Implement the SetMeshOutputCounts HLSL Function

[farzonl]

• Implement SetMeshOutputCounts clang builtin,
• Link SetMeshOutputCounts clang builtin with hlsl_intrinsics.h
• Add sema checks for SetMeshOutputCounts to CheckHLSLBuiltinFunctionCall in SemaChecking.cpp
• Add codegen for SetMeshOutputCounts to EmitHLSLBuiltinExpr in CGBuiltin.cpp
• Add codegen tests to clang/test/CodeGenHLSL/builtins/SetMeshOutputCounts.hlsl
• Add sema tests to clang/test/SemaHLSL/BuiltIns/SetMeshOutputCounts-errors.hlsl
• Create the int_dx_SetMeshOutputCounts intrinsic in IntrinsicsDirectX.td
• Create the DXILOpMapping of int_dx_SetMeshOutputCounts to 168 in DXIL.td
• Create the SetMeshOutputCounts.ll and SetMeshOutputCounts_errors.ll tests in llvm/test/CodeGen/DirectX/
• Create the int_spv_SetMeshOutputCounts intrinsic in IntrinsicsSPIRV.td
• In SPIRVInstructionSelector.cpp create the SetMeshOutputCounts lowering and map it to int_spv_SetMeshOutputCounts in SPIRVInstructionSelector::selectIntrinsic.
• Create SPIR-V backend test case in llvm/test/CodeGen/SPIRV/hlsl-intrinsics/SetMeshOutputCounts.ll

DirectX

DXIL Opcode	DXIL OpName	Shader Model	Shader Stages
168	SetMeshOutputCounts	6.5	('mesh',)

SPIR-V

OpSetMeshOutputsEXT:

Description:

Reserved.

Capability:
MeshShadingEXT

Reserved.

Word Count	Opcode	Results	Operands
3	5295	<id> Vertex Count	<id> Primitive Count

Test Case(s)

Example 1

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

[numthreads(1, 1, 1)]
[outputtopology("triangle")]
[shader("mesh")]
void fn(in uint gi : SV_GroupIndex, in uint vi : SV_ViewID) {

SetMeshOutputCounts(1, 1);
}

SPIRV Example(s):

Example 2

//dxc SetMeshOutputCounts_spirv_test.hlsl -T lib_6_8 -E fn -enable-16bit-types -spirv -fspv-target-env=universal1.5 -fcgl -O0

struct MeshPerVertex {
	float4 position : SV_Position;
};
[numthreads(1, 1, 1)]
[outputtopology("triangle")]
[shader("mesh")]
void fn(in uint gi : SV_GroupIndex, in uint vi : SV_ViewID,
	out vertices MeshPerVertex verts[3], out indices uint3 primitiveInd[3]) {

SetMeshOutputCounts(1, 1);
}

HLSL:

This function sets the actual number of outputs from the threadgroup.

Syntax

void SetMeshOutputCounts(
    uint numVertices,
    uint numPrimitives);

Remarks

At the beginning of the shader the implementation internally sets a
count of vertices and primitives to be exported from a threadgroup to 0.
It means that if a mesh shader returns without calling this function,
it will not output any mesh. This function sets the actual number of
outputs from the threadgroup.

Some restrictions on the function use and interactions with output arrays follow.

1. This function can only be called once per shader.

2. This call must occur before any writes to any of the
   shared output arrays.
   The validator will verify this is the case.

3. If the compiler can prove that this function is not called,
   then the threadgroup doesn't have any output.
   If the shader writes to any of the
   shared output arrays,
   compilation and shader validation will fail.
   If the shader does not call any of these functions,
   the compiler will issue a warning,
   and no rasterization work will be issued.

4. Only the input values from the first active thread are used.

5. This call must dominate all writes to
   shared output arrays.
   In other words, there must not be any execution path
   that even appears to reach any writes to any output array
   without first having executed this call.