Changes between Version 18 and Version 19 of GcnTimings


Ignore:
Timestamp:
06/05/16 20:00:50 (8 years ago)
Author:
trac
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GcnTimings

    v18 v19  
    157157<h3>Instruction scheduling</h3>
    158158<ul>
     159<li>if many wavefront executed in single CU (if many wavefronts) then scalar, vector and
     160data-share, memory (???) execution units can run independently (parallely) way,
     161achieving many instructions per cycles.</li>
    159162<li>between any integer V_ADD*, V_SUB*, V_FIRSTREADLINE_B32, V_READLANE_B32 operation
    160163and any scalar ALU instruction is 16-cycle delay. Masked if more waves than 4*CUs</li>
     
    650653<h3>DS Instruction timings</h3>
    651654<p>Timings of DS instructions includes only execution without waiting for completing
    652 LDS/GDS memory access on single wavefront. Timings of DS instructions are in this table:</p>
     655LDS/GDS memory access on single wavefront. Throughput indicates maximal possible
     656throughput that excludes any other delays and penalties.
     657Timings of DS instructions are in this table:</p>
    653658<table>
    654659<thead>