code wiki / (root) / nx_flash_attention_test.nx

nx_flash_attention_test.nx

buildroot/runtime/nx_flash_attention_test.nx

5628 B121 linesdepth 6pulls 8 transitivereach 0 importersview sourcekind gate/prooftopic flash
docsdependenciesstructsconstsfunctions

about

nx_flash_attention_test.nx -- tiled-attention equivalence + memory win.

dependencies 6 imports · 0 importers

nx_syscalls.nx nx_tier.nx nx_tensor.nx nx_attention.nx nx_flash_attention.nx nx_numeric_oracle.nx nx_flash_attention_test.nx

imports: nx_syscalls.nxnx_tier.nxnx_tensor.nxnx_attention.nxnx_flash_attention.nxnx_numeric_oracle.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main sys_mmap nx_t_alloc nx_dt_is_valid nx_dt_element_bytes sys_mmap ↻ nx_t_compute_strides_rowma nx_t_fill_zero nx_attn_forward sys_mmap ↻ nx_t_alloc ↻ nx_attn_score_matrix nx_t_is_contiguous nx_attn_scale_q10 nx_t_is_contiguous ↻ nx_attn_softmax_row_q10 nx_t_is_contiguous ↻ sys_mmap ↻ _attn_exp_lut_fill _attn_exp_q10_lut nx_attn_apply_to_v nx_t_is_contiguous ↻ nx_fa_forward nx_t_is_contiguous ↻ _fa_row sys_mmap ↻ _attn_exp_lut_fill ↻ _attn_exp_q10_lut ↻ nx_no_check_epsilon_rel_q1 nx_no_witness_clear _no_structural_check nx_t_is_contiguous ↻ nx_fa_working_memory_bytes nx_fa_savings_ratio_q10 sys_mmap ↻ nx_fa_working_memory_bytes ↻ nx_fa_verdict_is_valid

structs

none

consts

none

functions

10func main() -> nx_int
calls 9: sys_mmapnx_t_allocnx_t_fill_zeronx_attn_forwardnx_fa_forwardnx_no_check_epsilon_rel_q10+3