Prevent hoisting of non-hoistable instructions in non-function values…

… with code (#4250)

source/slang/slang-ir-redundancy-removal.cpp

@@ -25,9 +25,9 @@ struct RedundancyRemovalContext
                 for (UInt i = 0; i < inst->getOperandCount(); i++)
                 {
                     auto operand = inst->getOperand(i);
-                    if (getParentFunc(operand) != func)
+                    if (!hasDescendent(func, operand))
                     {
-                        // Global value won't prevent hoisting.
+                        // Only prevent hoisting from operands local to this function
                         continue;
                     }
                     auto operandParent = as<IRBlock>(operand->getParent());

source/slang/slang-ir.cpp

@@ -8530,6 +8530,18 @@ namespace Slang
         return nullptr;
     }
 
+    bool hasDescendent(IRInst* inst, IRInst* child)
+    {
+        auto parent = child->getParent();
+        while (parent)
+        {
+            if (inst == parent)
+                return true;
+            parent = parent->getParent();
+        }
+        return false;
+    }
+
     IRInst* getGenericReturnVal(IRInst* inst)
     {
         if (auto gen = as<IRGeneric>(inst))

source/slang/slang-ir.h

@@ -2538,6 +2538,9 @@ bool isBuiltin(IRInst* inst);
     // Get the enclosuing function of an instruction.
 IRFunc* getParentFunc(IRInst* inst);
 
+    // Is child a descendent of inst
+bool hasDescendent(IRInst* inst, IRInst* child);
+
     // True if moving this inst will not change the semantics of the program
 bool isMovableInst(IRInst* inst);
 

tests/bugs/generic-initializer-inlining.slang

@@ -0,0 +1,33 @@
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-cpu
+
+// CHECK:      1
+// CHECK-NEXT: 0
+// CHECK-NEXT: 0
+// CHECK-NEXT: 0
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<uint> outputBuffer;
+
+struct C<T>
+{
+    [ForceInline]
+    __init<U>(U value)
+    {
+        for (int i = 0; i < 1; ++i)
+        {}
+    }
+}
+
+static C<int> c = C<int>(1);
+
+int use<T>(T x)
+{
+    return 1;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint i : SV_GroupIndex)
+{
+    outputBuffer[0] = use(c);
+}
+