ywkang/kernbench2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
533e6992992482eec5d64d74d1d6b8dcd4c0c4d1
kernbench2/docs/diagrams/pe2pe_latency_plots/summary.csv
T
mukesh 1c5752a9ec Intercube allreduce: center root + bidirectional reduce 
Move the algorithmic root cube from the corner (cube_w-1,
cube_h-1) to the geometric center (cube_w//2, cube_h//2) and
have each phase converge bidirectionally so the intra-SIP
critical path drops from ~12 hops to ~8 hops on a 4×4 mesh
(left half W→E + right half E→W in row reduce; top half N→S +
bottom half S→N in col reduce; mirrored on broadcast).

Result on torus_2d 6 SIPs at 96 KB / PE on TCM:
  before (corner root)  : 22.0 µs
  after  (center root)  : 17.2 µs   (−22%)

Same shape on ring_1d (−7%) and mesh_2d_no_wrap (−12%); also
holds across SRAM and HBM (~−20% each).

Phase 1 test (test_intercube_root_center.py) asserts the
torus_2d 96 KB latency drops below 20.5 µs and that all 96
cubes still validate (correctness preserved).

Plot updates:
- overview.png: replace constant 10.6 µs theoretical line with
  user-supplied hand-derived curve (per-cube packet count =
  bytes_per_pe × 8 PEs ÷ 128 B; 1346 ns startup + 1.20 ns/pkt).
- All summary.csv numbers and per-topology PNGs regenerated.
- pe2pe_latency_plots and ipcq diagram emitter PNGs refreshed.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-27 21:28:58 -07:00

82 lines
6.6 KiB
CSV
Raw Blame History

hop,label,size_bytes,path,total_ns
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),128,ipcq,31.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),128,raw,12.019999999996799
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),256,ipcq,33.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),256,raw,13.019999999996799
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),384,ipcq,35.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),384,raw,14.019999999996799
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),512,ipcq,37.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),512,raw,15.019999999996799
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),768,ipcq,41.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),768,raw,17.0199999999968
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),1024,ipcq,45.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),1024,raw,19.0199999999968
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),2048,ipcq,61.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),2048,raw,27.0199999999968
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),4096,ipcq,93.6399999999976
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),4096,raw,43.0199999999968
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),8192,ipcq,157.64000000000306
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),8192,raw,75.02000000000407
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),10240,ipcq,189.64000000000306
h1_intra_horizontal,Intra-cube horizontal (pe0 to pe1),10240,raw,91.02000000000407
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),128,ipcq,31.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),128,raw,12.019999999996799
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),256,ipcq,33.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),256,raw,13.019999999996799
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),384,ipcq,35.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),384,raw,14.019999999996799
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),512,ipcq,37.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),512,raw,15.019999999996799
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),768,ipcq,41.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),768,raw,17.0199999999968
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),1024,ipcq,45.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),1024,raw,19.0199999999968
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),2048,ipcq,61.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),2048,raw,27.0199999999968
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),4096,ipcq,93.6399999999976
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),4096,raw,43.0199999999968
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),8192,ipcq,157.64000000000306
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),8192,raw,75.02000000000407
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),10240,ipcq,189.64000000000306
h2_intra_vertical,Intra-cube vertical (pe0 to pe4),10240,raw,91.02000000000407
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),128,ipcq,67.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),128,raw,68.53999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),256,ipcq,69.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),256,raw,70.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),384,ipcq,71.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),384,raw,71.53999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),512,ipcq,73.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),512,raw,73.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),768,ipcq,77.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),768,raw,76.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),1024,ipcq,81.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),1024,raw,79.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),2048,ipcq,97.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),2048,raw,91.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),4096,ipcq,129.65999999999804
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),4096,raw,115.03999999999724
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),8192,ipcq,193.65999999999985
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),8192,raw,163.04000000000087
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),10240,ipcq,225.65999999999985
h3_inter_cube_horizontal,Inter-cube horizontal (cube0 to cube1),10240,raw,187.04000000000087
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),128,ipcq,87.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),128,raw,88.53999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),256,ipcq,89.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),256,raw,90.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),384,ipcq,91.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),384,raw,91.53999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),512,ipcq,93.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),512,raw,93.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),768,ipcq,97.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),768,raw,96.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),1024,ipcq,101.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),1024,raw,99.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),2048,ipcq,117.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),2048,raw,111.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),4096,ipcq,149.65999999999804
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),4096,raw,135.03999999999724
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),8192,ipcq,213.65999999999985
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),8192,raw,183.04000000000087
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),10240,ipcq,245.65999999999985
h4_inter_cube_vertical,Inter-cube vertical (cube0 to cube4),10240,raw,207.04000000000087
Reference in New Issue View Git Blame Copy Permalink