Implement the f32tof16 HLSL Function

[farzonl]

• Implement f32tof16 clang builtin,
• Link f32tof16 clang builtin with hlsl_intrinsics.h
• Add sema checks for f32tof16 to CheckHLSLBuiltinFunctionCall in SemaChecking.cpp
• Add codegen for f32tof16 to EmitHLSLBuiltinExpr in CGBuiltin.cpp
• Add codegen tests to clang/test/CodeGenHLSL/builtins/f32tof16.hlsl
• Add sema tests to clang/test/SemaHLSL/BuiltIns/f32tof16-errors.hlsl
• Create the int_dx_f32tof16 intrinsic in IntrinsicsDirectX.td
• Create the DXILOpMapping of int_dx_f32tof16 to 130 in DXIL.td
• Create the f32tof16.ll and f32tof16_errors.ll tests in llvm/test/CodeGen/DirectX/
• Create the int_spv_f32tof16 intrinsic in IntrinsicsSPIRV.td
• In SPIRVInstructionSelector.cpp create the f32tof16 lowering and map it to int_spv_f32tof16 in SPIRVInstructionSelector::selectIntrinsic.
• Create SPIR-V backend test case in llvm/test/CodeGen/SPIRV/hlsl-intrinsics/f32tof16.ll

DirectX

DXIL Opcode	DXIL OpName	Shader Model	Shader Stages
130	LegacyF32ToF16	6.0	()

SPIR-V

PackHalf2x16:

Description:

PackHalf2x16

Result is the unsigned integer obtained by converting the components of
a two-component floating-point vector to the 16-bit OpTypeFloat, and
then packing these two 16-bit integers into a 32-bit unsigned integer.
The first vector component specifies the 16 least-significant bits of
the result; the second component specifies the 16 most-significant
bits.

The v operand must be a vector of 2 components whose type is a 32-bit
floating-point.

Result Type must be a 32-bit integer type.

Number	Operand 1	Operand 2	Operand 3	Operand 4
58	<id> v

Test Case(s)

Example 1

//dxc f32tof16_test.hlsl -T lib_6_8 -enable-16bit-types -O0

export uint4 fn(float4 p1) {
    return f32tof16(p1);
}

HLSL:

Converts an input into a float16 type.

Syntax

uint f32tof16(
  in float value
);

Parameters

value [in]

Type: float

The input value.

Return value

Type: uint

The converted value, stored in the low-half of the uint.

Remarks

The following overloaded versions are also available:

uint2 f32tof16(float2 value);
uint3 f32tof16(float3 value);
uint4 f32tof16(float4 value);

Minimum Shader Model

This function is supported in the following shader models.

Shader Model	Supported
Shader Model 4 and higher shader models	yes

 

Although this function is available in Shader Model 4 and higher shader models, because it is emulated in 4.0 and 4.1, it is less perfomant on these shader models than it is on Shader Model 5.

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

[farzonl]

@damyanp Adding refinment priority. This is used in DirectML shaders PixelPacking_Velocity.hlsli