Add internal-use __addressOf instrinsic to replace &

source/slang/core.meta.slang

@@ -2878,6 +2878,12 @@ __intrinsic_op(0)
 [require(cpp_cuda_spirv_llvm)]
 __prefix Ptr<T, Access::ReadWrite, AddressSpace::Device> operator&(__ref T value);
 
+__generic<T>
+__intrinsic_op(0)
+[KnownBuiltin( $((int)KnownBuiltinDeclName::InternalAddressOf))]
+[require(cpp_cuda_spirv_llvm)]
+Ptr<T, Access::ReadWrite, AddressSpace::Device> __addressOf(__ref T value);
+
 __generic<$(ptrTypeParameterList), TInt : __BuiltinIntegerType>
 __intrinsic_op($(kIROp_GetOffsetPtr))
 $(fullPtrType) operator+($(fullPtrType) value, TInt offset);

source/slang/slang-ast-support-types.h

@@ -235,6 +235,7 @@ FIDDLE() namespace Slang
         IDifferentiablePtr,
         NullDifferential,
         OperatorAddressOf,
+        InternalAddressOf,
         COUNT
     };
 

source/slang/slang-check-expr.cpp

@@ -3110,7 +3110,9 @@ Expr* SemanticsVisitor::CheckInvokeExprWithCheckedOperands(InvokeExpr* expr)
                                                 as<ConstantIntVal>(knownBuiltinAttr->name))
                                         {
                                             if (constantIntVal->getValue() ==
-                                                (int)KnownBuiltinDeclName::OperatorAddressOf)
+                                                    (int)KnownBuiltinDeclName::OperatorAddressOf ||
+                                                constantIntVal->getValue() ==
+                                                    (int)KnownBuiltinDeclName::InternalAddressOf)
                                             {
                                                 getSink()->diagnose(
                                                     argExpr,

source/standard-modules/neural/bindless-storage.slang

@@ -62,7 +62,7 @@ public struct BindlessAddress<T> : IPointerLikeAddress<T>
         {
         case cuda:
             // On CUDA, use packed vector atomic for types that support it (half2, bfloat16x2).
-            let scalarPtr = &handle[baseIndex + index];
+            let scalarPtr = __addressOf(handle[baseIndex + index]);
             let vecPtr = reinterpret<vector<T, 2>*>(scalarPtr);
             __atomic_reduce_add(vecPtr[0], value);
         default:

tests/bugs/gh-3601.slang

@@ -23,7 +23,7 @@ void funcThatTakesPointer(PP* p)
 }
 int* funcThatReturnsPointer(PP* p)
 {
-    return &p.data;
+    return __addressOf(p.data);
 }
 
 // CHECK: OpEntryPoint

tests/bugs/gh-7499.slang

@@ -20,7 +20,7 @@ export __extern_cpp int main()
     // Success is not crashing the compiler.
     int size = 0;
     int arr = 0;
-    f(&arr, size);
+    f(__addressOf(arr), size);
 
     return 0;
 }

tests/bugs/gh-8659.slang

@@ -9,7 +9,7 @@ RWStructuredBuffer<uint> outputBuffer;
 [numthreads(1,1,1)]
 void computeMain()
 {
-    let indices = (uint*)&ptr[0];
+    let indices = (uint*)__addressOf(ptr[0]);
     *indices = 100;
     // CHECK: 100
     outputBuffer[0] = ptr[0];

tests/cpu-program/gfx-smoke.slang

@@ -7,7 +7,7 @@ export __extern_cpp int main()
     gfx.DeviceDesc deviceDesc = {};
     deviceDesc.deviceType = gfx.DeviceType.CPU;
     Optional<gfx.IDevice> device;
-    gfx.gfxCreateDevice(&deviceDesc, device);
+    gfx.gfxCreateDevice(__addressOf(deviceDesc), device);
     if (device == none)
     {
         printf("fail\n");
@@ -17,7 +17,7 @@ export __extern_cpp int main()
     gfx.CommandQueueDesc queueDesc = {gfx::QueueType::Graphics};
     queueDesc.type = gfx.QueueType.Graphics;
     Optional<gfx.ICommandQueue> queue;
-    device.value.createCommandQueue(&queueDesc, queue);
+    device.value.createCommandQueue(__addressOf(queueDesc), queue);
 
     gfx.ShaderProgramDesc2 programDesc = {};
     NativeString s = R"(
@@ -36,15 +36,15 @@ export __extern_cpp int main()
     programDesc.sourceDataSize = s.length;
     programDesc.entryPointCount = 1;
     NativeString entryPointName = "computeMain";
-    programDesc.entryPointNames = &entryPointName;
+    programDesc.entryPointNames = __addressOf(entryPointName);
     Optional<gfx.IShaderProgram> program;
     Optional<slang.ISlangBlob> diagBlob;
-    device.value.createProgram2(&programDesc, program, diagBlob);
+    device.value.createProgram2(__addressOf(programDesc), program, diagBlob);
 
     Optional<gfx.IPipelineState> pipeline;
     gfx.ComputePipelineStateDesc pipelineDesc;
     pipelineDesc.program = NativeRef<gfx.IShaderProgram>(program.value);
-    device.value.createComputePipelineState(&pipelineDesc, pipeline);
+    device.value.createComputePipelineState(__addressOf(pipelineDesc), pipeline);
 
     Optional<gfx.ITransientResourceHeap> transientHeap;
     gfx.TransientResourceHeapDesc transientHeapDesc;
@@ -54,7 +54,7 @@ export __extern_cpp int main()
     transientHeapDesc.uavDescriptorCount = 1024;
     transientHeapDesc.samplerDescriptorCount = 256;
     transientHeapDesc.accelerationStructureDescriptorCount = 32;
-    device.value.createTransientResourceHeap(&transientHeapDesc, transientHeap);
+    device.value.createTransientResourceHeap(__addressOf(transientHeapDesc), transientHeap);
 
     Optional<gfx.IBufferResource> buffer;
     gfx.BufferResourceDesc bufferDesc = {};
@@ -64,12 +64,12 @@ export __extern_cpp int main()
     bufferDesc.elementSize = 4;
     bufferDesc.sizeInBytes = 256;
     bufferDesc.type = gfx.ResourceType.Buffer;
-    device.value.createBufferResource(&bufferDesc, nullptr, buffer);
+    device.value.createBufferResource(__addressOf(bufferDesc), nullptr, buffer);
 
     Optional<gfx.IResourceView> bufferView;
     gfx.ResourceViewDesc viewDesc;
     viewDesc.type = gfx.ResourceViewType.UnorderedAccess;
-    device.value.createBufferView(buffer.value, none, &viewDesc, bufferView);
+    device.value.createBufferView(buffer.value, none, __addressOf(viewDesc), bufferView);
 
     Optional<gfx.ICommandBuffer> commandBuffer;
     transientHeap.value.createCommandBuffer(commandBuffer);
@@ -80,13 +80,13 @@ export __extern_cpp int main()
     Optional<gfx.IShaderObject> entryPointObject;
     rootObject.value.getEntryPoint(0, entryPointObject);
     gfx.ShaderOffset offset = {};
-    entryPointObject.value.setResource(&offset, bufferView.value);
+    entryPointObject.value.setResource(__addressOf(offset), bufferView.value);
     encoder.value.dispatchCompute(1, 1, 1);
     encoder.value.endEncoding();
     commandBuffer.value.close();
 
     NativeRef<gfx.ICommandBuffer> commandBufferRef = NativeRef<gfx.ICommandBuffer>(commandBuffer.value);
-    queue.value.executeCommandBuffers(1, &commandBufferRef, none, 0);
+    queue.value.executeCommandBuffers(1, __addressOf(commandBufferRef), none, 0);
     queue.value.waitOnHost();
 
     Optional<slang.ISlangBlob> blob;

tests/cpu-program/pointer-basics.slang

@@ -2,13 +2,13 @@
 export __extern_cpp int main()
 {
     uint2 value;
-    int *pValue = (int*)&value;
+    int *pValue = (int*)__addressOf(value);
     *pValue = 1;
     (*pValue)++;
     ++pValue[0];
     ++pValue;
     *pValue = 1;
-    pValue = (int *)&value;
+    pValue = (int *)__addressOf(value);
     int64_t ptrVal = int64_t(pValue);
     pValue = (int *)ptrVal;
     if (pValue

tests/cpu-program/pointer-deref.slang

@@ -14,7 +14,7 @@ struct Record
 export __extern_cpp int main()
 {
     Record rec;
-    Record *pRec = &rec;
+    Record *pRec = __addressOf(rec);
     pRec.field = 1;
     pRec.sub.field2 = 2;
     pRec.sub.field3 = 3.0f;

tests/diagnostics/invalid-constant-pointer-taking.slang

@@ -11,7 +11,7 @@ StructuredBuffer<uint> constant_uint_buffer;
 [numthreads(1,1,1)]
 void computeMain(uint3 threadId : SV_DispatchThreadID)
 {
-    float* mutablePtr1 = &mutable_float_buffer[threadId.x];
+    float* mutablePtr1 = __addressOf(mutable_float_buffer[threadId.x]);
 
     // CHECK: ([[# @LINE+1]]): error 31160
     float* mutablePtr2 = __getAddress(mutable_float_buffer[threadId.x]);
@@ -20,7 +20,7 @@ void computeMain(uint3 threadId : SV_DispatchThreadID)
 
     // Constant pointers arent a thing in slang
     // CHECK: ([[# @LINE+1]]): error 30079:
-    float* ptr1 = &constant_float_buffer[threadId.x];
+    float* ptr1 = __addressOf(constant_float_buffer[threadId.x]);
 
     // CHECK: ([[# @LINE+1]]): error 31160
     float* ptr2 = __getAddress(constant_float_buffer[threadId.x]);

tests/language-feature/bitfield/msvc-repr.slang

@@ -31,6 +31,6 @@ void computeMain()
     s.d = 0xDEF0;
 
     // Write the struct to memory and read it back as uint
-    outputBuffer[0] = *((uint*)&s);
+    outputBuffer[0] = *((uint*)__addressOf(s));
 }
 

tests/language-feature/bitfield/repr-mixed.slang

@@ -31,6 +31,6 @@ void computeMain()
     m.d = 0xEF;
 
     // With default packing, all fields are in one backing field
-    outputBuffer[0] = *((uint*)&m) & 0xFFFFFF; // Mask to 24 bits
+    outputBuffer[0] = *((uint*)__addressOf(m)) & 0xFFFFFF; // Mask to 24 bits
 }
 

tests/language-feature/bitfield/repr.slang

@@ -31,6 +31,6 @@ void computeMain()
     s.d = 0xDEF0;
 
     // Write the struct to memory and read it back as uint
-    outputBuffer[0] = *((uint*)&s);
+    outputBuffer[0] = *((uint*)__addressOf(s));
 }
 

tests/language-feature/pointer/coherent-load-store-physical-storage-buffer.slang

@@ -20,5 +20,5 @@ void computeMain()
     Ptr<int> ptrIn = inputBuffer;
     Ptr<int> secondPtrIn = ptrIn;
     Ptr<int> ptrOut = outputBuffer;
-    storeCoherent<4, MemoryScope::Device>(ptrOut, loadCoherent<4, MemoryScope::Device>(&secondPtrIn[1]));
+    storeCoherent<4, MemoryScope::Device>(ptrOut, loadCoherent<4, MemoryScope::Device>(__addressOf(secondPtrIn[1])));
 }

tests/llvm/inline-llvm-ir.slang

@@ -32,7 +32,7 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
 {
     float4 a = float4(1,2,3,4);
     float f1 = reduceAdd(a);
-    memset(Ptr<void>(&a), 0, 8);
+    memset(Ptr<void>(__addressOf(a)), 0, 8);
     float f2 = reduceAdd(a);
 
     // CHECK: 10.0

tests/spirv/pointer.slang

@@ -18,7 +18,7 @@ void funcThatTakesPointer(PP* p)
 }
 int* funcThatReturnsPointer(PP* p)
 {
-    return &p.data;
+    return __addressOf(p.data);
 }
 
 void funcWithInOutParam(inout PP p)