SIMD signum function
by Richard Russell, 30th December 2014 (based on a blog post by Raymond Chen)
The signum function returns -1 for all negative numbers, +1 for all positive non-zero numbers and zero for an input of zero; it corresponds to the SGN() function in BBC BASIC. There is no direct equivalent in assembly language, but the following code snippets illustrate how it may be calculated on multiple data values using MMX and SSE2 code. In all cases the input is in mm0 or xmm0 and the output is in mm1 or xmm1:
Signum of eight 8-bit values
pxor mm1, mm1
pxor mm2, mm2
pcmpgtb mm1, mm0 ; mm1 = pcmpgt(0, x)
pcmpgtb mm0, mm2 ; mm0 = pcmpgt(x, 0)
psubb mm1, mm0 ; mm1 = signum
Signum of four 16-bit values
pxor mm1, mm1
pxor mm2, mm2
pcmpgtw mm1, mm0 ; mm1 = pcmpgt(0, x)
pcmpgtw mm0, mm2 ; mm0 = pcmpgt(x, 0)
psubw mm1, mm0 ; mm1 = signum
Signum of two 32-bit values
pxor mm1, mm1
pxor mm2, mm2
pcmpgtd mm1, mm0 ; mm1 = pcmpgt(0, x)
pcmpgtd mm0, mm2 ; mm0 = pcmpgt(x, 0)
psubd mm1, mm0 ; mm1 = signum
Signum of sixteen 8-bit values (requires ASMLIB2 library)
pxor xmm1, xmm1
pxor xmm2, xmm2
pcmpgtb xmm1, xmm0 ; xmm1 = pcmpgt(0, x)
pcmpgtb xmm0, xmm2 ; xmm0 = pcmpgt(x, 0)
psubb xmm1, xmm0 ; xmm1 = signum
Signum of eight 16-bit values (requires ASMLIB2 library)
pxor xmm1, xmm1
pxor xmm2, xmm2
pcmpgtw xmm1, xmm0 ; xmm1 = pcmpgt(0, x)
pcmpgtw xmm0, xmm2 ; xmm0 = pcmpgt(x, 0)
psubw xmm1, xmm0 ; xmm1 = signum
Signum of four 32-bit values (requires ASMLIB2 library)
pxor xmm1, xmm1
pxor xmm2, xmm2
pcmpgtd xmm1, xmm0 ; xmm1 = pcmpgt(0, x)
pcmpgtd xmm0, xmm2 ; xmm0 = pcmpgt(x, 0)
psubd xmm1, xmm0 ; xmm1 = signum