floating point cmp, bit twiddling expertise needed

Index > Main > floating point cmp, bit twiddling expertise needed

Author

Thread

tthsqe

Joined: 20 May 2009
Posts: 767

tthsqe 01 Dec 2013, 11:54

We can determine if two 128-bit sse registers xmm0 and xmm1 have exactly the same bit pattern with

Code:

pcmpeqb xmm0,xmm1
pmovmskb  eax,xmm0
cmp  ax,-1

but unfortunately this cannot be extended to 256 bits because vpcmpeqb is not supported on 256-bit operands in AVX.
Is it possible to get the same functionality in AVX by using the floating point operations? (The floating point ops are supported in AVX.)

Code:

vcmpeqpd ymm2,ymm0,ymm1
vmovmskpd  eax,ymm2
cmp  eax,01111b

This doesn't work because 0x8000000000000000, and 0x0000000000000000 are treated as equal. Sad

01 Dec 2013, 11:54

cod3b453

Joined: 25 Aug 2004
Posts: 618

cod3b453 01 Dec 2013, 13:38

I don't have a processor with AVX to try this but maybe:

Code:

vptest ymm0,ymm1
j[c/z] ...

01 Dec 2013, 13:38

tthsqe

Joined: 20 May 2009
Posts: 767

tthsqe 01 Dec 2013, 14:03

ah, this is very good! Thanks, cod3b453! I completely forgot about ptest.
Actually, I think the following are correct

Code:

; zero test
; if (ymm0=0), then a, else b
 vptest  ymm0,qqword[mask]
     jz  a
b:
a:

; another zero test
; if (ymm0=0), then a, else b
 vxorps  ymm1,ymm1,ymm1
 vptest  ymm1,ymm0
     jc  a
b:
a:

; equality test
; if (ymm0=ymm1), then a, else b
 vxorps  ymm2,ymm0,ymm1
 vptest  ymm2,qqword[mask]
     jz  a
b:
a:

align 32
mask: dq -1,-1,-1,-1

01 Dec 2013, 14:03

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum