ivampiresp/Spdk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvme: remove avx optimizations when copying command

Browse Source 
Using AVX512 or AVX2 ends up being a small pessimization.
I think AVX works better for copies when there are
multiple cachelines to copy.  I see a 2-3% improvement
in high IOPs benchmarks when reverting to SSE.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I3d70a1e359e98cec2a9da41ccf9af2de9baa5868

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456247
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This commit is contained in:
Jim Harris 2019-05-29 02:34:28 -07:00
parent c85164bd69
commit d09874f3a2
1 changed files with 1 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
lib/nvme/nvme_pcie.c
View File

@ -1095,18 +1095,7 @@ static inline void
nvme_pcie_copy_command(struct spdk_nvme_cmd *dst, const struct spdk_nvme_cmd *src)
{
/* dst and src are known to be non-overlapping and 64-byte aligned. */
#if defined(__AVX512F__)
__m512i *d512 = (__m512i *)dst;
const __m512i *s512 = (const __m512i *)src;
_mm512_stream_si512(d512, _mm512_load_si512(s512));
#elif defined(__AVX__)
__m256i *d256 = (__m256i *)dst;
const __m256i *s256 = (const __m256i *)src;
_mm256_stream_si256(&d256[0], _mm256_load_si256(&s256[0]));
_mm256_stream_si256(&d256[1], _mm256_load_si256(&s256[1]));
#elif defined(__SSE2__)
#if defined(__SSE2__)
__m128i *d128 = (__m128i *)dst;
const __m128i *s128 = (const __m128i *)src;