Implement the `InterlockedAdd` HLSL Function

[farzonl]

• Implement InterlockedAdd clang builtin,
• Link InterlockedAdd clang builtin with hlsl_intrinsics.h
• Add sema checks for InterlockedAdd to CheckHLSLBuiltinFunctionCall in SemaChecking.cpp
• Add codegen for InterlockedAdd to EmitHLSLBuiltinExpr in CGBuiltin.cpp
• Add codegen tests to clang/test/CodeGenHLSL/builtins/InterlockedAdd.hlsl
• Add sema tests to clang/test/SemaHLSL/BuiltIns/InterlockedAdd-errors.hlsl
• Create the int_dx_InterlockedAdd intrinsic in IntrinsicsDirectX.td
• Create the DXILOpMapping of int_dx_InterlockedAdd to 160 in DXIL.td
• Create the InterlockedAdd.ll and InterlockedAdd_errors.ll tests in llvm/test/CodeGen/DirectX/
• Create the int_spv_InterlockedAdd intrinsic in IntrinsicsSPIRV.td
• In SPIRVInstructionSelector.cpp create the InterlockedAdd lowering and map it to int_spv_InterlockedAdd in SPIRVInstructionSelector::selectIntrinsic.
• Create SPIR-V backend test case in llvm/test/CodeGen/SPIRV/hlsl-intrinsics/InterlockedAdd.ll

DirectX

DXIL Opcode	DXIL OpName	Shader Model	Shader Stages
160	CreateHandleForLib	6.3	()

SPIR-V

OpAtomicIAdd:

Description:

Perform the following steps atomically with respect to any other atomic
accesses within Scope to the same location:

1. load through Pointer to get an Original Value,
2. get a New Value by integer addition of Original Value and
   Value, and
3. store the New Value back through Pointer.

The instruction’s result is the Original Value.

Result Type must be an integer type scalar.

The type of Value must be the same as Result Type. The type of the
value pointed to by Pointer must be the same as Result Type.

Memory is a memory Scope.

Word Count	Opcode	Results	Operands
7	234	<id> Result Type	Result <id>	<id> Pointer	Scope <id> Memory	Memory Semantics <id> Semantics	<id> Value

Test Case(s)

Example 1

//dxc InterlockedAdd_test.hlsl -T lib_6_8 -enable-16bit-types -O0

RWStructuredBuffer<int64_t> buffer : register(u0);
[numthreads(1, 1, 1)]
export void fn(uint3 dispatchThreadID : SV_DispatchThreadID, int64_t p1) {
int index = dispatchThreadID.x;
    return InterlockedAdd(buffer[index], p1);
}

Example 2

//dxc InterlockedAdd_1_test.hlsl -T lib_6_8 -enable-16bit-types -O0

RWStructuredBuffer<int64_t> buffer : register(u0);
[numthreads(1, 1, 1)]
export void fn(uint3 dispatchThreadID : SV_DispatchThreadID, int64_t p1, uint64_t p2) {
int index = dispatchThreadID.x;
    return InterlockedAdd(buffer[index], p1, p2);
}

Example 3

//dxc InterlockedAdd_2_test.hlsl -T lib_6_8 -enable-16bit-types -O0

RWStructuredBuffer<int> buffer : register(u0);
[numthreads(1, 1, 1)]
export void fn(uint3 dispatchThreadID : SV_DispatchThreadID, int p1) {
int index = dispatchThreadID.x;
    return InterlockedAdd(buffer[index], p1);
}

Example 4

//dxc InterlockedAdd_3_test.hlsl -T lib_6_8 -enable-16bit-types -O0

RWStructuredBuffer<int> buffer : register(u0);
[numthreads(1, 1, 1)]
export void fn(uint3 dispatchThreadID : SV_DispatchThreadID, int p1, uint p2) {
int index = dispatchThreadID.x;
    return InterlockedAdd(buffer[index], p1, p2);
}

HLSL:

Performs a guaranteed atomic add of value to the dest resource variable.

Syntax

void InterlockedAdd(
  in  R dest,
  in  T value,
  out T original_value
);

Parameters

dest [in]

Type: R

The destination address.

value [in]

Type: T

The input value.

original_value [out]

Type: T

Optional. The original input value.

Return value

This function does not return a value.

Remarks

This operation can only be performed on int or uint typed resources and shared memory variables. There are two possible uses for this function. The first is when R is a shared memory variable type. In this case, the function performs an atomic add of value to the shared memory register referenced by dest. The second scenario is when R is a resource variable type. In this scenario, the function performs an atomic add of value to the resource location referenced by dest. The overloaded function has an additional output variable which will be set to the original value of dest. This overloaded operation is only available when R is readable and writable.

Interlocked operations do not imply any memory fence/barrier.

Minimum Shader Model

This function is supported in the following shader models.

Shader Model	Supported
Shader Model 5 and higher shader models	yes

 

This function is supported in the following types of shaders:

Vertex	Hull	Domain	Geometry	Pixel	Compute
x	x	x	x	x	x

 

See also

Intrinsic Functions

Shader Model 5