NASM 2.05 based x86 Instruction Reference[ch235]
A.5.211 PF2IW: Packed Single-Precision FP to Integer Word Convert PF2IW mm1,mm2/m64 ; 0F 0F /r 1C [PENT,3DNOW] PF2IW converts two single-precision FP values in the source operand to signed 16-bit integers, using truncation, and stores them in the destination operand. Source values that are outside the range supported by the destination are saturated to the largest absolute value of the same sign. - In the K6-2 and K6-III, the 16-bit value is zero-extended to 32-bits before storing. - In the K6-2+, K6-III+ and Athlon processors, the value is sign- extended to 32-bits before storing.