Lerking/ibench
1
0
Fork 0
mirror of https://github.com/RRZE-HPC/ibench.git synced 2025-07-21 21:01:10 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' of github.com:RRZE-HPC/ibench into dev

Browse Source
This commit is contained in:
JanLJL
2025-02-26 11:40:52 +01:00
parent c8e47e3994 a658684a70
commit 9e344c4a75
1 changed files with 15 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
src/AVX-512/vfmadd213sd-zzz.S
View File

@@ -23,17 +23,25 @@ latency:
vpcmpeqw xmm0, xmm0, xmm0 # all ones
vpsllq xmm0, xmm0, 54 # logical left shift: 11111110..0 (54 = 64 - (11 - 1))
vpsrlq xmm0, xmm0, 2 # logical right shift: 1 bit for sign; leading mantissa bit is zero
# expand from SSE to AVX
vinsertf128 ymm0, ymm0, xmm0, 0x1
# expand from AVX to AVX-512
vinsertf64x4 zmm0, zmm0, ymm0, 0x1
# copy SP 1.0
vmovaps xmm1, xmm0
vmovaps zmm1, zmm0
# Mark registers AVX-512
vmovaps zmm0, zmm0
vmovaps zmm1, zmm1
loop:
inc i
INSTR xmm0, xmm1, xmm1
INSTR xmm0, xmm1, xmm1
INSTR xmm0, xmm1, xmm1
INSTR zmm0, zmm1, zmm1
INSTR zmm0, zmm1, zmm1
INSTR zmm0, zmm1, zmm1
cmp i, N
INSTR xmm0, xmm1, xmm1
INSTR xmm0, xmm1, xmm1
INSTR xmm0, xmm1, xmm1
INSTR zmm0, zmm1, zmm1
INSTR zmm0, zmm1, zmm1
INSTR zmm0, zmm1, zmm1
jl loop
done:
mov rsp, rbp