Re: [Qemu-devel] [v3 1/3] cutils: add avx2 instruction optimization

[Li, Liang Z]

> On 12/08/2015 04:08 AM, Liang Li wrote:
> > +++ b/util/buffer-zero-avx2.c
> > @@ -0,0 +1,54 @@
> > +#include "qemu-common.h"
> > +
> > +#if defined CONFIG_IFUNC && defined CONFIG_AVX2 #include
> > +<immintrin.h>
> > +#define AVX2_VECTYPE        __m256i
> > +#define AVX2_SPLAT(p)       _mm256_set1_epi8(*(p))
> > +#define AVX2_ALL_EQ(v1, v2) \
> > +    (_mm256_movemask_epi8(_mm256_cmpeq_epi8(v1, v2)) ==
> 0xFFFFFFFF)
> > +#define AVX2_VEC_OR(v1, v2) (_mm256_or_si256(v1, v2))
> > +
> > +inline bool
> > +can_use_buffer_find_nonzero_offset_avx2(const void *buf, size_t len)
> > +{
> > +    return (len % (BUFFER_FIND_NONZERO_OFFSET_UNROLL_FACTOR
> > +                   * sizeof(AVX2_VECTYPE)) == 0
> > +            && ((uintptr_t) buf) % sizeof(AVX2_VECTYPE) == 0); }
> 
> I'm not keen on adding a new file for this.  You ought to be able to use
> __attribute__((target("-mavx2"))) on any compiler that supports the
> command-line option.  Which means you can do this all in one file with static
> functions.
> 

I think you means the ' __attribute__((target("avx2")))', I have tried this 
way, the issue here is:
 without the ' -mavx2' option for gcc, there are compiling error:  '__m256i 
undeclared', the __attribute__((target("avx2")))
can't solve this issue.  Any idea?

If I put these avx2 Intrinsics and the sse2 Intrinsics in a single file, the 
sse2  Intrinsics will be compiled to the avx2 instructions, this is not we want.

> Nor am I keen on marking a function inline when we know it must be out-of-
> line because of the ifunc usage.

Inline can be removed.

Thanks 

Liang
> 
> r~