Compare Intrinsics for Floating Point Vectors

This topic only applies to Intel® Many Integrated Core Architecture (Intel® MIC Architecture).

The intrinsic functions to perform compare operations between 512-bit floating point vectors are VCMPPS and VCMPPD for float32 and float64 vectors respectively.

These instructions must comply with 3.1.1.1 of the DX10 specifications, which says:

• The comparisons EQ, GT, GE, LT, and LE; when either or both operands are NaN, returns FALSE.

• Comparisons ignore the sign of 0 (so +0 equals -0).

• The comparison NE, when either or both operands is NaN, returns TRUE.

• Comparisons of any non-NaN value against +/-∞ return the correct result.

Also from the DX10 specifications:

• Denorms are flushed before comparison (original source registers are untouched).

• +0 equals -0. Comparison with NaN returns false.

• Infinity of like signs, are considered equals. Infinity values of either sign are considered ordered values.