Shader Model 6.4

The new shader model adds the following 3 intrinsics to accelerate dot products on reduced precision data:

##Unsigned Integer Dot-Product of 4 Elements and Accumulate: uint32 dot4add_u8packed( uint32 a, uint32 b, uint32 acc ); // ubyte4 a, b;  A 4-dimensional unsigned integer dot-product with add. Multiplies together each corresponding pair of unsigned 8-bit int bytes in the two input DWORDs, and sums the results into the 32-bit unsigned integer accumulator. This instruction operates within a single 32-bit wide SIMD lane. The inputs are also assumed to be 32-bit quantities.   ##Signed Integer Dot-Product of 4 Elements and Accumulate: int32  dot4add_i8packed( uint32 a, uint32 b,  int32 acc );          // signed byte4 a, b;  A 4-dimensional signed integer dot-product with add. Multiplies together each corresponding pair of signed 8-bit int bytes in the two input DWORDs, and sums the results into the 32-bit signed integer accumulator. This instruction operates within a single 32-bit wide SIMD lane. The inputs are also assumed to be 32-bit quantities.   ##Single Precision Floating Point 2-Element Dot-Product and Accumulate: float dot2add( half2 a, half2 b, float acc );  A 2-dimensional floating point dot product of half2 vectors with add. Multiplies the elements of the two half-precision float input vectors together and sums the results into the 32-bit float accumulator. These instructions operate within a single 32-bit wide SIMD lane. The inputs are 16-bit quantities packed into the same lane. This is not considered a ‘fused’ operation, and so need not emit INF due to fp16 overflow unless the [precise] flag is set This covered under the low-precision feature bit (indicating native half and short support).