Thanks for using Compiler Explorer
Sponsors
Jakt
C++
Ada
Analysis
Android Java
Android Kotlin
Assembly
C
C3
Carbon
C++ (Circle)
CIRCT
Clean
CMake
CMakeScript
COBOL
C++ for OpenCL
MLIR
Cppx
Cppx-Blue
Cppx-Gold
Cpp2-cppfront
Crystal
C#
CUDA C++
D
Dart
Elixir
Erlang
Fortran
F#
GLSL
Go
Haskell
HLSL
Hook
Hylo
IL
ispc
Java
Julia
Kotlin
LLVM IR
LLVM MIR
Modula-2
Nim
Objective-C
Objective-C++
OCaml
OpenCL C
Pascal
Pony
Python
Racket
Ruby
Rust
Snowball
Scala
Solidity
Spice
Swift
LLVM TableGen
Toit
TypeScript Native
V
Vala
Visual Basic
WASM
Zig
Javascript
GIMPLE
hlsl source #1
Output
Compile to binary object
Link to binary
Execute the code
Intel asm syntax
Demangle identifiers
Verbose demangling
Filters
Unused labels
Library functions
Directives
Comments
Horizontal whitespace
Debug intrinsics
Compiler
Clang (trunk)
DXC (trunk)
DXC 1.6.2112
DXC 1.7.2207
DXC 1.7.2212
DXC 1.7.2308
DXC 1.8.2306-preview
DXC 1.8.2403
DXC 1.8.2403.1
DXC 1.8.2403.2
DXC 1.8.2405
DXC 1.8.2407
RGA 2.6.1 (DXC 1.6.2112)
RGA 2.6.1 (DXC 1.7.2207)
RGA 2.6.2 (DXC 1.6.2112)
RGA 2.6.2 (DXC 1.7.2207)
RGA 2.6.2 (DXC trunk)
RGA 2.9.0 (DXC trunk)
Options
Source code
namespace spirv { template<typename M, typename T, typename StorageClass> [[vk::ext_instruction(/*spv::OpAccessChain*/65)]] vk::SpirvOpaqueType</* OpTypePointer*/ 32,StorageClass,M> accessChain([[vk::ext_reference]] vk::SpirvOpaqueType</* OpTypePointer*/ 32,StorageClass,T> base, [[vk::ext_literal]] uint32_t index0); template<class T, class U> [[vk::ext_instruction(/*spv::OpBitcast*/124)]] T bitcast(U); } namespace bda { template<typename T> using __spv_ptr_t = vk::SpirvOpaqueType< /* OpTypePointer */ 32, /* PhysicalStorageBuffer */ vk::Literal<vk::integral_constant<uint,5349> >, T >; namespace impl { // this only exists to workaround DXC issue XYZW TODO link template<class T> [[vk::ext_capability(/*PhysicalStorageBufferAddresses */ 5347 )]] [[vk::ext_instruction(/*spv::OpBitcast*/124)]] [[vk::ext_decorate(/*AliasedPointer*/5356)]] T bitcast(uint64_t); template<typename T, typename P, uint32_t alignment> [[vk::ext_capability(/*PhysicalStorageBufferAddresses */ 5347 )]] [[vk::ext_instruction( /*OpLoad*/61 )]] T load(P pointer, [[vk::ext_literal]] uint32_t __aligned=/*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment=alignment); template<typename T, typename P, uint32_t alignment> [[vk::ext_capability(/*PhysicalStorageBufferAddresses */ 5347 )]] [[vk::ext_instruction( /*OpStore*/62 )]] void store(P pointer, T obj, [[vk::ext_literal]] uint32_t __aligned=/*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment=alignment); // TODO: atomics for different types template<typename T, typename P> // integers operate on 2s complement so same op for signed and unsigned [[vk::ext_instruction(/*spv::OpAtomicIAdd*/234)]] T atomicIAdd([[vk::ext_decorate(/*AliasedPointer*/5356)]] P ptr, uint32_t memoryScope, uint32_t memorySemantics, T value); } // TODO: maybe make normal and restrict separate distinct types instead of templates template<typename T, bool _restrict=false> struct __ptr; template<typename T, uint32_t alignment, bool _restrict> struct __base_ref { // TODO: // static_assert(alignment>=alignof(T)); // TODO: Some OpDecorateMember Name on the variable would be nice so I don't just see `%spirvIntrinsicType_N` using spv_ptr_t = __spv_ptr_t<T>; [[vk::ext_decorate(/*AliasedPointer*/5356)]] spv_ptr_t ptr; // TODO: Would like to use `spv_ptr_t` or OpAccessChain result instead of `uint64_t` void __init([[vk::ext_decorate(/*AliasedPointer*/5356)]] const spv_ptr_t _ptr) { ptr = _ptr; } __ptr<T,_restrict> addrof() { __ptr<T,_restrict> retval; retval.addr = spirv::bitcast<uint64_t>(ptr); return retval; } // TODO: make load/store `enable_if<nbl::hlsl::is_fundamental_v<T>,return type>` and remove the `#if 0` // NOTE: `enable_if` because of https://github.com/microsoft/DirectXShaderCompiler/issues/6541 T load() { return impl::load<T,__spv_ptr_t<T>,alignment>(ptr); } void store(const T val) { impl::store<T,__spv_ptr_t<T>,alignment>(ptr); } }; template<typename T, uint32_t alignment/*= TODO: alignof(T)*/, bool _restrict=false> struct __ref : __base_ref<T,alignment,_restrict> { using base_t = __base_ref<T,alignment,_restrict>; using this_t = __ref<T,alignment,_restrict>; }; // TODO: specializations for simple builtin types that have atomics template<uint32_t alignment, bool _restrict> struct __ref<uint32_t,alignment,_restrict> : __base_ref<uint32_t,alignment,_restrict> { using base_t = __base_ref<uint32_t,alignment,_restrict>; using this_t = __ref<uint32_t,alignment,_restrict>; // TODO: you need an `init(...)` to foward to `__init(...)` the reason is having one conformant factory for structs later on [[vk::ext_capability(/*PhysicalStorageBufferAddresses */ 5347 )]] uint32_t atomicAdd(const uint32_t value) { return impl::atomicIAdd<uint32_t>(base_t::ptr,/*spv::ScopeDevice*/1,/*spv::MemorySemanticsMaskNone*/0,value); } }; template<typename T, bool _restrict> struct __ptr { using this_t = __ptr<T,_restrict>; uint64_t addr; static this_t create(const uint64_t _addr) { this_t retval; retval.addr = _addr; return retval; } template<uint32_t alignment=4/*TODO: alignof(T)*/> __ref<T,4,_restrict> deref() { // TODO: assert(addr&uint64_t(alignment-1)==0); using retval_t = __ref<T,4,_restrict>; retval_t retval; // TODO: call `init` instead retval.__init(impl::bitcast<typename retval_t::spv_ptr_t>(addr)); return retval; } }; } uint64_t address; struct PSInput { float4 color : COLOR; }; template<typename KeyAccessor> struct counting { void init(const uint64_t address) { key_ptr = KeyAccessor::create(address); } void f() { bda::__ref<uint32_t,4,false> tmp = key_ptr.deref(); tmp.atomicAdd(1); } KeyAccessor key_ptr; }; [numthreads(256,1,1)] void main() { counting<bda::__ptr<uint32_t> > counter; counter.init(address); counter.f(); }
Become a Patron
Sponsor on GitHub
Donate via PayPal
Source on GitHub
Mailing list
Installed libraries
Wiki
Report an issue
How it works
Contact the author
CE on Mastodon
About the author
Statistics
Changelog
Version tree