add OpLoad and OpStore SPIR-V intrinsics for non-BDA pointer types

Also note that `nbl::hlsl::bit_cast` needs a lot of fixing

Author: devsh

include/nbl/builtin/hlsl/bda/__ref.hlsl

@@ -1,27 +1,15 @@
 // Copyright (C) 2018-2024 - DevSH Graphics Programming Sp. z O.O.
 // This file is part of the "Nabla Engine".
 // For conditions of distribution and use, see copyright notice in nabla.h
-
-#include "nbl/builtin/hlsl/cpp_compat.hlsl"
-#include "nbl/builtin/hlsl/spirv_intrinsics/core.hlsl"
-
 #ifndef _NBL_BUILTIN_HLSL_BDA_REF_INCLUDED_
 #define _NBL_BUILTIN_HLSL_BDA_REF_INCLUDED_
 
+#include "nbl/builtin/hlsl/functional.hlsl"
+
 namespace nbl
 {
 namespace hlsl
 {
-
-// TODO: make a common `nbl/builtin/hlsl/__ref.hlsl`
-// TODO: also refactor `bda::__base_ref` into just `__ref` and make it a typedef
-template<uint32_t StorageClass, typename T>
-using __spv_ptr_t = spirv::pointer_t<StorageClass,T>;
-
-template<uint32_t StorageClass, typename T>
-[[vk::ext_instruction(spv::OpCopyObject)]]
-__spv_ptr_t<StorageClass,T> addrof([[vk::ext_reference]] T v);
-
 namespace bda
 {
 template<typename T>
@@ -30,6 +18,7 @@ using __spv_ptr_t = spirv::pointer_t<spv::StorageClassPhysicalStorageBuffer,T>;
 template<typename T>
 struct __ptr;
 
+// TODO: refactor this in terms of `nbl::hlsl::` when they fix the composite struct inline SPIR-V BDA issue
 template<typename T, uint32_t alignment, bool _restrict>
 struct __base_ref
 {
@@ -52,12 +41,12 @@ struct __base_ref
 
     T load()
     {
-        return spirv::load < T, __spv_ptr_t<T>, alignment > (__get_spv_ptr());
+        return spirv::load<T,alignment>(__get_spv_ptr());
     }
 
     void store(const T val)
     {
-        spirv::store < T, __spv_ptr_t<T>, alignment > (__get_spv_ptr(), val);
+        spirv::store<T,alignment>(__get_spv_ptr(),val);
     }
 };
 

include/nbl/builtin/hlsl/memory_accessor.hlsl

@@ -116,7 +116,7 @@ struct StructureOfArrays : impl::StructureOfArraysBase<IndexType,ElementStride,S
     enable_if_t<sizeof(T)%sizeof(access_t)==0,void> get(const index_t ix, NBL_REF_ARG(T) value)
     {
         NBL_CONSTEXPR uint64_t SubElementCount = sizeof(T)/sizeof(access_t);
-        // `vector` for now, we'll use `array` later
+        // `vector` for now, we'll use `array` later when `bit_cast` gets fixed
         vector<access_t,SubElementCount> aux;
         for (index_t i=0; i<SubElementCount; i++)
             accessor.get(base_t::getIx(ix,i),aux[i]);
@@ -127,7 +127,7 @@ struct StructureOfArrays : impl::StructureOfArraysBase<IndexType,ElementStride,S
     enable_if_t<sizeof(T)%sizeof(access_t)==0,void> set(const index_t ix, NBL_CONST_REF_ARG(T) value)
     { 
         NBL_CONSTEXPR uint64_t SubElementCount = sizeof(T)/sizeof(access_t);
-        // `vector` for now, we'll use `array` later
+        // `vector` for now, we'll use `array` later when `bit_cast` gets fixed
         vector<access_t,SubElementCount> aux;
         aux = bit_cast<vector<access_t,SubElementCount>,T>(value);
         for (index_t i=0; i<SubElementCount; i++)

include/nbl/builtin/hlsl/spirv_intrinsics/core.hlsl

@@ -169,15 +169,24 @@ template<typename T, typename Ptr_T> // DXC Workaround
 [[vk::ext_instruction(spv::OpAtomicCompareExchange)]]
 enable_if_t<is_spirv_type_v<Ptr_T>, T> atomicCompareExchange(Ptr_T ptr, uint32_t memoryScope, uint32_t memSemanticsEqual, uint32_t memSemanticsUnequal, T value, T comparator);
 
-template<typename T, typename P, uint32_t alignment>
+
+template<typename T, uint32_t alignment>
 [[vk::ext_capability(spv::CapabilityPhysicalStorageBufferAddresses)]]
 [[vk::ext_instruction(spv::OpLoad)]]
-enable_if_t<is_spirv_type_v<P>, T> load(P pointer, [[vk::ext_literal]] uint32_t __aligned = /*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment = alignment);
+T load(pointer_t<spv::StorageClassPhysicalStorageBuffer,T> pointer, [[vk::ext_literal]] uint32_t __aligned = /*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment = alignment);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpLoad)]]
+enable_if_t<is_spirv_type_v<P>,T> load(P pointer);
 
-template<typename T, typename P, uint32_t alignment >
+template<typename T, uint32_t alignment>
 [[vk::ext_capability(spv::CapabilityPhysicalStorageBufferAddresses)]]
 [[vk::ext_instruction(spv::OpStore)]]
-enable_if_t<is_spirv_type_v<P>, void> store(P pointer, T obj, [[vk::ext_literal]] uint32_t __aligned = /*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment = alignment);
+void store(pointer_t<spv::StorageClassPhysicalStorageBuffer,T>  pointer, T obj, [[vk::ext_literal]] uint32_t __aligned = /*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment = alignment);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpStore)]]
+enable_if_t<is_spirv_type_v<P>,void> store(P pointer, T obj);
 
 //! Std 450 Extended set operations
 template<typename SquareMatrix>