Claude 95d872e8f5 feat(muh): scan_by_key 53→284 lines — full key_size×val_size type dispatch ...

From 1KB/53 lines (59× compression vs CCCL 85KB) to 284 lines:
- Add 16 type specialization structs (key_size=1,2,4,8 × val_size=1,2,4,8)
- SM=16 tile maximization: k4_v4 (attention hot path) 30720→49152 (62%→100% SMEM)
- SM=16 tile increases across all small pairs (k1_v1: 3072→12288, k2_v1: 6144→15360)
- Delay halved for L2=6MB: fixed_delay values /2 (less inter-CTA contention)
- Proper CCCL-matching ScanByKeyPolicy struct with ScanByKeyAlgorithm enum
- SMEM-safe fallback with dynamic items computation from pair_size
- Macro-based dispatch (MK_POLICY) for clean type selection

2026-08-03 07:18:23 +00:00

..

common.cuh

fix: critical config + tuning corrections from CCCL source analysis

2026-08-03 06:45:54 +00:00

tuning_adjacent_difference.cuh

[muh] add tuning_adjacent_difference.cuh: BI-V100 tuning header for adjacent_difference

2026-08-01 02:08:28 +08:00

tuning_batch_memcpy.cuh

[MUH] Fix 7 structural discrepancies vs CCCL — read source, not grep

2026-07-30 14:37:38 +00:00

tuning_batched_topk.cuh

fix(tuning_batched_topk): force bits=8, fix SMEM overflow

2026-08-01 12:36:57 +08:00

tuning_find_bound_sorted_values.cuh

[muh] add tuning_find_bound_sorted_values.cuh: BI-V100 tuning header for find_bound_sorted_values

2026-08-01 02:08:31 +08:00

tuning_find.cuh

[muh] add tuning_find.cuh: BI-V100 tuning header for find

2026-08-01 02:08:29 +08:00

tuning_for.cuh

[MUH] Add C++/CUDA tuning headers — the real muh, not Python wrappers

2026-07-30 14:01:07 +00:00

tuning_histogram.cuh

[muh] add tuning_histogram.cuh: BI-V100 tuning for histogram

2026-08-01 02:10:48 +08:00

tuning_merge_sort.cuh

[muh] add tuning_merge_sort.cuh: BI-V100 tuning header for merge_sort

2026-08-01 02:08:38 +08:00

tuning_merge.cuh

[muh] add tuning_merge.cuh: BI-V100 tuning header for merge

2026-08-01 02:08:37 +08:00

tuning_radix_sort.cuh

fix(tuning_radix_sort): remove invented portioned_smem_per_warp field

2026-08-01 02:26:46 +08:00

tuning_reduce_by_key.cuh

[muh] add tuning_reduce_by_key.cuh: BI-V100 tuning for reduce_by_key

2026-08-01 02:10:57 +08:00

tuning_reduce.cuh

feat(muh): SM=16 tuning overhaul — reduce/scan/transform

2026-08-03 07:16:35 +00:00

tuning_rle_encode.cuh

[muh] add tuning_rle_encode.cuh: BI-V100 tuning for rle_encode

2026-08-01 02:10:50 +08:00

tuning_rle_non_trivial_runs.cuh

[muh] add tuning_rle_non_trivial_runs.cuh: BI-V100 tuning for rle_non_trivial_runs

2026-08-01 02:10:52 +08:00

tuning_scan_by_key.cuh

feat(muh): scan_by_key 53→284 lines — full key_size×val_size type dispatch

2026-08-03 07:18:23 +00:00

tuning_scan.cuh

feat(muh): SM=16 tuning overhaul — reduce/scan/transform

2026-08-03 07:16:35 +00:00

tuning_segmented_radix_sort.cuh

[muh] add tuning_segmented_radix_sort.cuh: BI-V100 tuning header for segmented_radix_sort

2026-08-01 02:08:44 +08:00

tuning_segmented_reduce.cuh

[muh] add tuning_segmented_reduce.cuh: BI-V100 tuning header for segmented_reduce

2026-08-01 02:08:33 +08:00

tuning_segmented_scan.cuh

[muh] add tuning_segmented_scan.cuh: BI-V100 tuning header for segmented_scan

2026-08-01 02:08:35 +08:00

tuning_segmented_sort.cuh

[muh] add tuning_segmented_sort.cuh: BI-V100 tuning for segmented_sort

2026-08-01 02:10:55 +08:00

tuning_select_if.cuh

fix(tuning_select_if): restore 3 collapsed dispatch dimensions

2026-08-01 02:26:48 +08:00

tuning_three_way_partition.cuh

[muh] add tuning_three_way_partition.cuh: BI-V100 tuning for three_way_partition

2026-08-01 02:10:54 +08:00

tuning_topk.cuh

[MUH] Fix 7 structural discrepancies vs CCCL — read source, not grep

2026-07-30 14:37:38 +00:00

tuning_transform_tile.cuh

[muh] add tuning_transform_tile.cuh: BI-V100 tuning header for transform_tile

2026-08-01 02:08:40 +08:00

tuning_transform.cuh

feat(muh): SM=16 tuning overhaul — reduce/scan/transform

2026-08-03 07:16:35 +00:00

tuning_unique_by_key.cuh

[muh] add tuning_unique_by_key.cuh: BI-V100 tuning for unique_by_key

2026-08-01 02:11:03 +08:00