Multiply-adds and subtracts packed single-precision floating-point values using three float32 vectors. The corresponding FMA instruction is VFMADDSUB<XXX>PS, where XXX could be 132, 213, or 231.
Syntax
For 128-bit vector
extern __m128 _mm_fmaddsub_ps(__m128 a, __m128 b, __m128 c); |
For 256-bit vector
extern __m256 _mm256_fmaddsub_ps(__m256 a, __m256 b, __m256 c); |
Arguments
a | float32 vector used for the operation |
b | float32 vector also used for the operation |
c | float32 vector also used for the operation |
Description
Performs a set of SIMD multiply-add-subtract computation on packed single-precision floating-point values using three source vectors/operands, a, b, and c. Corresponding values in two operands, a and b, are multiplied and infinite precision intermediate results are obtained. The odd values in the third operand, c, are added to the intermediate results while the even values are subtracted from them. The final results are rounded to the nearest float32 values.
The compiler defaults to using the VFMADDSUB213PS instruction and uses the other forms VFMADDSUB132PS or VFMADDSUB231PS only if a low level optimization decides it is useful or necessary. For example, the compiler could change the default if it finds that another instruction form saves a register or eliminates a move.