Mamba2 fused intra chunk diagonal computation
Fused intra-chunk diagonal computation for Mamba2 SSD that combines segment_sum for causal masking, B/C matrix contraction to form attention-like weights G, and application of exponential decay mask L to compute Y_diag. This is the most memory-intensive part of the chunked parallel scan.
Reported evidence · last observed 2026-07-10. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 14 cases · mean latencyhead_dim = 128 · n_groups = 8 · num_heads = 32 · chunk_size = 128 · state_size = 12814 cases
Not measured on H100 for this workload. Challenges →
112.2µs89% SOL1.00×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
212.5µs89% SOL1.03×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
313.3µs88% SOL1.09×Reported · license unknown · no source2026-08-16
Measured exactly what you asked. Reported by source; not independently reproduced.
413.5µs87% SOL1.11×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
513.5µs88% SOL1.11×Reported · license unknown · no source2026-07-12
Measured exactly what you asked. Reported by source; not independently reproduced.
613.7µs87% SOL1.13×Reported · license unknown · no source2026-07-27
Measured exactly what you asked. Reported by source; not independently reproduced.
714.9µs86% SOL1.22×Reported · license unknown · no source2026-07-27
Measured exactly what you asked. Reported by source; not independently reproduced.
816.7µs84% SOL1.37×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
917.1µs83% SOL1.41×Reported · license unknown · no source2026-08-22
Measured exactly what you asked. Reported by source; not independently reproduced.
1017.2µs83% SOL1.42×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
1117.3µs83% SOL1.42×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1217.9µs82% SOL1.47×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1318.2µs82% SOL1.50×Reported · license unknown · no source2026-08-16
Measured exactly what you asked. Reported by source; not independently reproduced.
1418.5µs82% SOL1.52×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.
1519.0µs81% SOL1.56×Reported · license unknown · no source2026-08-13
Measured exactly what you asked. Reported by source; not independently reproduced.
1620.0µs80% SOL1.65×Reported · license unknown · no source2026-08-12
Measured exactly what you asked. Reported by source; not independently reproduced.
1720.1µs80% SOL1.65×Reported · license unknown · no source2026-08-12
Measured exactly what you asked. Reported by source; not independently reproduced.
1820.1µs80% SOL1.65×Reported · license unknown · no source2026-08-12
Measured exactly what you asked. Reported by source; not independently reproduced.
1921.2µs79% SOL1.75×Reported · license unknown · no source2026-07-16
Measured exactly what you asked. Reported by source; not independently reproduced.
2021.3µs80% SOL1.75×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
2121.7µs78% SOL1.79×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
2221.8µs78% SOL1.80×Reported · license unknown · no source2026-07-24
Measured exactly what you asked. Reported by source; not independently reproduced.
2335.5µs67% SOL2.92×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
2438.5µs65% SOL3.16×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
25109.1µs39% SOL8.97×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
26120.5µs37% SOL9.91×Reported · license unknown · no source2026-07-24
Measured exactly what you asked. Reported by source; not independently reproduced.