[HN Gopher] A story of a large loop with a long instruction depe...
       ___________________________________________________________________
        
       A story of a large loop with a long instruction dependency chain
        
       Author : signa11
       Score  : 19 points
       Date   : 2024-03-01 14:38 UTC (8 hours ago)
        
 (HTM) web link (johnnysswlab.com)
 (TXT) w3m dump (johnnysswlab.com)
        
       | robinhouston wrote:
       | Since the site seems to have been hugged to death, here's an
       | archive link:
       | https://web.archive.org/web/20240229063944/https://johnnyssw...
        
         | temende wrote:
         | At least now they're not exceeding their bandwidth cap ;)
        
       | jart wrote:
       | I know when I was coding an avx2 matmul last month, having
       | multiple dot product dependency chains operating in parallel was
       | the single biggest thing that brought it in the same league of
       | performance as MKL. It was like a night and day difference the
       | first time I ran the program after doing that. Using lookaside L1
       | cache didn't help me very much, since it worked better to share
       | register loads across operations.
        
       ___________________________________________________________________
       (page generated 2024-03-01 23:01 UTC)