builtin: update inline spirv hlsl #739

alichraghi · 2024-09-03T23:56:10Z

Description

Follow up to Devsh-Graphics-Programming/SPIRV-Headers#3

Testing

Ran arithmetic test

Signed-off-by: Ali Cheraghi <[email protected]>

devshgraphicsprogrammingjenkins · 2024-09-03T23:57:02Z

[CI]: Can one of the admins verify this patch?

devshgraphicsprogramming · 2024-09-06T10:12:28Z

include/nbl/builtin/hlsl/glsl_compat/core.hlsl

+// TODO: redundant T
 template<typename T>
 struct bitfieldExtract<T, true, true>
 {
    static T __call( T val, uint32_t offsetBits, uint32_t numBits )
    {
-        return spirv::bitFieldSExtract<T>( val, offsetBits, numBits );
+        return spirv::bitFieldExtract( val, offsetBits, numBits );
    }
 };

+// TODO: redundant T
 template<typename T>
 struct bitfieldExtract<T, false, true>
 {
    static T __call( T val, uint32_t offsetBits, uint32_t numBits )
    {
-        return spirv::bitFieldUExtract<T>( val, offsetBits, numBits );
+        return spirv::bitFieldExtract( val, offsetBits, numBits );
    } 
 };


Inline SPIR-V should have exact same names as Op enums, except without Op and the leading letter should be lowercase

alos why do you think T is redundant?

P.S. now I realize the overload choice chould have been done nicer with enable_if_t

devshgraphicsprogramming · 2024-09-07T19:01:13Z

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

+
+//! Std 450 Extended set operations
+template<typename SquareMatrix>
+[[vk::ext_instruction(GLSLstd450MatrixInverse)]]


they need an extra parameter microsoft/DirectXShaderCompiler#6751

devshgraphicsprogramming · 2024-09-07T19:02:02Z

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

+template<typename T, typename U>
+[[vk::ext_capability(spv::CapabilityPhysicalStorageBufferAddresses)]]
+[[vk::ext_instruction(spv::OpBitcast)]]
+enable_if_t<is_spirv_type_v<T> && is_spirv_type_v<U>, T> bitcast(U);


hmm you should probably have spirv::is_pointer_v for this

devshgraphicsprogramming · 2024-09-07T19:02:46Z

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

+template<class T, class U>
+[[vk::ext_instruction(spv::OpBitcast)]]
+T bitcast(U);


needs enable_if to check that T is fundamental or a builtin HLSL vector (read the SPIR-V spec) + sizeof(T)==sizeof(U)

devshgraphicsprogramming · 2024-09-07T19:03:12Z

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

+namespace builtin
+{[[vk::ext_builtin_output(spv::BuiltInPosition)]]


crappy formatting, the { should have a line all to itself

devshgraphicsprogramming · 2024-09-07T19:03:27Z

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

 [[vk::ext_builtin_input(spv::BuiltInNumWorkgroups)]]
-static const uint32_t3 NumWorkGroups;
-// TODO: Doesn't work, find out why and file issue on DXC!
-//[[vk::ext_builtin_input(spv::BuiltInWorkgroupSize)]]
-//static const uint32_t3 WorkgroupSize;
+static const uint32_t3 NumWorkgroups;


look at the comment

you mean that WorkgroupSize should be enabled?

that you can't define that builtin because DXC shits its pants

what can i do? add that TODO back in generator?

devshgraphicsprogramming · 2024-09-07T19:04:22Z

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

+//! Execution Modes
+namespace execution_mode
+{
+	void invocations()
+	{
+		vk::ext_execution_mode(spv::ExecutionModeInvocations);
+	}


have you tested this works with godbolt?

AFAIK this only works if you call directly from entry point

nvm we did it this way before, all is good

devshgraphicsprogramming · 2024-09-07T19:07:32Z

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

+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpLoad)]]
+enable_if_t<is_spirv_type_v<P>, T> load(P pointer, [[vk::ext_literal]] uint32_t memoryAccess);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpLoad)]]
+enable_if_t<is_spirv_type_v<P>, T> load(P pointer, [[vk::ext_literal]] uint32_t memoryAccess, [[vk::ext_literal]] uint32_t memoryAccessParam);
+
+template<typename T, typename P, uint32_t alignment>
+[[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);


why do you gen these overloads?

it's defined in the grammer. MemoryAccess operand may have zero or more parameters.

devshgraphicsprogramming · 2024-09-07T19:09:35Z

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

+template<typename T>
+[[vk::ext_instruction(spv::OpLoad)]]
+T load(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, [[vk::ext_literal]] uint32_t memoryAccess);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpLoad)]]
+T load(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, [[vk::ext_literal]] uint32_t memoryAccess, [[vk::ext_literal]] uint32_t memoryAccessParam);
+
+template<typename T, uint32_t alignment>
+[[vk::ext_instruction(spv::OpLoad)]]
+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>
+[[vk::ext_instruction(spv::OpLoad)]]
+T load(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer);


This is a BDA load, and as such:

it will match the loads above because pointer_t is a spirv type, and never get called or worse, clash being ambiguous

BDA load/store ALWAYS take aligned operands and require alignment to be specified (read the spec)

BDA load overloads need to emit the PhysicalStorageBuffer capability

it will match the loads above because pointer_t is a spirv type, and never get called or worse, clash being ambiguous

hmm do you have any idea beside renaming the BDA overload to something else?

devshgraphicsprogramming · 2024-09-07T19:10:04Z

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

+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpStore)]]
+enable_if_t<is_spirv_type_v<P>, void> store(P pointer, T object, [[vk::ext_literal]] uint32_t memoryAccess);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpStore)]]
+enable_if_t<is_spirv_type_v<P>, void> store(P pointer, T object, [[vk::ext_literal]] uint32_t memoryAccess, [[vk::ext_literal]] uint32_t memoryAccessParam);
+
+template<typename T, typename P, uint32_t alignment>
+[[vk::ext_instruction(spv::OpStore)]]
+enable_if_t<is_spirv_type_v<P>, void> store(P pointer, T object, [[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 object);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpStore)]]
+void store(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, T object, [[vk::ext_literal]] uint32_t memoryAccess);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpStore)]]
+void store(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, T object, [[vk::ext_literal]] uint32_t memoryAccess, [[vk::ext_literal]] uint32_t memoryAccessParam);
+
+template<typename T, uint32_t alignment>
+[[vk::ext_instruction(spv::OpStore)]]
+void store(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, T object, [[vk::ext_literal]] uint32_t __aligned = /*Aligned*/0x00000002, [[vk::ext_literal]] uint32_t __alignment = alignment);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpStore)]]
+void store(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer, T object);


same comments as for load

devshgraphicsprogramming · 2024-09-07T19:12:26Z

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

+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpGenericPtrMemSemantics)]]
+enable_if_t<is_spirv_type_v<P>, T> genericPtrMemSemantics(P pointer);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpGenericPtrMemSemantics)]]
+T genericPtrMemSemantics(pointer_t<spv::StorageClassPhysicalStorageBuffer, T> pointer);


don't emit stuff that requires OpenCL / kernel environment

devshgraphicsprogramming · 2024-09-07T19:14:02Z

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

+template<typename T>
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]
+[[vk::ext_instruction(spv::OpBitFieldInsert)]]
+T bitFieldInsert(T base, T insert, uint32_t offset, uint32_t count);
+
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]
+[[vk::ext_instruction(spv::OpBitFieldSExtract)]]
+int32_t bitFieldExtract(int32_t base, uint32_t offset, uint32_t count);
+
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]
+[[vk::ext_instruction(spv::OpBitFieldSExtract)]]
+int64_t bitFieldExtract(int64_t base, uint32_t offset, uint32_t count);
+
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]
+[[vk::ext_instruction(spv::OpBitFieldUExtract)]]
+uint32_t bitFieldExtract(uint32_t base, uint32_t offset, uint32_t count);
+
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]


you can probably skip emitting caps and extensions that DXC would emit anyway when targetting vulkan 1.3

there's a cleaner way to do OpBitFieldUExtract and OpBitFieldSExtract than having every single possible overload imaginable.

template<typename U> enable_if_t<is_unsigned_v<U>,U>

and

template<typename S> enable_if_t<is_signed_v<S>,S>

perhaps. note that we have to emit an overload for 64bit ints/floats anyways

devshgraphicsprogramming · 2024-09-07T19:18:54Z

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

+template<typename T>
+[[vk::ext_capability(spv::CapabilityBitInstructions)]]
+[[vk::ext_instruction(spv::OpBitReverse)]]
+T bitReverse(T base);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpBitCount)]]
+T bitCount(T base);


you probably want to express type constraints with enable_if_t on any templated instruction

devshgraphicsprogramming · 2024-09-07T19:19:43Z

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

+[[vk::ext_instruction(spv::OpControlBarrier)]]
+void controlBarrier(uint32_t executionScope, uint32_t memoryScope,  uint32_t semantics);
+
+[[vk::ext_instruction(spv::OpMemoryBarrier)]]
+void memoryBarrier(uint32_t memoryScope,  uint32_t semantics);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicLoad)]]
+T atomicLoad([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicLoad)]]
+enable_if_t<is_spirv_type_v<P>, T> atomicLoad(P pointer, uint32_t memoryScope,  uint32_t semantics);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicStore)]]
+void atomicStore([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics, T value);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicStore)]]
+enable_if_t<is_spirv_type_v<P>, void> atomicStore(P pointer, uint32_t memoryScope,  uint32_t semantics, T value);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicExchange)]]
+T atomicExchange([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics, T value);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicExchange)]]
+enable_if_t<is_spirv_type_v<P>, T> atomicExchange(P pointer, uint32_t memoryScope,  uint32_t semantics, T value);

-template<typename T> // integers operate on 2s complement so same op for signed and unsigned
-[[vk::ext_capability(spv::CapabilityInt64Atomics)]]
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicCompareExchange)]]
+T atomicCompareExchange([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t equal,  uint32_t unequal, T value, T comparator);


btw does the SPIR-V spec say that any of the memory operands need to be literals or not?

No, just checked

devshgraphicsprogramming · 2024-09-07T19:23:23Z

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

+[[vk::ext_instruction(spv::OpAtomicSMin)]]
+int32_t atomicMin([[vk::ext_reference]] int32_t pointer, uint32_t memoryScope,  uint32_t semantics, int32_t value);

-template<typename T, typename Ptr_T> // DXC Workaround
-[[vk::ext_capability(spv::CapabilityInt64Atomics)]]
-[[vk::ext_instruction(spv::OpAtomicISub)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && (is_same_v<T,uint64_t> || is_same_v<T,int64_t>), T> atomicISub(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+[[vk::ext_instruction(spv::OpAtomicSMin)]]
+int64_t atomicMin([[vk::ext_reference]] int64_t pointer, uint32_t memoryScope,  uint32_t semantics, int64_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicSMin)]]
+enable_if_t<is_spirv_type_v<P>, int32_t> atomicMin(P pointer, uint32_t memoryScope,  uint32_t semantics, int32_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicSMin)]]
+enable_if_t<is_spirv_type_v<P>, int64_t> atomicMin(P pointer, uint32_t memoryScope,  uint32_t semantics, int64_t value);
+
+[[vk::ext_instruction(spv::OpAtomicUMin)]]
+uint32_t atomicMin([[vk::ext_reference]] uint32_t pointer, uint32_t memoryScope,  uint32_t semantics, uint32_t value);
+
+[[vk::ext_instruction(spv::OpAtomicUMin)]]
+uint64_t atomicMin([[vk::ext_reference]] uint64_t pointer, uint32_t memoryScope,  uint32_t semantics, uint64_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicUMin)]]
+enable_if_t<is_spirv_type_v<P>, uint32_t> atomicMin(P pointer, uint32_t memoryScope,  uint32_t semantics, uint32_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicUMin)]]
+enable_if_t<is_spirv_type_v<P>, uint64_t> atomicMin(P pointer, uint32_t memoryScope,  uint32_t semantics, uint64_t value);
+
+[[vk::ext_instruction(spv::OpAtomicSMax)]]
+int32_t atomicMax([[vk::ext_reference]] int32_t pointer, uint32_t memoryScope,  uint32_t semantics, int32_t value);
+
+[[vk::ext_instruction(spv::OpAtomicSMax)]]
+int64_t atomicMax([[vk::ext_reference]] int64_t pointer, uint32_t memoryScope,  uint32_t semantics, int64_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicSMax)]]
+enable_if_t<is_spirv_type_v<P>, int32_t> atomicMax(P pointer, uint32_t memoryScope,  uint32_t semantics, int32_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicSMax)]]
+enable_if_t<is_spirv_type_v<P>, int64_t> atomicMax(P pointer, uint32_t memoryScope,  uint32_t semantics, int64_t value);
+
+[[vk::ext_instruction(spv::OpAtomicUMax)]]
+uint32_t atomicMax([[vk::ext_reference]] uint32_t pointer, uint32_t memoryScope,  uint32_t semantics, uint32_t value);
+
+[[vk::ext_instruction(spv::OpAtomicUMax)]]
+uint64_t atomicMax([[vk::ext_reference]] uint64_t pointer, uint32_t memoryScope,  uint32_t semantics, uint64_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicUMax)]]
+enable_if_t<is_spirv_type_v<P>, uint32_t> atomicMax(P pointer, uint32_t memoryScope,  uint32_t semantics, uint32_t value);
+
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicUMax)]]
+enable_if_t<is_spirv_type_v<P>, uint64_t> atomicMax(P pointer, uint32_t memoryScope,  uint32_t semantics, uint64_t value);


btw the 64bit atomics need a special capability (64bit atomics)

devshgraphicsprogramming · 2024-09-07T19:24:08Z

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

+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicCompareExchange)]]
+enable_if_t<is_spirv_type_v<P>, T> atomicCompareExchange(P pointer, uint32_t memoryScope,  uint32_t equal,  uint32_t unequal, T value, T comparator);
+
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicCompareExchangeWeak)]]
+T atomicCompareExchangeWeak([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t equal,  uint32_t unequal, T value, T comparator);
+
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicCompareExchangeWeak)]]
+enable_if_t<is_spirv_type_v<P>, T> atomicCompareExchangeWeak(P pointer, uint32_t memoryScope,  uint32_t equal,  uint32_t unequal, T value, T comparator);
+


to do atomics of any other size than 4, you need special caps or extensions :(

devshgraphicsprogramming · 2024-09-07T19:25:15Z

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

 template<typename T>
 [[vk::ext_instruction(spv::OpAtomicAnd)]]
-enable_if_t<is_same_v<T,uint32_t> || is_same_v<T,int32_t>, T> atomicAnd([[vk::ext_reference]] T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+T atomicAnd([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics, T value);

-template<typename T, typename Ptr_T> // DXC Workaround
+template<typename T, typename P>
 [[vk::ext_instruction(spv::OpAtomicAnd)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && (is_same_v<T,uint32_t> || is_same_v<T,int32_t>), T> atomicAnd(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+enable_if_t<is_spirv_type_v<P>, T> atomicAnd(P pointer, uint32_t memoryScope,  uint32_t semantics, T value);

 template<typename T>
 [[vk::ext_instruction(spv::OpAtomicOr)]]
-enable_if_t<is_same_v<T,uint32_t> || is_same_v<T,int32_t>, T> atomicOr([[vk::ext_reference]] T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+T atomicOr([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics, T value);

-template<typename T, typename Ptr_T> // DXC Workaround
+template<typename T, typename P>
 [[vk::ext_instruction(spv::OpAtomicOr)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && (is_same_v<T,uint32_t> || is_same_v<T,int32_t>), T> atomicOr(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+enable_if_t<is_spirv_type_v<P>, T> atomicOr(P pointer, uint32_t memoryScope,  uint32_t semantics, T value);

 template<typename T>
 [[vk::ext_instruction(spv::OpAtomicXor)]]
-enable_if_t<is_same_v<T,uint32_t> || is_same_v<T,int32_t>, T> atomicXor([[vk::ext_reference]] T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+T atomicXor([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics, T value);

-template<typename T, typename Ptr_T> // DXC Workaround
+template<typename T, typename P>
 [[vk::ext_instruction(spv::OpAtomicXor)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && (is_same_v<T,uint32_t> || is_same_v<T,int32_t>), T> atomicXor(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+enable_if_t<is_spirv_type_v<P>, T> atomicXor(P pointer, uint32_t memoryScope,  uint32_t semantics, T value);


same as with the compare or swap atomics, caps/extensions for sizes other than 32bit

devshgraphicsprogramming · 2024-09-07T19:26:03Z

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

+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicFlagTestAndSet)]]
+T atomicFlagTestAndSet([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics);

-template<typename Signed, typename Ptr_T> // DXC Workaround
-[[vk::ext_instruction(spv::OpAtomicSMin)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && is_same_v<Signed,int32_t>, Signed> atomicSMin(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, Signed value);
+template<typename T, typename P>
+[[vk::ext_instruction(spv::OpAtomicFlagTestAndSet)]]
+enable_if_t<is_spirv_type_v<P>, T> atomicFlagTestAndSet(P pointer, uint32_t memoryScope,  uint32_t semantics);

-template<typename Unsigned>
-[[vk::ext_instruction( spv::OpAtomicUMin )]]
-enable_if_t<is_same_v<Unsigned,uint32_t>, Unsigned> atomicUMin([[vk::ext_reference]] Unsigned ptr, uint32_t memoryScope, uint32_t memorySemantics, Unsigned value);
+template<typename T>
+[[vk::ext_instruction(spv::OpAtomicFlagClear)]]
+void atomicFlagClear([[vk::ext_reference]] T pointer, uint32_t memoryScope,  uint32_t semantics);

-template<typename Unsigned, typename Ptr_T> // DXC Workaround
-[[vk::ext_instruction(spv::OpAtomicUMin)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && is_same_v<Unsigned,uint32_t>, Unsigned> atomicUMin(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, Unsigned value);
+template<typename P>
+[[vk::ext_instruction(spv::OpAtomicFlagClear)]]
+enable_if_t<is_spirv_type_v<P>, void> atomicFlagClear(P pointer, uint32_t memoryScope,  uint32_t semantics);


SPIR-V non OpenCL env allows for flag ops?

also I'm sure there would be some type constraint on T

devshgraphicsprogramming · 2024-09-07T19:26:58Z

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

+[[vk::ext_capability(spv::CapabilityGroupNonUniform)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformElect)]]
+bool groupNonUniformElect(uint32_t executionScope);

-template<typename Signed, typename Ptr_T> // DXC Workaround
-[[vk::ext_instruction(spv::OpAtomicSMax)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && is_same_v<Signed,int32_t>, Signed> atomicSMax(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, Signed value);
+[[vk::ext_capability(spv::CapabilityGroupNonUniformVote)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformAll)]]
+bool groupNonUniformAll(uint32_t executionScope, bool predicate);

-template<typename Unsigned>
-[[vk::ext_instruction( spv::OpAtomicUMax )]]
-enable_if_t<is_same_v<Unsigned,uint32_t>, Unsigned> atomicUMax([[vk::ext_reference]] uint32_t ptr, uint32_t memoryScope, uint32_t memorySemantics, Unsigned value);
+[[vk::ext_capability(spv::CapabilityGroupNonUniformVote)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformAny)]]
+bool groupNonUniformAny(uint32_t executionScope, bool predicate);

-template<typename Unsigned, typename Ptr_T> // DXC Workaround
-[[vk::ext_instruction(spv::OpAtomicUMax)]]
-enable_if_t<is_spirv_type_v<Ptr_T> && is_same_v<Unsigned,uint32_t>, Unsigned> atomicUMax(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, Unsigned value);
+[[vk::ext_capability(spv::CapabilityGroupNonUniformVote)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformAllEqual)]]
+bool groupNonUniformAllEqual(uint32_t executionScope, bool value);

 template<typename T>
-[[vk::ext_instruction(spv::OpAtomicExchange)]]
-T atomicExchange([[vk::ext_reference]] T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBroadcast)]]
+T groupNonUniformBroadcast(uint32_t executionScope, T value, uint32_t id);

-template<typename T, typename Ptr_T> // DXC Workaround
-[[vk::ext_instruction(spv::OpAtomicExchange)]]
-enable_if_t<is_spirv_type_v<Ptr_T>, T> atomicExchange(Ptr_T ptr, uint32_t memoryScope, uint32_t memorySemantics, T value);
+template<typename T>
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBroadcastFirst)]]
+T groupNonUniformBroadcastFirst(uint32_t executionScope, T value);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBallot)]]
+uint32_t4 groupNonUniformBallot(uint32_t executionScope, bool predicate);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformInverseBallot)]]
+bool groupNonUniformInverseBallot(uint32_t executionScope, uint32_t4 value);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBallotBitExtract)]]
+bool groupNonUniformBallotBitExtract(uint32_t executionScope, uint32_t4 value, uint32_t index);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBallotBitCount)]]
+uint32_t groupNonUniformBallotBitCount(uint32_t executionScope, [[vk::ext_literal]] uint32_t operation, uint32_t4 value);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBallotFindLSB)]]
+uint32_t groupNonUniformBallotFindLSB(uint32_t executionScope, uint32_t4 value);
+
+[[vk::ext_capability(spv::CapabilityGroupNonUniformBallot)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformBallotFindMSB)]]
+uint32_t groupNonUniformBallotFindMSB(uint32_t executionScope, uint32_t4 value);

 template<typename T>
-[[vk::ext_instruction(spv::OpAtomicCompareExchange)]]
-T atomicCompareExchange([[vk::ext_reference]] T ptr, uint32_t memoryScope, uint32_t memSemanticsEqual, uint32_t memSemanticsUnequal, T value, T comparator);
+[[vk::ext_capability(spv::CapabilityGroupNonUniformShuffle)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformShuffle)]]
+T groupNonUniformShuffle(uint32_t executionScope, T value, uint32_t id);

-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>
+[[vk::ext_capability(spv::CapabilityGroupNonUniformShuffle)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformShuffleXor)]]
+T groupNonUniformShuffleXor(uint32_t executionScope, T value, uint32_t mask);

+template<typename T>
+[[vk::ext_capability(spv::CapabilityGroupNonUniformShuffleRelative)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformShuffleUp)]]
+T groupNonUniformShuffleUp(uint32_t executionScope, T value, uint32_t delta);

-template<typename T, uint32_t alignment>
-[[vk::ext_capability(spv::CapabilityPhysicalStorageBufferAddresses)]]
-[[vk::ext_instruction(spv::OpLoad)]]
-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>
+[[vk::ext_capability(spv::CapabilityGroupNonUniformShuffleRelative)]]
+[[vk::ext_instruction(spv::OpGroupNonUniformShuffleDown)]]
+T groupNonUniformShuffleDown(uint32_t executionScope, T value, uint32_t delta);


does execution scope need to be a literal?

Nop. just checked

devshgraphicsprogramming · 2024-09-07T19:29:31Z

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

+[[vk::ext_capability(spv::CapabilityAtomicFloat16MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+float atomicMinEXT_AtomicFloat16MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat32MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+float atomicMinEXT_AtomicFloat32MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat64MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+float atomicMinEXT_AtomicFloat64MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);


wrong types, first should be float16_t second should be float32_t, last should be float64_t

devshgraphicsprogramming · 2024-09-07T19:30:07Z

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

+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+float atomicMinEXT_AtomicFloat16VectorNV([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);


this would be for a vector<float16_t,N> not float

devshgraphicsprogramming · 2024-09-07T19:30:28Z

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

+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMinEXT_AtomicFloat16MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat32MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMinEXT_AtomicFloat32MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat64MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMinEXT_AtomicFloat64MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFMinEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMinEXT_AtomicFloat16VectorNV(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);


same issue as with the non template P overloads

devshgraphicsprogramming · 2024-09-07T19:30:47Z

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

+[[vk::ext_capability(spv::CapabilityAtomicFloat16MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+float atomicMaxEXT_AtomicFloat16MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat32MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+float atomicMaxEXT_AtomicFloat32MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat64MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+float atomicMaxEXT_AtomicFloat64MinMaxEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+float atomicMaxEXT_AtomicFloat16VectorNV([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMaxEXT_AtomicFloat16MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);

-template<typename Signed>
-[[vk::ext_instruction( spv::OpBitFieldSExtract )]]
-enable_if_t<is_signed_v<Signed>, Signed> bitFieldSExtract( Signed val, uint32_t offsetBits, uint32_t numBits );
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat32MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMaxEXT_AtomicFloat32MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);

-template<typename Integral>
-[[vk::ext_instruction( spv::OpBitFieldInsert )]]
-Integral bitFieldInsert( Integral base, Integral insert, uint32_t offset, uint32_t count );
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat64MinMaxEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMaxEXT_AtomicFloat64MinMaxEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFMaxEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicMaxEXT_AtomicFloat16VectorNV(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);


same issues as with your min

devshgraphicsprogramming · 2024-09-07T19:31:09Z

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

+[[vk::ext_capability(spv::CapabilityAtomicFloat16AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+float atomicAddEXT_AtomicFloat16AddEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat32AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+float atomicAddEXT_AtomicFloat32AddEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat64AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+float atomicAddEXT_AtomicFloat64AddEXT([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+float atomicAddEXT_AtomicFloat16VectorNV([[vk::ext_reference]] float pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicAddEXT_AtomicFloat16AddEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat32AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicAddEXT_AtomicFloat32AddEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat64AddEXT)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicAddEXT_AtomicFloat64AddEXT(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);
+
+template<typename P>
+[[vk::ext_capability(spv::CapabilityAtomicFloat16VectorNV)]]
+[[vk::ext_instruction(spv::OpAtomicFAddEXT)]]
+enable_if_t<is_spirv_type_v<P>, float> atomicAddEXT_AtomicFloat16VectorNV(P pointer, uint32_t memoryScope,  uint32_t semantics, float value);


same float sizing issues as with the min-max stuff

devshgraphicsprogramming · 2024-09-07T19:31:18Z

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

+[[vk::ext_capability(spv::CapabilitySplitBarrierINTEL)]]
+[[vk::ext_instruction(spv::OpControlBarrierArriveINTEL)]]
+void controlBarrierArriveINTEL(uint32_t executionScope, uint32_t memoryScope,  uint32_t semantics);
+
+[[vk::ext_capability(spv::CapabilitySplitBarrierINTEL)]]
+[[vk::ext_instruction(spv::OpControlBarrierWaitINTEL)]]
+void controlBarrierWaitINTEL(uint32_t executionScope, uint32_t memoryScope,  uint32_t semantics);


these are OpenCL AFAIK

devshgraphicsprogramming · 2024-09-07T19:32:36Z

include/nbl/builtin/hlsl/spirv_intrinsics/fragment_shader_pixel_interlock.hlsl

-[[vk::ext_capability(spv::CapabilityFragmentShaderPixelInterlockEXT)]]
-[[vk::ext_extension("SPV_EXT_fragment_shader_interlock")]]
-[[vk::ext_instruction(spv::OpBeginInvocationInterlockEXT)]]
-void beginInvocationInterlockEXT();
-
-[[vk::ext_capability(spv::CapabilityFragmentShaderPixelInterlockEXT)]]
-[[vk::ext_extension("SPV_EXT_fragment_shader_interlock")]]
-[[vk::ext_instruction(spv::OpEndInvocationInterlockEXT)]]
-void endInvocationInterlockEXT();


these are missing in your new codegen, please test with example 62 CAD as well as arithmetic unit test

devshgraphicsprogramming · 2024-09-07T19:33:04Z

include/nbl/builtin/hlsl/spirv_intrinsics/subgroup_arithmetic.hlsl

-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformIAdd  )]]
-int32_t groupAdd(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, int32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformIAdd  )]]
-uint32_t groupAdd(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, uint32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformFAdd  )]]
-float32_t groupAdd(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, float32_t value);
-
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformIMul )]]
-int32_t groupMul(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, int32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformIMul )]]
-uint32_t groupMul(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, uint32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformFMul )]]
-float32_t groupMul(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, float32_t value);
-
-template<typename T>
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformBitwiseAnd )]]
-T groupBitwiseAnd(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, T value);
-
-template<typename T>
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformBitwiseOr )]]
-T groupBitwiseOr(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, T value);
-
-template<typename T>
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformBitwiseXor )]]
-T groupBitwiseXor(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, T value);
-
-// The MIN and MAX operations in SPIR-V have different Ops for each arithmetic type
-// so we implement them distinctly
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformSMin )]]
-int32_t groupBitwiseMin(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, int32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformUMin )]]
-uint32_t groupBitwiseMin(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, uint32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformFMin )]]
-float32_t groupBitwiseMin(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, float32_t value);
-
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformSMax )]]
-int32_t groupBitwiseMax(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, int32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformUMax )]]
-uint32_t groupBitwiseMax(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, uint32_t value);
-[[vk::ext_capability( spv::CapabilityGroupNonUniformArithmetic )]]
-[[vk::ext_instruction( spv::OpGroupNonUniformFMax )]]
-float32_t groupBitwiseMax(uint32_t groupScope, [[vk::ext_literal]] uint32_t operation, float32_t value);


@kpentaris oogle this if the new codegen looks sane

Signed-off-by: Ali Cheraghi <[email protected]>

src/nbl/builtin/CMakeLists.txt

Signed-off-by: Ali Cheraghi <[email protected]>

builtin: update inline spirv hlsl

d94c9cc

Signed-off-by: Ali Cheraghi <[email protected]>

devshgraphicsprogramming reviewed Sep 6, 2024

View reviewed changes

devshgraphicsprogramming reviewed Sep 7, 2024

View reviewed changes

builtin: remove wrong comment

6d51fca

Signed-off-by: Ali Cheraghi <[email protected]>

builtin: temp cmake

76c1617

Signed-off-by: Ali Cheraghi <[email protected]>

alichraghi force-pushed the master branch from 47721ba to 76c1617 Compare September 9, 2024 17:04

builtin: update glsl_compat layer to represent new spirv_intrisics

65064e8

Signed-off-by: Ali Cheraghi <[email protected]>

alichraghi force-pushed the master branch from ca1ae43 to 65064e8 Compare September 10, 2024 10:59

builtin: saner CMakeLists.txt (still not working)

16fc27d

Signed-off-by: Ali Cheraghi <[email protected]>

alichraghi force-pushed the master branch from c6a7ba9 to 16fc27d Compare September 10, 2024 15:11

AnastaZIuk reviewed Sep 11, 2024

View reviewed changes

src/nbl/builtin/CMakeLists.txt Outdated Show resolved Hide resolved

builtin: fix cmake

f816d46

Signed-off-by: Ali Cheraghi <[email protected]>

alichraghi force-pushed the master branch from 8e11f47 to f816d46 Compare September 11, 2024 21:41

		namespace builtin
		{[[vk::ext_builtin_output(spv::BuiltInPosition)]]

builtin: update inline spirv hlsl #739

Are you sure you want to change the base?

builtin: update inline spirv hlsl #739

Conversation

alichraghi commented Sep 3, 2024

Description

Testing

devshgraphicsprogrammingjenkins commented Sep 3, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

devshgraphicsprogramming Sep 7, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

devshgraphicsprogramming Sep 7, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

devshgraphicsprogramming Sep 7, 2024 •

edited

Loading

devshgraphicsprogramming Sep 7, 2024 •

edited

Loading