has operands, recurse on the first operand.
+//
+GenTree* GenTree::gtFirstNodeInOperandOrder()
+{
+    GenTree*             op = this;
+    GenTree::VisitResult visitResult;
+    do
+    {
+        visitResult = op->VisitOperands([&op](GenTree* operand) {
+            op = operand;
+            return GenTree::VisitResult::Abort;
+        });
+    } while (visitResult == GenTree::VisitResult::Abort);
+
+    return op;
+}

src/coreclr/jit/gentree.h

@@ -1984,6 +1984,8 @@ struct GenTree
 
     inline GenTree* gtCommaStoreVal();
 
+    GenTree* gtFirstNodeInOperandOrder();
+
     // Return the child of this node if it is a GT_RELOAD or GT_COPY; otherwise simply return the node itself
     inline GenTree* gtSkipReloadOrCopy();
 

src/coreclr/jit/lower.cpp

@@ -2012,8 +2012,24 @@ void Lowering::LowerArgsForCall(GenTreeCall* call)
 #endif // defined(TARGET_X86) && defined(FEATURE_IJW)
 
     LegalizeArgPlacement(call);
+    AfterLowerArgsForCall(call);
 }
 
+#if !defined(TARGET_WASM)
+
+//------------------------------------------------------------------------
+// AfterLowerArgsForCall: post processing after call args are lowered
+//
+// Arguments:
+//    call - Call node
+//
+void Lowering::AfterLowerArgsForCall(GenTreeCall* call)
+{
+    // no-op for non-Wasm targets
+}
+
+#endif // !defined(TARGET_WASM)
+
 #if defined(TARGET_X86) && defined(FEATURE_IJW)
 //------------------------------------------------------------------------
 // LowerSpecialCopyArgs: Lower special copy arguments for P/Invoke IL stubs

src/coreclr/jit/lower.h

@@ -211,6 +211,7 @@ class Lowering final : public Phase
     GenTree* LowerVirtualVtableCall(GenTreeCall* call);
     GenTree* LowerVirtualStubCall(GenTreeCall* call);
     void     LowerArgsForCall(GenTreeCall* call);
+    void     AfterLowerArgsForCall(GenTreeCall* call);
 #if defined(TARGET_X86) && defined(FEATURE_IJW)
     void LowerSpecialCopyArgs(GenTreeCall* call);
     void InsertSpecialCopyArg(GenTreePutArgStk* putArgStk, CORINFO_CLASS_HANDLE argType, unsigned lclNum);

src/coreclr/jit/lowerwasm.cpp

@@ -121,7 +121,12 @@ GenTree* Lowering::LowerNeg(GenTreeOp* node)
     //
     GenTree* x    = node->gtGetOp1();
     GenTree* zero = m_compiler->gtNewZeroConNode(node->TypeGet());
-    BlockRange().InsertBefore(x, zero);
+
+    // To preserve stack order we must insert the zero before the entire
+    // tree rooted at x.
+    //
+    GenTree* insertBefore = x->gtFirstNodeInOperandOrder();
+    BlockRange().InsertBefore(insertBefore, zero);
     LowerNode(zero);
     node->ChangeOper(GT_SUB);
     node->gtOp1 = zero;
@@ -221,6 +226,11 @@ void Lowering::LowerPutArgStk(GenTreePutArgStk* putArgNode)
 void Lowering::LowerCast(GenTree* tree)
 {
     assert(tree->OperIs(GT_CAST));
+
+    if (tree->gtOverflow())
+    {
+        tree->gtGetOp1()->gtLIRFlags |= LIR::Flags::MultiplyUsed;
+    }
     ContainCheckCast(tree->AsCast());
 }
 
@@ -463,3 +473,19 @@ void Lowering::AfterLowerBlock()
     Stackifier stackifier(this);
     stackifier.StackifyCurrentBlock();
 }
+
+//------------------------------------------------------------------------
+// AfterLowerArgsForCall: post processing after call args are lowered
+//
+// Arguments:
+//    call - Call node
+//
+void Lowering::AfterLowerArgsForCall(GenTreeCall* call)
+{
+    if (call->NeedsNullCheck())
+    {
+        // Prepare for explicit null check
+        CallArg* thisArg = call->gtArgs.GetThisArg();
+        thisArg->GetNode()->gtLIRFlags |= LIR::Flags::MultiplyUsed;
+    }
+}