dylan e2fc3f270f fix: corex_batched_gemm use TCU OpClassTensorOp + Cu10 + float accum ...

Previous version used default SIMT path (25ms).
Fixed version matches moe_cutlass_batched.cu (2.462ms):
  - ElementAccumulator = float (was half_t)
  - OpClassTensorOp (was default OpClassSimt)
  - arch::Cu10 (was default Sm61)
  - RowMajor layout (was ColumnMajor)
  - torch::sigmoid(x)*x instead of torch::silu (not in corex torch)

Also fixed bind.cpp: removed col-major transposition logic,
kernel now RowMajor so A(M,K) @ B(K,N) = C(M,N) directly.

2026-08-15 12:35:56 +00:00

..

cuda

fix: corex_batched_gemm use TCU OpClassTensorOp + Cu10 + float accum

2026-08-15 12:35:56 +00:00

ilu

data: complete SGEMM upstream from 3 repos (siboehm+wangzyon+edtallison) + xllm fused_qknorm_rope + xattention kernels

2026-08-15 07:00:09 +00:00

npu

fix(critical): fold max_completion_tokens + max_num_seqs=2 + max_model_len=80000 + xllm_latest layer import

2026-08-13 03:19:39 +00:00

build_test_cutlass_batched.sh

feat: Cu10 TensorOp batched HGEMM via Iluvatar CUTLASS framework

2026-08-15 05:44:15 +00:00

build_test_hgemm_warp.sh

feat: hgemm_warptiling.cu — siboehm kernel 10 ported to WARPSIZE=64 FP16

2026-08-14 17:05:59 +00:00

build_test_hgemm.sh

feat: hgemm_blocktiling.cu — FP16 GEMM kernel for MoE expert dispatch on BI-V100

2026-08-14 16:22:00 +00:00

rebuild_test_k10.sh

test: clean rebuild + debug output for kernel 10 correctness

2026-08-15 05:14:18 +00:00