Star708
1011
代码
代码
Issues74
Pull Requests194
Actions
项目讨论
Wiki
分析
Star708
1011
已合并
统一Capi fp4/fp8类型的命名风格 #2205
pangfd创建于 5月15日
统一Capi fp4/fp8类型的命名风格 #2205
已合并
pangfd创建于 5月15日
从已删除 :cherry-pick-mr-2123-1778815757753-auto合入到cann/asc-devkitmaster
共 30 个文件变更+703-651
Mdocs/api/c_api/cube_compute/asc_mmad.md+8-8
@@ -37,10 +37,10 @@ $$
37 ```cpp37 ```cpp
38 // 如下原型仅支持Ascend 950PR/Ascend 950DT38 // 如下原型仅支持Ascend 950PR/Ascend 950DT
39 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ bfloat16_t* a_matrix, __cb__ bfloat16_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);39 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ bfloat16_t* a_matrix, __cb__ bfloat16_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
40- __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);40+ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
41- __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);41+ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
42- __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);42+ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
43- __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);43+ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
44 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ half* a_matrix, __cb__ half* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);44 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ half* a_matrix, __cb__ half* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
45 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float* a_matrix, __cb__ float* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);45 __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float* a_matrix, __cb__ float* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
46 __aicore__ inline void asc_mmad(__cc__ int32_t* c_matrix, __ca__ int8_t* a_matrix, __cb__ int8_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);46 __aicore__ inline void asc_mmad(__cc__ int32_t* c_matrix, __ca__ int8_t* a_matrix, __cb__ int8_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
@@ -67,10 +67,10 @@ $$
67 ```cpp67 ```cpp
68 // 如下原型仅支持Ascend 950PR/Ascend 950DT68 // 如下原型仅支持Ascend 950PR/Ascend 950DT
69 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ bfloat16_t* a_matrix, __cb__ bfloat16_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);69 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ bfloat16_t* a_matrix, __cb__ bfloat16_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
70- __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);70+ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
71- __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);71+ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
72- __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);72+ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
73- __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);73+ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
74 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ half* a_matrix, __cb__ half* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);74 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ half* a_matrix, __cb__ half* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
75 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float* a_matrix, __cb__ float* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);75 __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float* a_matrix, __cb__ float* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
76 __aicore__ inline void asc_mmad_sync(__cc__ int32_t* c_matrix, __ca__ int8_t* a_matrix, __cb__ int8_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);76 __aicore__ inline void asc_mmad_sync(__cc__ int32_t* c_matrix, __ca__ int8_t* a_matrix, __cb__ int8_t* b_matrix, uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
Mdocs/api/c_api/cube_datamove/asc_copy_gm2l1/asc_copy_gm2l1_arch_3510.md+8-8
@@ -31,8 +31,8 @@
31 ```cpp31 ```cpp
32 __aicore__ inline void asc_copy_gm2l1(__cbuf__ bfloat16_t *dst, __gm__ bfloat16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)32 __aicore__ inline void asc_copy_gm2l1(__cbuf__ bfloat16_t *dst, __gm__ bfloat16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
33 __aicore__ inline void asc_copy_gm2l1(__cbuf__ float *dst, __gm__ float *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)33 __aicore__ inline void asc_copy_gm2l1(__cbuf__ float *dst, __gm__ float *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
34- __aicore__ inline void asc_copy_gm2l1(__cbuf__ float8_e4m3_t *dst, __gm__ float8_e4m3_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)34+ __aicore__ inline void asc_copy_gm2l1(__cbuf__ fp8_e4m3fn_t *dst, __gm__ fp8_e4m3fn_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
35- __aicore__ inline void asc_copy_gm2l1(__cbuf__ float8_e5m2_t *dst, __gm__ float8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)35+ __aicore__ inline void asc_copy_gm2l1(__cbuf__ fp8_e5m2_t *dst, __gm__ fp8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
36 __aicore__ inline void asc_copy_gm2l1(__cbuf__ half *dst, __gm__ half *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)36 __aicore__ inline void asc_copy_gm2l1(__cbuf__ half *dst, __gm__ half *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
37 __aicore__ inline void asc_copy_gm2l1(__cbuf__ hifloat8_t *dst, __gm__ hifloat8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)37 __aicore__ inline void asc_copy_gm2l1(__cbuf__ hifloat8_t *dst, __gm__ hifloat8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
38 __aicore__ inline void asc_copy_gm2l1(__cbuf__ int16_t *dst, __gm__ int16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)38 __aicore__ inline void asc_copy_gm2l1(__cbuf__ int16_t *dst, __gm__ int16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
@@ -41,8 +41,8 @@
41 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint16_t *dst, __gm__ uint16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)41 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint16_t *dst, __gm__ uint16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
42 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint32_t *dst, __gm__ uint32_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)42 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint32_t *dst, __gm__ uint32_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
43 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint8_t *dst, __gm__ uint8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)43 __aicore__ inline void asc_copy_gm2l1(__cbuf__ uint8_t *dst, __gm__ uint8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
44- __aicore__ inline void asc_copy_gm2l1(__cbuf__ float4_e1m2x2_t *dst, __gm__ float4_e1m2x2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)44+ __aicore__ inline void asc_copy_gm2l1(__cbuf__ fp4x2_e1m2_t *dst, __gm__ fp4x2_e1m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
45- __aicore__ inline void asc_copy_gm2l1(__cbuf__ float4_e2m1x2_t *dst, __gm__ float4_e2m1x2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)45+ __aicore__ inline void asc_copy_gm2l1(__cbuf__ fp4x2_e2m1_t *dst, __gm__ fp4x2_e2m1_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
46 __aicore__ inline void asc_copy_gm2l1(__cbuf__ void *dst, __gm__ void *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)46 __aicore__ inline void asc_copy_gm2l1(__cbuf__ void *dst, __gm__ void *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
47 ```47 ```
48 48 
@@ -51,8 +51,8 @@
51 ```cpp51 ```cpp
52 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ bfloat16_t *dst, __gm__ bfloat16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)52 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ bfloat16_t *dst, __gm__ bfloat16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
53 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float *dst, __gm__ float *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)53 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float *dst, __gm__ float *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
54- __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float8_e4m3_t *dst, __gm__ float8_e4m3_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)54+ __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ fp8_e4m3fn_t *dst, __gm__ fp8_e4m3fn_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
55- __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float8_e5m2_t *dst, __gm__ float8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)55+ __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ fp8_e5m2_t *dst, __gm__ fp8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
56 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ half *dst, __gm__ half *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)56 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ half *dst, __gm__ half *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
57 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ hifloat8_t *dst, __gm__ hifloat8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)57 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ hifloat8_t *dst, __gm__ hifloat8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
58 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ int16_t *dst, __gm__ int16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)58 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ int16_t *dst, __gm__ int16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
@@ -61,8 +61,8 @@
61 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint16_t *dst, __gm__ uint16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)61 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint16_t *dst, __gm__ uint16_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
62 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint32_t *dst, __gm__ uint32_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)62 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint32_t *dst, __gm__ uint32_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
63 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint8_t *dst, __gm__ uint8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)63 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ uint8_t *dst, __gm__ uint8_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
64- __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float4_e1m2x2_t *dst, __gm__ float4_e1m2x2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)64+ __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ fp4x2_e1m2_t *dst, __gm__ fp4x2_e1m2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
65- __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ float4_e2m1x2_t *dst, __gm__ float4_e2m1x2_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)65+ __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ fp4x2_e2m1_t *dst, __gm__ fp4x2_e2m1_t *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
66 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ void *dst, __gm__ void *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)66 __aicore__ inline void asc_copy_gm2l1_sync(__cbuf__ void *dst, __gm__ void *src, uint32_t m_start_position, uint32_t k_start_position, uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
67 ```67 ```
68 68 
Mdocs/api/c_api/cube_datamove/asc_copy_gm2l1_dn2nz.md+4-4
@@ -22,8 +22,8 @@ DN->NZ的搬运形式如下图：
22 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)22 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
23 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)23 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
24 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)24 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
25- __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)25+ __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
26- __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)26+ __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
27 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)27 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
28 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)28 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
29 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)29 __aicore__ inline void asc_copy_gm2l1_dn2nz(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
@@ -39,8 +39,8 @@ DN->NZ的搬运形式如下图：
39 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)39 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
40 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)40 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
41 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)41 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
42- __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)42+ __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
43- __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)43+ __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
44 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)44 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
45 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)45 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
46 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)46 __aicore__ inline void asc_copy_gm2l1_dn2nz_sync(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
Mdocs/api/c_api/cube_datamove/asc_copy_gm2l1_nd2nz/asc_copy_gm2l1_nd2nz_arch_3510.md+4-4
@@ -22,8 +22,8 @@ ND->NZ的搬运形式如下图：
22 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)22 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
23 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)23 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
24 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)24 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
25- __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)25+ __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
26- __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)26+ __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
27 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)27 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
28 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)28 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
29 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)29 __aicore__ inline void asc_copy_gm2l1_nd2nz(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
@@ -39,8 +39,8 @@ ND->NZ的搬运形式如下图：
39 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)39 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ int8_t* dst, __gm__ int8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
40 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)40 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ uint8_t* dst, __gm__ uint8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
41 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)41 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ hifloat8_t* dst, __gm__ hifloat8_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
42- __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)42+ __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
43- __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)43+ __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
44 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)44 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ int16_t* dst, __gm__ int16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
45 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)45 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ uint16_t* dst, __gm__ uint16_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
46 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)46 __aicore__ inline void asc_copy_gm2l1_nd2nz_sync(__cbuf__ half* dst, __gm__ half* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value, uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
Mdocs/api/c_api/cube_datamove/asc_copy_l0c2gm/asc_copy_l0c2gm_arch_3510.md+8-8
@@ -63,11 +63,11 @@ quant_post可选量化模式分别为：
63 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,63 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
64 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,64 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
65 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)65 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
66- __aicore__ inline void asc_copy_l0c2gm(__gm__ float8_e4m3_t *dst_addr, __cc__ float *src_addr,66+ __aicore__ inline void asc_copy_l0c2gm(__gm__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr,
67 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,67 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
68 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,68 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
69 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)69 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
70- __aicore__ inline void asc_copy_l0c2gm(__gm__ float8_e5m2_t *dst_addr, __cc__ float *src_addr,70+ __aicore__ inline void asc_copy_l0c2gm(__gm__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr,
71 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,71 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
72 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,72 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
73 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)73 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
@@ -95,11 +95,11 @@ quant_post可选量化模式分别为：
95 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,95 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
96 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,96 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
97 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)97 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
98- __aicore__ inline void asc_copy_l0c2gm(__gm__ float8_e4m3_t *dst_addr, __cc__ int32_t *src_addr,98+ __aicore__ inline void asc_copy_l0c2gm(__gm__ fp8_e4m3fn_t *dst_addr, __cc__ int32_t *src_addr,
99 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,99 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
100 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,100 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
101 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)101 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
102- __aicore__ inline void asc_copy_l0c2gm(__gm__ float8_e5m2_t *dst_addr, __cc__ int32_t *src_addr,102+ __aicore__ inline void asc_copy_l0c2gm(__gm__ fp8_e5m2_t *dst_addr, __cc__ int32_t *src_addr,
103 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,103 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
104 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,104 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
105 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)105 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
@@ -140,11 +140,11 @@ quant_post可选量化模式分别为：
140 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,140 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
141 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,141 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
142 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)142 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
143- __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ float8_e4m3_t* dst_addr, __cc__ float* src_addr,143+ __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ fp8_e4m3fn_t* dst_addr, __cc__ float* src_addr,
144 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,144 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
145 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,145 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
146 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)146 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
147- __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ float8_e5m2_t* dst_addr, __cc__ float* src_addr,147+ __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ fp8_e5m2_t* dst_addr, __cc__ float* src_addr,
148 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,148 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
149 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,149 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
150 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)150 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
@@ -172,11 +172,11 @@ quant_post可选量化模式分别为：
172 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,172 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
173 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,173 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
174 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)174 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
175- __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ float8_e4m3_t* dst_addr, __cc__ int32_t* src_addr,175+ __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ fp8_e4m3fn_t* dst_addr, __cc__ int32_t* src_addr,
176 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,176 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
177 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,177 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
178 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)178 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
179- __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ float8_e5m2_t* dst_addr, __cc__ int32_t* src_addr,179+ __aicore__ inline void asc_copy_l0c2gm_sync(__gm__ fp8_e5m2_t* dst_addr, __cc__ int32_t* src_addr,
180 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,180 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
181 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,181 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
182 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)182 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en, bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
Mdocs/api/c_api/cube_datamove/asc_copy_l0c2ub.md+8-8
@@ -66,12 +66,12 @@ quant_post可选量化模式分别为：
66 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,66 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
67 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,67 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
68 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)68 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
69- __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float8_e4m3_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,69+ __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,
70 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,70 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
71 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,71 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
72 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,72 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
73 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)73 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
74- __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,74+ __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,
75 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,75 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
76 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,76 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
77 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,77 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -106,12 +106,12 @@ quant_post可选量化模式分别为：
106 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,106 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
107 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,107 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
108 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)108 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
109- __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float8_e4m3_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,109+ __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,
110 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,110 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
111 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,111 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
112 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,112 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
113 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)113 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
114- __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float8_e5m2_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,114+ __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,
115 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,115 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
116 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,116 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
117 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,117 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -161,12 +161,12 @@ quant_post可选量化模式分别为：
161 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,161 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
162 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,162 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
163 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)163 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
164- __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float8_e4m3_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,164+ __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ fp8_e4m3fn_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,
165 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,165 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
166 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,166 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
167 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,167 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
168 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)168 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
169- __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float8_e5m2_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,169+ __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ fp8_e5m2_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,
170 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,170 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
171 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,171 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
172 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,172 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -201,12 +201,12 @@ quant_post可选量化模式分别为：
201 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,201 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
202 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,202 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
203 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)203 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
204- __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float8_e4m3_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,204+ __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ fp8_e4m3fn_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,
205 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,205 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
206 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,206 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
207 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,207 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
208 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)208 bool C0_pad_en, bool broadcast_en, bool NZ2DN_en)
209- __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float8_e5m2_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,209+ __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ fp8_e5m2_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,
210 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,210 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
211 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,211 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
212 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,212 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
Mdocs/api/c_api/cube_datamove/asc_copy_l12l0a/asc_copy_l12l0a_arch_3510.md+44-23
@@ -15,9 +15,10 @@
15- 常规搬运 2D格式15- 常规搬运 2D格式
16 16 
17 ```cpp17 ```cpp
18+ 
18 __aicore__ inline void asc_copy_l12l0a(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)19 __aicore__ inline void asc_copy_l12l0a(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
19- __aicore__ inline void asc_copy_l12l0a(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)20+ __aicore__ inline void asc_copy_l12l0a(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
20- __aicore__ inline void asc_copy_l12l0a(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)21+ __aicore__ inline void asc_copy_l12l0a(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
21 __aicore__ inline void asc_copy_l12l0a(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)22 __aicore__ inline void asc_copy_l12l0a(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
22 __aicore__ inline void asc_copy_l12l0a(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)23 __aicore__ inline void asc_copy_l12l0a(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
23 __aicore__ inline void asc_copy_l12l0a(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)24 __aicore__ inline void asc_copy_l12l0a(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -27,17 +28,19 @@
27 __aicore__ inline void asc_copy_l12l0a(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)28 __aicore__ inline void asc_copy_l12l0a(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
28 __aicore__ inline void asc_copy_l12l0a(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)29 __aicore__ inline void asc_copy_l12l0a(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
29 __aicore__ inline void asc_copy_l12l0a(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)30 __aicore__ inline void asc_copy_l12l0a(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
31+ __aicore__ inline void asc_copy_l12l0a(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
30 __aicore__ inline void asc_copy_l12l0a(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)32 __aicore__ inline void asc_copy_l12l0a(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
31 __aicore__ inline void asc_copy_l12l0a(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)33 __aicore__ inline void asc_copy_l12l0a(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
32- 34+ 
33 ```35 ```
34 36 
35- 同步常规搬运 2D格式37- 同步常规搬运 2D格式
36 38 
37 ```cpp39 ```cpp
40+ 
38 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)41 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
39- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)42+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
40- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)43+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
41 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)44 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
42 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)45 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
43 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)46 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -47,17 +50,19 @@
47 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)50 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
48 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)51 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
49 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)52 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
53+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
50 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)54 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
51 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)55 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
52- 56+ 
53 ```57 ```
54 58 
55- 转置搬运 2D格式59- 转置搬运 2D格式
56 60 
57 ```cpp61 ```cpp
62+ 
58 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)63 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
59- __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)64+ __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
60- __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)65+ __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
61 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)66 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
62 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)67 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
63 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)68 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -67,17 +72,19 @@
67 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)72 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
68 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)73 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
69 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)74 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
75+ __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
70 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)76 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
71 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)77 __aicore__ inline void asc_copy_l12l0a_transpose(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
72- 78+ 
73 ```79 ```
74- 80+ 
75- 同步转置搬运 2D格式81- 同步转置搬运 2D格式
76 82 
77 ```cpp83 ```cpp
84+ 
78 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)85 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
79- __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)86+ __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
80- __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)87+ __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
81 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)88 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
82 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)89 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
83 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)90 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -87,35 +94,49 @@
87 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)94 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
88 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)95 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
89 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)96 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
97+ __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
90 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)98 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
91 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)99 __aicore__ inline void asc_copy_l12l0a_transpose_sync(__ca__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
92 100
93 ```101 ```
102+ 
94- 高维切分搬运 3D格式103- 高维切分搬运 3D格式
95 104 
96 ```cpp105 ```cpp
97- __aicore__ inline void asc_copy_l12l0a(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)106+ 
98- __aicore__ inline void asc_copy_l12l0a(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
99 __aicore__ inline void asc_copy_l12l0a(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)107 __aicore__ inline void asc_copy_l12l0a(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
100- __aicore__ inline void asc_copy_l12l0a(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)108+ __aicore__ inline void asc_copy_l12l0a(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
109+ __aicore__ inline void asc_copy_l12l0a(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
110+ __aicore__ inline void asc_copy_l12l0a(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
111+ __aicore__ inline void asc_copy_l12l0a(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
112+ __aicore__ inline void asc_copy_l12l0a(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
113+ __aicore__ inline void asc_copy_l12l0a(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
101 __aicore__ inline void asc_copy_l12l0a(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)114 __aicore__ inline void asc_copy_l12l0a(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
102- __aicore__ inline void asc_copy_l12l0a(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)115+ __aicore__ inline void asc_copy_l12l0a(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
103 __aicore__ inline void asc_copy_l12l0a(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)116 __aicore__ inline void asc_copy_l12l0a(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
117+ __aicore__ inline void asc_copy_l12l0a(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
104 __aicore__ inline void asc_copy_l12l0a(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)118 __aicore__ inline void asc_copy_l12l0a(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
119+ 
105 ```120 ```
106 121 
107- 同步高维切分搬运 3D格式122- 同步高维切分搬运 3D格式
108 123 
109- ```cpp124+```cpp
110- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)125+ 
111- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
112 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)126 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
113- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)127+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
128+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
129+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
130+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
131+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
132+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
114 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)133 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
115- __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)134+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
116 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)135 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
136+ __aicore__ inline void asc_copy_l12l0a_sync(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
117 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)137 __aicore__ inline void asc_copy_l12l0a_sync(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
118- ```138+ 
139+```
119 140 
120## 参数说明141## 参数说明
121 142 
@@ -211,4 +232,4 @@ uint8_t k_step = 4;
211int16_t src_stride = 1;232int16_t src_stride = 1;
212uint16_t dst_stride = 1;233uint16_t dst_stride = 1;
213asc_copy_l12l0a(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);234asc_copy_l12l0a(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
214-```235+```
Mdocs/api/c_api/cube_datamove/asc_copy_l12l0b/asc_copy_l12l0b_arch_3510.md+46-16
@@ -13,10 +13,12 @@
13## 函数原型13## 函数原型
14 14 
15- 常规搬运 2D格式15- 常规搬运 2D格式
16+ 
16 ```cpp17 ```cpp
18+ 
17 __aicore__ inline void asc_copy_l12l0b(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)19 __aicore__ inline void asc_copy_l12l0b(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
18- __aicore__ inline void asc_copy_l12l0b(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)20+ __aicore__ inline void asc_copy_l12l0b(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
19- __aicore__ inline void asc_copy_l12l0b(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)21+ __aicore__ inline void asc_copy_l12l0b(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
20 __aicore__ inline void asc_copy_l12l0b(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)22 __aicore__ inline void asc_copy_l12l0b(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
21 __aicore__ inline void asc_copy_l12l0b(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)23 __aicore__ inline void asc_copy_l12l0b(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
22 __aicore__ inline void asc_copy_l12l0b(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)24 __aicore__ inline void asc_copy_l12l0b(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -26,16 +28,19 @@
26 __aicore__ inline void asc_copy_l12l0b(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)28 __aicore__ inline void asc_copy_l12l0b(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
27 __aicore__ inline void asc_copy_l12l0b(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)29 __aicore__ inline void asc_copy_l12l0b(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
28 __aicore__ inline void asc_copy_l12l0b(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)30 __aicore__ inline void asc_copy_l12l0b(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
31+ __aicore__ inline void asc_copy_l12l0b(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
29 __aicore__ inline void asc_copy_l12l0b(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)32 __aicore__ inline void asc_copy_l12l0b(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
30 __aicore__ inline void asc_copy_l12l0b(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)33 __aicore__ inline void asc_copy_l12l0b(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
31- 34+ 
32 ```35 ```
33 36 
34- 同步常规搬运 2D格式37- 同步常规搬运 2D格式
38+ 
35 ```cpp39 ```cpp
40+ 
36 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)41 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
37- __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)42+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
38- __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)43+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
39 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)44 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
40 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)45 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
41 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)46 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -45,16 +50,19 @@
45 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)50 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
46 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)51 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
47 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)52 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
53+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
48 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)54 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
49 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)55 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
50- 56+ 
51 ```57 ```
52 58 
53- 转置搬运 2D格式59- 转置搬运 2D格式
60+ 
54 ```cpp61 ```cpp
62+ 
55 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)63 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
56- __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)64+ __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
57- __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)65+ __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
58 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)66 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
59 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)67 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
60 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)68 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -64,16 +72,19 @@
64 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)72 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
65 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)73 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
66 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)74 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
75+ __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
67 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)76 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
68 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)77 __aicore__ inline void asc_copy_l12l0b_transpose(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
69- 78+ 
70 ```79 ```
71- 80+ 
72- 同步转置搬运 2D格式81- 同步转置搬运 2D格式
82+ 
73 ```cpp83 ```cpp
84+ 
74 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)85 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
75- __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)86+ __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
76- __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)87+ __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
77 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)88 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
78 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)89 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
79 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)90 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
@@ -83,29 +94,48 @@
83 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)94 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
84 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)95 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
85 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)96 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
97+ __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
86 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)98 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
87 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)99 __aicore__ inline void asc_copy_l12l0b_transpose_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t m_start_position, uint16_t k_start_position, uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
88- 100+ 
89 ```101 ```
90 102 
91- 高维切分搬运 3D格式103- 高维切分搬运 3D格式
92 104 
93 ```cpp105 ```cpp
106+ 
107+ __aicore__ inline void asc_copy_l12l0b(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
108+ __aicore__ inline void asc_copy_l12l0b(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
109+ __aicore__ inline void asc_copy_l12l0b(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
110+ __aicore__ inline void asc_copy_l12l0b(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
111+ __aicore__ inline void asc_copy_l12l0b(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
112+ __aicore__ inline void asc_copy_l12l0b(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
113+ __aicore__ inline void asc_copy_l12l0b(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
94 __aicore__ inline void asc_copy_l12l0b(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)114 __aicore__ inline void asc_copy_l12l0b(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
95 __aicore__ inline void asc_copy_l12l0b(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)115 __aicore__ inline void asc_copy_l12l0b(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
96- __aicore__ inline void asc_copy_l12l0b(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
97 __aicore__ inline void asc_copy_l12l0b(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)116 __aicore__ inline void asc_copy_l12l0b(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
117+ __aicore__ inline void asc_copy_l12l0b(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
98 __aicore__ inline void asc_copy_l12l0b(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)118 __aicore__ inline void asc_copy_l12l0b(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
119+ 
99 ```120 ```
100 121 
101- 同步高维切分搬运 3D格式122- 同步高维切分搬运 3D格式
102 123 
103 ```cpp124 ```cpp
125+ 
126+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
127+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
128+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
129+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
130+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
131+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
132+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
104 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)133 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
105 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)134 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
106- __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
107 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)135 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
136+ __aicore__ inline void asc_copy_l12l0b_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
108 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)137 __aicore__ inline void asc_copy_l12l0b_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt, uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w, uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
138+ 
109 ```139 ```
110 140 
111## 参数说明141## 参数说明
@@ -202,4 +232,4 @@ uint8_t k_step = 4;
202int16_t src_stride = 1;232int16_t src_stride = 1;
203uint16_t dst_stride = 1;233uint16_t dst_stride = 1;
204asc_copy_l12l0b(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);234asc_copy_l12l0b(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
205-```235+```
Mdocs/api/c_api/cube_datamove/asc_copy_l12l0b_trans/asc_copy_l12l0b_trans_arch_3510.md+7-6
@@ -78,8 +78,8 @@
78 78 
79 ```cpp79 ```cpp
80 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);80 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
81- __aicore__ inline void asc_copy_l12l0b_trans(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);81+ __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
82- __aicore__ inline void asc_copy_l12l0b_trans(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);82+ __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
83 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ half* dst, __cbuf__ half* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);83 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ half* dst, __cbuf__ half* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
84 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ float* dst, __cbuf__ float* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);84 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ float* dst, __cbuf__ float* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
85 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);85 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
@@ -87,17 +87,17 @@
87 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);87 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
88 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);88 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
89 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);89 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
90+ __aicore__ inline void asc_copy_l12l0b_trans(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
90 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);91 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
91 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);92 __aicore__ inline void asc_copy_l12l0b_trans(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
92- 
93 ```93 ```
94 94 
95- 同步转置搬运95- 同步转置搬运
96 96 
97 ```cpp97 ```cpp
98 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);98 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
99- __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);99+ __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
100- __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);100+ __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
101 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);101 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ half* dst, __cbuf__ half* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
102 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);102 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ float* dst, __cbuf__ float* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
103 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);103 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
@@ -105,6 +105,7 @@
105 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);105 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
106 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);106 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
107 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);107 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
108+ __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ int4b_t* dst, __cbuf__ int4b_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
108 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);109 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
109 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);110 __aicore__ inline void asc_copy_l12l0b_trans_sync(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride, uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride);
110 111 
@@ -156,4 +157,4 @@ uint16_t dst_frac_stride = 15;
156uint16_t src_frac_stride = 8;157uint16_t src_frac_stride = 8;
157 158 
158asc_copy_l12l0b_trans(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);159asc_copy_l12l0b_trans(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);
159-```160+```
Mdocs/api/c_api/reg/reg_load/asc_scatter.md+2-2
@@ -19,7 +19,7 @@ __simd_callee__ inline void asc_scatter(vector_int16_t& dst, __ubuf__ int16_t* s
19__simd_callee__ inline void asc_scatter(vector_uint16_t& dst, __ubuf__ uint16_t* src, vector_uint16_t index, vector_bool mask)19__simd_callee__ inline void asc_scatter(vector_uint16_t& dst, __ubuf__ uint16_t* src, vector_uint16_t index, vector_bool mask)
20__simd_callee__ inline void asc_scatter(vector_int32_t& dst, __ubuf__ int32_t* src, vector_uint32_t index, vector_bool mask)20__simd_callee__ inline void asc_scatter(vector_int32_t& dst, __ubuf__ int32_t* src, vector_uint32_t index, vector_bool mask)
21__simd_callee__ inline void asc_scatter(vector_uint32_t& dst, __ubuf__ uint32_t* src, vector_uint32_t index, vector_bool mask)21__simd_callee__ inline void asc_scatter(vector_uint32_t& dst, __ubuf__ uint32_t* src, vector_uint32_t index, vector_bool mask)
22-__simd_callee__ inline void asc_scatter(vector_bf16& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)22+__simd_callee__ inline void asc_scatter(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)
23__simd_callee__ inline void asc_scatter(vector_half& dst, __ubuf__ half* src, vector_uint16_t index, vector_bool mask)23__simd_callee__ inline void asc_scatter(vector_half& dst, __ubuf__ half* src, vector_uint16_t index, vector_bool mask)
24__simd_callee__ inline void asc_scatter(vector_float& dst, __ubuf__ float* src, vector_uint32_t index, vector_bool mask)24__simd_callee__ inline void asc_scatter(vector_float& dst, __ubuf__ float* src, vector_uint32_t index, vector_bool mask)
25```25```
@@ -56,4 +56,4 @@ vector_uint16_t index;
56vector_bool mask = asc_create_mask_b16(PAT_ALL);56vector_bool mask = asc_create_mask_b16(PAT_ALL);
57asc_loadalign(src, src_addr); // src_addr是外部输入的UB内存空间地址。57asc_loadalign(src, src_addr); // src_addr是外部输入的UB内存空间地址。
58asc_scatter(dst, src, index, mask);58asc_scatter(dst, src, index, mask);
59-```59+```
Mimpl/c_api/instr_impl/npu_arch_3510/cube_compute_impl.h+13-13
@@ -172,7 +172,7 @@ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ bfloat16_t*
172 c_matrix_source, c_matrix_init_val);172 c_matrix_source, c_matrix_init_val);
173}173}
174 174 
175-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e4m3_t* b_matrix,175+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix,
176 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,176 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
177 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)177 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)
178{178{
@@ -180,8 +180,8 @@ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_
180 c_matrix_source, c_matrix_init_val);180 c_matrix_source, c_matrix_init_val);
181}181}
182 182 
183-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,183+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
184- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,184+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
185 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,185 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
186 bool c_matrix_init_val)186 bool c_matrix_init_val)
187{187{
@@ -189,7 +189,7 @@ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_
189 c_matrix_source, c_matrix_init_val);189 c_matrix_source, c_matrix_init_val);
190}190}
191 191 
192-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e5m2_t* b_matrix,192+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix,
193 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,193 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
194 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)194 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)
195{195{
@@ -197,8 +197,8 @@ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_
197 c_matrix_source, c_matrix_init_val);197 c_matrix_source, c_matrix_init_val);
198}198}
199 199 
200-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,200+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
201- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,201+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
202 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,202 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
203 bool c_matrix_init_val)203 bool c_matrix_init_val)
204{204{
@@ -206,7 +206,7 @@ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_
206 c_matrix_source, c_matrix_init_val);206 c_matrix_source, c_matrix_init_val);
207}207}
208 208 
209-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e4m3_t* b_matrix,209+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix,
210 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,210 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
211 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)211 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)
212{212{
@@ -214,8 +214,8 @@ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_
214 c_matrix_source, c_matrix_init_val);214 c_matrix_source, c_matrix_init_val);
215}215}
216 216 
217-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,217+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
218- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,218+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
219 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,219 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
220 bool c_matrix_init_val)220 bool c_matrix_init_val)
221{221{
@@ -223,7 +223,7 @@ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_
223 c_matrix_source, c_matrix_init_val);223 c_matrix_source, c_matrix_init_val);
224}224}
225 225 
226-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e5m2_t* b_matrix,226+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix,
227 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,227 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
228 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)228 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val)
229{229{
@@ -231,8 +231,8 @@ __aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_
231 c_matrix_source, c_matrix_init_val);231 c_matrix_source, c_matrix_init_val);
232}232}
233 233 
234-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,234+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
235- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,235+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
236 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,236 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
237 bool c_matrix_init_val)237 bool c_matrix_init_val)
238{238{
@@ -288,4 +288,4 @@ __aicore__ inline void asc_mmad_sync(__cc__ int32_t* c_matrix, __ca__ int8_t* a_
288 c_matrix_source, c_matrix_init_val);288 c_matrix_source, c_matrix_init_val);
289}289}
290 290 
291-#endif291+#endif
Mimpl/c_api/instr_impl/npu_arch_3510/cube_compute_impl/asc_mmad_impl.h+16-16
@@ -44,8 +44,8 @@ __aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ bfloat1
44 asc_sync_post_process();44 asc_sync_post_process();
45}45}
46 46 
47-__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,47+__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
48- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,48+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
49 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,49 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
50 bool c_matrix_init_val)50 bool c_matrix_init_val)
51{51{
@@ -55,8 +55,8 @@ __aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_
55 }55 }
56}56}
57 57 
58-__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,58+__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
59- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,59+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
60 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,60 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,
61 bool c_matrix_source, bool c_matrix_init_val)61 bool c_matrix_source, bool c_matrix_init_val)
62{62{
@@ -65,8 +65,8 @@ __aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_
65 asc_sync_post_process();65 asc_sync_post_process();
66}66}
67 67 
68-__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,68+__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
69- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,69+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
70 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,70 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
71 bool c_matrix_init_val)71 bool c_matrix_init_val)
72{72{
@@ -76,8 +76,8 @@ __aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_
76 }76 }
77}77}
78 78 
79-__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,79+__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
80- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,80+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
81 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,81 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,
82 bool c_matrix_source, bool c_matrix_init_val)82 bool c_matrix_source, bool c_matrix_init_val)
83{83{
@@ -86,8 +86,8 @@ __aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_
86 asc_sync_post_process();86 asc_sync_post_process();
87}87}
88 88 
89-__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,89+__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
90- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,90+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
91 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,91 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
92 bool c_matrix_init_val)92 bool c_matrix_init_val)
93{93{
@@ -97,8 +97,8 @@ __aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_
97 }97 }
98}98}
99 99 
100-__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,100+__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
101- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,101+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
102 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,102 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,
103 bool c_matrix_source, bool c_matrix_init_val)103 bool c_matrix_source, bool c_matrix_init_val)
104{104{
@@ -107,8 +107,8 @@ __aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_
107 asc_sync_post_process();107 asc_sync_post_process();
108}108}
109 109 
110-__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,110+__aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
111- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,111+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
112 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,112 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
113 bool c_matrix_init_val)113 bool c_matrix_init_val)
114{114{
@@ -118,8 +118,8 @@ __aicore__ inline void asc_mmad_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_
118 }118 }
119}119}
120 120 
121-__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,121+__aicore__ inline void asc_mmad_sync_impl(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
122- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,122+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
123 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,123 uint16_t right_width, uint8_t unit_flag, bool disable_gemv,
124 bool c_matrix_source, bool c_matrix_init_val)124 bool c_matrix_source, bool c_matrix_init_val)
125{125{
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_gm2l1_dn2nz_impl.h+4-4
@@ -34,7 +34,7 @@ __aicore__ inline void asc_copy_gm2l1_dn2nz_sync_impl(__cbuf__ bfloat16_t* dst,
34 asc_sync_post_process();34 asc_sync_post_process();
35}35}
36 36 
37-__aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,37+__aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
38 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)38 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
39{39{
40 if ASC_IS_AIC {40 if ASC_IS_AIC {
@@ -42,14 +42,14 @@ __aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ float8_e4m3_t* dst, __
42 }42 }
43}43}
44 44 
45-__aicore__ inline void asc_copy_gm2l1_dn2nz_sync_impl(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,45+__aicore__ inline void asc_copy_gm2l1_dn2nz_sync_impl(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
46 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)46 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
47{47{
48 asc_copy_gm2l1_dn2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);48 asc_copy_gm2l1_dn2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);
49 asc_sync_post_process();49 asc_sync_post_process();
50}50}
51 51 
52-__aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,52+__aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
53 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)53 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
54{54{
55 if ASC_IS_AIC {55 if ASC_IS_AIC {
@@ -57,7 +57,7 @@ __aicore__ inline void asc_copy_gm2l1_dn2nz_impl(__cbuf__ float8_e5m2_t* dst, __
57 }57 }
58}58}
59 59 
60-__aicore__ inline void asc_copy_gm2l1_dn2nz_sync_impl(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,60+__aicore__ inline void asc_copy_gm2l1_dn2nz_sync_impl(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
61 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)61 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
62{62{
63 asc_copy_gm2l1_dn2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);63 asc_copy_gm2l1_dn2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_gm2l1_impl.h+13-13
@@ -66,8 +66,8 @@ __aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ float *dst, __gm__ floa
66 asc_sync_post_process();66 asc_sync_post_process();
67}67}
68 68 
69-// float8_e4m3_t69+// fp8_e4m3fn_t
70-__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float8_e4m3_t *dst, __gm__ float8_e4m3_t *src, uint32_t m_start_position, uint32_t k_start_position,70+__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ fp8_e4m3fn_t *dst, __gm__ fp8_e4m3fn_t *src, uint32_t m_start_position, uint32_t k_start_position,
71 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)71 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
72{72{
73 if ASC_IS_AIC {73 if ASC_IS_AIC {
@@ -75,15 +75,15 @@ __aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float8_e4m3_t *dst, __gm__ f
75 }75 }
76}76}
77 77 
78-__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ float8_e4m3_t *dst, __gm__ float8_e4m3_t *src, uint32_t m_start_position, uint32_t k_start_position,78+__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ fp8_e4m3fn_t *dst, __gm__ fp8_e4m3fn_t *src, uint32_t m_start_position, uint32_t k_start_position,
79 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)79 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
80{80{
81 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);81 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);
82 asc_sync_post_process();82 asc_sync_post_process();
83}83}
84 84 
85-// float8_e5m2_t85+// fp8_e5m2_t
86-__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float8_e5m2_t *dst, __gm__ float8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position,86+__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ fp8_e5m2_t *dst, __gm__ fp8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position,
87 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)87 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
88{88{
89 if ASC_IS_AIC {89 if ASC_IS_AIC {
@@ -91,7 +91,7 @@ __aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float8_e5m2_t *dst, __gm__ f
91 }91 }
92}92}
93 93 
94-__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ float8_e5m2_t *dst, __gm__ float8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position,94+__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ fp8_e5m2_t *dst, __gm__ fp8_e5m2_t *src, uint32_t m_start_position, uint32_t k_start_position,
95 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)95 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
96{96{
97 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);97 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);
@@ -226,8 +226,8 @@ __aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ uint8_t *dst, __gm__ ui
226 asc_sync_post_process();226 asc_sync_post_process();
227}227}
228 228 
229-// float4_e1m2x2_t229+// fp4x2_e1m2_t
230-__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float4_e1m2x2_t *dst, __gm__ float4_e1m2x2_t *src, uint32_t m_start_position, uint32_t k_start_position,230+__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ fp4x2_e1m2_t *dst, __gm__ fp4x2_e1m2_t *src, uint32_t m_start_position, uint32_t k_start_position,
231 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)231 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
232{232{
233 if ASC_IS_AIC {233 if ASC_IS_AIC {
@@ -235,15 +235,15 @@ __aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float4_e1m2x2_t *dst, __gm__
235 }235 }
236}236}
237 237 
238-__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ float4_e1m2x2_t *dst, __gm__ float4_e1m2x2_t *src, uint32_t m_start_position, uint32_t k_start_position,238+__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ fp4x2_e1m2_t *dst, __gm__ fp4x2_e1m2_t *src, uint32_t m_start_position, uint32_t k_start_position,
239 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)239 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
240{240{
241 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);241 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);
242 asc_sync_post_process();242 asc_sync_post_process();
243}243}
244 244 
245-// float4_e2m1x2_t245+// fp4x2_e2m1_t
246-__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float4_e2m1x2_t *dst, __gm__ float4_e2m1x2_t *src, uint32_t m_start_position, uint32_t k_start_position,246+__aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ fp4x2_e2m1_t *dst, __gm__ fp4x2_e2m1_t *src, uint32_t m_start_position, uint32_t k_start_position,
247 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)247 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
248{248{
249 if ASC_IS_AIC {249 if ASC_IS_AIC {
@@ -251,7 +251,7 @@ __aicore__ inline void asc_copy_gm2l1_impl(__cbuf__ float4_e2m1x2_t *dst, __gm__
251 }251 }
252}252}
253 253 
254-__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ float4_e2m1x2_t *dst, __gm__ float4_e2m1x2_t *src, uint32_t m_start_position, uint32_t k_start_position,254+__aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ fp4x2_e2m1_t *dst, __gm__ fp4x2_e2m1_t *src, uint32_t m_start_position, uint32_t k_start_position,
255 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)255 uint16_t dst_stride, uint16_t m_step, uint16_t k_step, uint8_t decomp_mode, uint8_t l2_cache_ctl)
256{256{
257 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);257 asc_copy_gm2l1_impl(dst, src, m_start_position, k_start_position, dst_stride, m_step, k_step, decomp_mode, l2_cache_ctl);
@@ -279,4 +279,4 @@ __aicore__ inline void asc_copy_gm2l1_sync_impl(__cbuf__ void *dst, __gm__ void
279#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)279#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)
280#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS280#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS
281#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC281#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC
282-#endif282+#endif
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_gm2l1_nd2nz_impl.h+4-4
@@ -35,7 +35,7 @@ __aicore__ inline void asc_copy_gm2l1_nd2nz_sync_impl(__cbuf__ bfloat16_t* dst,
35 asc_sync_post_process();35 asc_sync_post_process();
36}36}
37 37 
38-__aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,38+__aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
39 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)39 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
40{40{
41 if ASC_IS_AIC {41 if ASC_IS_AIC {
@@ -43,14 +43,14 @@ __aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ float8_e4m3_t* dst, __
43 }43 }
44}44}
45 45 
46-__aicore__ inline void asc_copy_gm2l1_nd2nz_sync_impl(__cbuf__ float8_e4m3_t* dst, __gm__ float8_e4m3_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,46+__aicore__ inline void asc_copy_gm2l1_nd2nz_sync_impl(__cbuf__ fp8_e4m3fn_t* dst, __gm__ fp8_e4m3fn_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
47 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)47 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
48{48{
49 asc_copy_gm2l1_nd2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);49 asc_copy_gm2l1_nd2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);
50 asc_sync_post_process();50 asc_sync_post_process();
51}51}
52 52 
53-__aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,53+__aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
54 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)54 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
55{55{
56 if ASC_IS_AIC {56 if ASC_IS_AIC {
@@ -58,7 +58,7 @@ __aicore__ inline void asc_copy_gm2l1_nd2nz_impl(__cbuf__ float8_e5m2_t* dst, __
58 }58 }
59}59}
60 60 
61-__aicore__ inline void asc_copy_gm2l1_nd2nz_sync_impl(__cbuf__ float8_e5m2_t* dst, __gm__ float8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,61+__aicore__ inline void asc_copy_gm2l1_nd2nz_sync_impl(__cbuf__ fp8_e5m2_t* dst, __gm__ fp8_e5m2_t* src, uint64_t loop1_src_stride, uint8_t l2_cache_ctl, uint16_t n_value,
62 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)62 uint32_t d_value, uint64_t loop4_src_stride, bool smallc0_en)
63{63{
64 asc_copy_gm2l1_nd2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);64 asc_copy_gm2l1_nd2nz_impl(dst, src, loop1_src_stride, l2_cache_ctl, n_value, d_value, loop4_src_stride, smallc0_en);
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_l0c2gm_impl.h+12-12
@@ -75,8 +75,8 @@ __aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ half* dst_addr, __cc__ f
75 asc_sync_post_process();75 asc_sync_post_process();
76}76}
77 77 
78-// float8_e4m3_t float78+// fp8_e4m3fn_t float
79-__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e4m3_t *dst_addr, __cc__ float *src_addr,79+__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr,
80 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,80 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
81 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,81 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
82 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,82 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -90,7 +90,7 @@ __aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e4m3_t *dst_addr, __cc
90 }90 }
91}91}
92 92 
93-__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e4m3_t* dst_addr, __cc__ float* src_addr,93+__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ fp8_e4m3fn_t* dst_addr, __cc__ float* src_addr,
94 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,94 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
95 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,95 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
96 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,96 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -103,8 +103,8 @@ __aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e4m3_t* dst_addr,
103 asc_sync_post_process();103 asc_sync_post_process();
104}104}
105 105 
106-// float8_e5m2_t float106+// fp8_e5m2_t float
107-__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e5m2_t *dst_addr, __cc__ float *src_addr,107+__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr,
108 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,108 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
109 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,109 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
110 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,110 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -118,7 +118,7 @@ __aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e5m2_t *dst_addr, __cc
118 }118 }
119}119}
120 120 
121-__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e5m2_t* dst_addr, __cc__ float* src_addr,121+__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ fp8_e5m2_t* dst_addr, __cc__ float* src_addr,
122 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,122 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
123 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,123 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
124 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,124 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -299,8 +299,8 @@ __aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ half* dst_addr, __cc__ i
299 asc_sync_post_process();299 asc_sync_post_process();
300}300}
301 301 
302-// float8_e4m3_t int32_t302+// fp8_e4m3fn_t int32_t
303-__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e4m3_t *dst_addr, __cc__ int32_t *src_addr,303+__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ fp8_e4m3fn_t *dst_addr, __cc__ int32_t *src_addr,
304 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,304 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
305 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,305 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
306 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,306 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -314,7 +314,7 @@ __aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e4m3_t *dst_addr, __cc
314 }314 }
315}315}
316 316 
317-__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e4m3_t* dst_addr, __cc__ int32_t* src_addr,317+__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ fp8_e4m3fn_t* dst_addr, __cc__ int32_t* src_addr,
318 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,318 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
319 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,319 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
320 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,320 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -327,8 +327,8 @@ __aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e4m3_t* dst_addr,
327 asc_sync_post_process();327 asc_sync_post_process();
328}328}
329 329 
330-// float8_e5m2_t int32_t330+// fp8_e5m2_t int32_t
331-__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e5m2_t *dst_addr, __cc__ int32_t *src_addr,331+__aicore__ inline void asc_copy_l0c2gm_impl(__gm__ fp8_e5m2_t *dst_addr, __cc__ int32_t *src_addr,
332 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,332 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
333 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,333 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
334 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,334 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
@@ -342,7 +342,7 @@ __aicore__ inline void asc_copy_l0c2gm_impl(__gm__ float8_e5m2_t *dst_addr, __cc
342 }342 }
343}343}
344 344 
345-__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ float8_e5m2_t* dst_addr, __cc__ int32_t* src_addr,345+__aicore__ inline void asc_copy_l0c2gm_sync_impl(__gm__ fp8_e5m2_t* dst_addr, __cc__ int32_t* src_addr,
346 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,346 uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t l2_cache_ctl,
347 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,347 uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en,
348 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,348 bool NZ2ND_en, uint64_t quant_post, uint8_t relu_post, bool clip_relu_post,
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_l0c2ub_impl.h+12-12
@@ -71,8 +71,8 @@ __aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ half* dst_addr, __cc__
71 asc_sync_post_process();71 asc_sync_post_process();
72}72}
73 73 
74-// float8_e4m3_t float74+// fp8_e4m3fn_t float
75-__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e4m3_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,75+__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,
76 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,76 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
77 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,77 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
78 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,78 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -85,7 +85,7 @@ __aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e4m3_t *dst_addr, __
85 }85 }
86}86}
87 87 
88-__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e4m3_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,88+__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ fp8_e4m3fn_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,
89 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,89 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
90 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,90 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
91 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,91 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -97,8 +97,8 @@ __aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e4m3_t* dst_add
97 asc_sync_post_process();97 asc_sync_post_process();
98}98}
99 99 
100-// float8_e5m2_t float100+// fp8_e5m2_t float
101-__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,101+__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size,
102 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,102 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
103 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,103 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
104 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,104 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -111,7 +111,7 @@ __aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e5m2_t *dst_addr, __
111 }111 }
112}112}
113 113 
114-__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e5m2_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,114+__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ fp8_e5m2_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size,
115 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,115 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
116 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,116 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
117 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,117 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -279,8 +279,8 @@ __aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ half* dst_addr, __cc__
279 asc_sync_post_process();279 asc_sync_post_process();
280}280}
281 281 
282-// float8_e4m3_t int32_t282+// fp8_e4m3fn_t int32_t
283-__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e4m3_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,283+__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,
284 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,284 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
285 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,285 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
286 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,286 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -293,7 +293,7 @@ __aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e4m3_t *dst_addr, __
293 }293 }
294}294}
295 295 
296-__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e4m3_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,296+__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ fp8_e4m3fn_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,
297 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,297 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
298 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,298 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
299 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,299 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -305,8 +305,8 @@ __aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e4m3_t* dst_add
305 asc_sync_post_process();305 asc_sync_post_process();
306}306}
307 307 
308-// float8_e5m2_t int32_t308+// fp8_e5m2_t int32_t
309-__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e5m2_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,309+__aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size,
310 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,310 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
311 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,311 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
312 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,312 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
@@ -319,7 +319,7 @@ __aicore__ inline void asc_copy_l0c2ub_impl(__ubuf__ float8_e5m2_t *dst_addr, __
319 }319 }
320}320}
321 321 
322-__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ float8_e5m2_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,322+__aicore__ inline void asc_copy_l0c2ub_sync_impl(__ubuf__ fp8_e5m2_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size,
323 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,323 uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre,
324 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,324 uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, bool NZ2ND_en, uint64_t quant_post,
325 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,325 uint8_t relu_post, bool clip_relu_post, uint8_t eltwise_op, bool eltwise_antq_en,
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_l12l0a_impl.h+60-60
@@ -21,229 +21,229 @@
21 21 
22__aicore__ inline void asc_copy_l12l0a_impl(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,22__aicore__ inline void asc_copy_l12l0a_impl(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
23 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,23 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
24- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)24+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
25{25{
26 if ASC_IS_AIC {26 if ASC_IS_AIC {
27 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,27 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
28- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);28+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
29 }29 }
30}30}
31 31 
32__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,32__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
33 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,33 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
34- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)34+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
35{35{
36 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,36 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
37- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);37+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
38 asc_sync_post_process();38 asc_sync_post_process();
39}39}
40 40 
41-__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,41+__aicore__ inline void asc_copy_l12l0a_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
42 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,42 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
43- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)43+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
44{44{
45 if ASC_IS_AIC {45 if ASC_IS_AIC {
46 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,46 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
47- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);47+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
48 }48 }
49}49}
50 50 
51-__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,51+__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
52 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,52 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
53- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)53+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
54{54{
55 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,55 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
56- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);56+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
57 asc_sync_post_process();57 asc_sync_post_process();
58}58}
59 59 
60-__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,60+__aicore__ inline void asc_copy_l12l0a_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
61 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,61 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
62- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)62+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
63{63{
64 if ASC_IS_AIC {64 if ASC_IS_AIC {
65 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,65 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
66- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);66+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
67 }67 }
68}68}
69 69 
70-__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,70+__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
71 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,71 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
72- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)72+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
73{73{
74 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,74 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
75- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);75+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
76 asc_sync_post_process();76 asc_sync_post_process();
77}77}
78 78 
79__aicore__ inline void asc_copy_l12l0a_impl(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,79__aicore__ inline void asc_copy_l12l0a_impl(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
80 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,80 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
81- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)81+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
82{82{
83 if ASC_IS_AIC {83 if ASC_IS_AIC {
84 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,84 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
85- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);85+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
86 }86 }
87}87}
88 88 
89__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,89__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
90 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,90 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
91- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)91+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
92{92{
93 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,93 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
94- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);94+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
95 asc_sync_post_process();95 asc_sync_post_process();
96}96}
97 97 
98__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,98__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
99 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,99 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
100- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)100+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
101{101{
102 if ASC_IS_AIC {102 if ASC_IS_AIC {
103 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,103 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
104- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);104+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
105 }105 }
106}106}
107 107 
108__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,108__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
109 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,109 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
110- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)110+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
111{111{
112 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,112 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
113- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);113+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
114 asc_sync_post_process();114 asc_sync_post_process();
115}115}
116 116 
117__aicore__ inline void asc_copy_l12l0a_impl(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,117__aicore__ inline void asc_copy_l12l0a_impl(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
118 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,118 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
119- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)119+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
120{120{
121 if ASC_IS_AIC {121 if ASC_IS_AIC {
122 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,122 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
123- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);123+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
124 }124 }
125}125}
126 126 
127__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,127__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
128 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,128 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
129- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)129+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
130{130{
131 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,131 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
132- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);132+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
133 asc_sync_post_process();133 asc_sync_post_process();
134}134}
135 135 
136__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,136__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
137 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,137 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
138- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)138+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
139{139{
140 if ASC_IS_AIC {140 if ASC_IS_AIC {
141 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,141 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
142- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);142+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
143 }143 }
144}144}
145 145 
146__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,146__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
147 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,147 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
148- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)148+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
149{149{
150 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,150 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
151- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);151+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
152 asc_sync_post_process();152 asc_sync_post_process();
153}153}
154 154 
155__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,155__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
156 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,156 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
157- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)157+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
158{158{
159 if ASC_IS_AIC {159 if ASC_IS_AIC {
160 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,160 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
161- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);161+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
162 }162 }
163}163}
164 164 
165__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,165__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
166 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,166 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
167- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)167+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
168{168{
169 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,169 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
170- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);170+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
171 asc_sync_post_process();171 asc_sync_post_process();
172}172}
173 173 
174__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,174__aicore__ inline void asc_copy_l12l0a_impl(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
175 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,175 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
176- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)176+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
177{177{
178 if ASC_IS_AIC {178 if ASC_IS_AIC {
179 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,179 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
180- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);180+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
181 }181 }
182}182}
183 183 
184__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,184__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
185 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,185 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
186- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)186+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
187{187{
188 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,188 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
189- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);189+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
190 asc_sync_post_process();190 asc_sync_post_process();
191}191}
192 192 
193__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,193__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
194 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,194 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
195- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)195+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
196{196{
197 if ASC_IS_AIC {197 if ASC_IS_AIC {
198 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,198 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
199- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);199+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
200 }200 }
201}201}
202 202 
203__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,203__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
204 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,204 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
205- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)205+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
206{206{
207 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,207 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
208- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);208+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
209 asc_sync_post_process();209 asc_sync_post_process();
210}210}
211 211 
212__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,212__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
213 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,213 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
214- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)214+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
215{215{
216 if ASC_IS_AIC {216 if ASC_IS_AIC {
217 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,217 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
218- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);218+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
219 }219 }
220}220}
221 221 
222__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,222__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
223 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,223 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
224- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)224+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
225{225{
226 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,226 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
227- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);227+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
228 asc_sync_post_process();228 asc_sync_post_process();
229}229}
230 230 
231__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,231__aicore__ inline void asc_copy_l12l0a_impl(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
232 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,232 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
233- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)233+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
234{234{
235 if ASC_IS_AIC {235 if ASC_IS_AIC {
236 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,236 img2colv2_cbuf_to_ca(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
237- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);237+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
238 }238 }
239}239}
240 240 
241__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,241__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
242 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,242 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
243- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)243+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
244{244{
245 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,245 asc_copy_l12l0a_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
246- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);246+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
247 asc_sync_post_process();247 asc_sync_post_process();
248}248}
249 249 
@@ -262,7 +262,7 @@ __aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ bfloat16_t* dst, __cbuf_
262 asc_sync_post_process();262 asc_sync_post_process();
263}263}
264 264 
265-__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,265+__aicore__ inline void asc_copy_l12l0a_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
266 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)266 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
267{267{
268 if ASC_IS_AIC {268 if ASC_IS_AIC {
@@ -270,14 +270,14 @@ __aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e4m3_t* dst, __cbuf__
270 }270 }
271}271}
272 272 
273-__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,273+__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
274 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)274 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
275{275{
276 asc_copy_l12l0a_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);276 asc_copy_l12l0a_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
277 asc_sync_post_process();277 asc_sync_post_process();
278}278}
279 279 
280-__aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,280+__aicore__ inline void asc_copy_l12l0a_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
281 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)281 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
282{282{
283 if ASC_IS_AIC {283 if ASC_IS_AIC {
@@ -285,7 +285,7 @@ __aicore__ inline void asc_copy_l12l0a_impl(__ca__ float8_e5m2_t* dst, __cbuf__
285 }285 }
286}286}
287 287 
288-__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,288+__aicore__ inline void asc_copy_l12l0a_sync_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
289 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)289 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
290{290{
291 asc_copy_l12l0a_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);291 asc_copy_l12l0a_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
@@ -489,7 +489,7 @@ __aicore__ inline void asc_copy_l12l0a_transpose_sync_impl(__ca__ bfloat16_t* ds
489 asc_sync_post_process();489 asc_sync_post_process();
490}490}
491 491 
492-__aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,492+__aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
493 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)493 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
494{494{
495 if ASC_IS_AIC {495 if ASC_IS_AIC {
@@ -497,14 +497,14 @@ __aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ float8_e4m3_t* dst,
497 }497 }
498}498}
499 499 
500-__aicore__ inline void asc_copy_l12l0a_transpose_sync_impl(__ca__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,500+__aicore__ inline void asc_copy_l12l0a_transpose_sync_impl(__ca__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
501 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)501 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
502{502{
503 asc_copy_l12l0a_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);503 asc_copy_l12l0a_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
504 asc_sync_post_process();504 asc_sync_post_process();
505}505}
506 506 
507-__aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,507+__aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
508 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)508 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
509{509{
510 if ASC_IS_AIC {510 if ASC_IS_AIC {
@@ -512,7 +512,7 @@ __aicore__ inline void asc_copy_l12l0a_transpose_impl(__ca__ float8_e5m2_t* dst,
512 }512 }
513}513}
514 514 
515-__aicore__ inline void asc_copy_l12l0a_transpose_sync_impl(__ca__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,515+__aicore__ inline void asc_copy_l12l0a_transpose_sync_impl(__ca__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
516 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)516 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
517{517{
518 asc_copy_l12l0a_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);518 asc_copy_l12l0a_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_l12l0b_impl.h+60-60
@@ -20,229 +20,229 @@
20#include "instr_impl/npu_arch_3510/utils_impl/utils_impl.h"20#include "instr_impl/npu_arch_3510/utils_impl/utils_impl.h"
21__aicore__ inline void asc_copy_l12l0b_impl(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,21__aicore__ inline void asc_copy_l12l0b_impl(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
22 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,22 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
23- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)23+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
24{24{
25 if ASC_IS_AIC {25 if ASC_IS_AIC {
26 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,26 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
27- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);27+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
28 }28 }
29}29}
30 30 
31__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,31__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ bfloat16_t* dst, __cbuf__ bfloat16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
32 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,32 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
33- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)33+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
34{34{
35 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,35 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
36- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);36+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
37 asc_sync_post_process();37 asc_sync_post_process();
38}38}
39 39 
40-__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,40+__aicore__ inline void asc_copy_l12l0b_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
41 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,41 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
42- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)42+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
43{43{
44 if ASC_IS_AIC {44 if ASC_IS_AIC {
45 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,45 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
46- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);46+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
47 }47 }
48}48}
49 49 
50-__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,50+__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
51 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,51 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
52- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)52+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
53{53{
54 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,54 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
55- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);55+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
56 asc_sync_post_process();56 asc_sync_post_process();
57}57}
58 58 
59-__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,59+__aicore__ inline void asc_copy_l12l0b_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
60 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,60 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
61- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)61+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
62{62{
63 if ASC_IS_AIC {63 if ASC_IS_AIC {
64 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,64 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
65- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);65+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
66 }66 }
67}67}
68 68 
69-__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,69+__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
70 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,70 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
71- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)71+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
72{72{
73 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,73 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
74- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);74+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
75 asc_sync_post_process();75 asc_sync_post_process();
76}76}
77 77 
78__aicore__ inline void asc_copy_l12l0b_impl(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,78__aicore__ inline void asc_copy_l12l0b_impl(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
79 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,79 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
80- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)80+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
81{81{
82 if ASC_IS_AIC {82 if ASC_IS_AIC {
83 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,83 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
84- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);84+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
85 }85 }
86}86}
87 87 
88__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,88__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ half* dst, __cbuf__ half* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
89 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,89 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
90- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)90+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
91{91{
92 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,92 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
93- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);93+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
94 asc_sync_post_process();94 asc_sync_post_process();
95}95}
96 96 
97__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,97__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
98 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,98 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
99- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)99+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
100{100{
101 if ASC_IS_AIC {101 if ASC_IS_AIC {
102 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,102 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
103- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);103+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
104 }104 }
105}105}
106 106 
107__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,107__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float* dst, __cbuf__ float* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
108 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,108 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
109- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)109+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
110{110{
111 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,111 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
112- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);112+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
113 asc_sync_post_process();113 asc_sync_post_process();
114}114}
115 115 
116__aicore__ inline void asc_copy_l12l0b_impl(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,116__aicore__ inline void asc_copy_l12l0b_impl(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
117 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,117 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
118- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)118+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
119{119{
120 if ASC_IS_AIC {120 if ASC_IS_AIC {
121 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,121 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
122- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);122+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
123 }123 }
124}124}
125 125 
126__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,126__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ hifloat8_t* dst, __cbuf__ hifloat8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
127 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,127 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
128- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)128+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
129{129{
130 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,130 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
131- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);131+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
132 asc_sync_post_process();132 asc_sync_post_process();
133}133}
134 134 
135__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,135__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
136 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,136 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
137- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)137+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
138{138{
139 if ASC_IS_AIC {139 if ASC_IS_AIC {
140 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,140 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
141- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);141+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
142 }142 }
143}143}
144 144 
145__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,145__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int16_t* dst, __cbuf__ int16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
146 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,146 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
147- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)147+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
148{148{
149 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,149 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
150- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);150+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
151 asc_sync_post_process();151 asc_sync_post_process();
152}152}
153 153 
154__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,154__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
155 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,155 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
156- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)156+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
157{157{
158 if ASC_IS_AIC {158 if ASC_IS_AIC {
159 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,159 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
160- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);160+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
161 }161 }
162}162}
163 163 
164__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,164__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int32_t* dst, __cbuf__ int32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
165 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,165 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
166- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)166+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
167{167{
168 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,168 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
169- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);169+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
170 asc_sync_post_process();170 asc_sync_post_process();
171}171}
172 172 
173__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,173__aicore__ inline void asc_copy_l12l0b_impl(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
174 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,174 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
175- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)175+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
176{176{
177 if ASC_IS_AIC {177 if ASC_IS_AIC {
178 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,178 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
179- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);179+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
180 }180 }
181}181}
182 182 
183__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,183__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ int8_t* dst, __cbuf__ int8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
184 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,184 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
185- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)185+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
186{186{
187 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,187 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
188- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);188+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
189 asc_sync_post_process();189 asc_sync_post_process();
190}190}
191 191 
192__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,192__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
193 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,193 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
194- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)194+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
195{195{
196 if ASC_IS_AIC {196 if ASC_IS_AIC {
197 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,197 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
198- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);198+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
199 }199 }
200}200}
201 201 
202__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,202__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint16_t* dst, __cbuf__ uint16_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
203 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,203 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
204- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)204+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
205{205{
206 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,206 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
207- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);207+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
208 asc_sync_post_process();208 asc_sync_post_process();
209}209}
210 210 
211__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,211__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
212 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,212 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
213- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)213+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
214{214{
215 if ASC_IS_AIC {215 if ASC_IS_AIC {
216 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,216 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
217- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);217+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
218 }218 }
219}219}
220 220 
221__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,221__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint32_t* dst, __cbuf__ uint32_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
222 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,222 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
223- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)223+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
224{224{
225 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,225 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
226- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);226+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
227 asc_sync_post_process();227 asc_sync_post_process();
228}228}
229 229 
230__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,230__aicore__ inline void asc_copy_l12l0b_impl(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
231 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,231 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
232- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)232+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
233{233{
234 if ASC_IS_AIC {234 if ASC_IS_AIC {
235 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,235 img2colv2_cbuf_to_cb(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
236- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);236+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
237 }237 }
238}238}
239 239 
240__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,240__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ uint8_t* dst, __cbuf__ uint8_t* src, uint16_t k_extension, uint16_t m_extension, uint16_t k_start_pt,
241 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,241 uint16_t m_start_pt, uint8_t stride_w, uint8_t stride_h, uint8_t filter_w, uint8_t filter_h, uint8_t dilation_filter_w,
242- uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool fmatrix_ctrl, uint16_t channel_size)242+ uint8_t dilation_filter_h, bool filter_size_w, bool filter_size_h, bool transpose, bool f_matrix_ctrl, uint16_t channel_size)
243{243{
244 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,244 asc_copy_l12l0b_impl(dst, src, k_extension, m_extension, k_start_pt, m_start_pt, stride_w, stride_h, filter_w, filter_h,
245- dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, fmatrix_ctrl, channel_size);245+ dilation_filter_w, dilation_filter_h, filter_size_w, filter_size_h, transpose, f_matrix_ctrl, channel_size);
246 asc_sync_post_process();246 asc_sync_post_process();
247}247}
248 248 
@@ -261,7 +261,7 @@ __aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ bfloat16_t* dst, __cbuf_
261 asc_sync_post_process();261 asc_sync_post_process();
262}262}
263 263 
264-__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,264+__aicore__ inline void asc_copy_l12l0b_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
265 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)265 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
266{266{
267 if ASC_IS_AIC {267 if ASC_IS_AIC {
@@ -269,14 +269,14 @@ __aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e4m3_t* dst, __cbuf__
269 }269 }
270}270}
271 271 
272-__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,272+__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
273 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)273 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
274{274{
275 asc_copy_l12l0b_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);275 asc_copy_l12l0b_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
276 asc_sync_post_process();276 asc_sync_post_process();
277}277}
278 278 
279-__aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,279+__aicore__ inline void asc_copy_l12l0b_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
280 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)280 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
281{281{
282 if ASC_IS_AIC {282 if ASC_IS_AIC {
@@ -284,7 +284,7 @@ __aicore__ inline void asc_copy_l12l0b_impl(__cb__ float8_e5m2_t* dst, __cbuf__
284 }284 }
285}285}
286 286 
287-__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,287+__aicore__ inline void asc_copy_l12l0b_sync_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
288 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)288 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
289{289{
290 asc_copy_l12l0b_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);290 asc_copy_l12l0b_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
@@ -487,7 +487,7 @@ __aicore__ inline void asc_copy_l12l0b_transpose_sync_impl(__cb__ bfloat16_t* ds
487 asc_sync_post_process();487 asc_sync_post_process();
488}488}
489 489 
490-__aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,490+__aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
491 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)491 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
492{492{
493 if ASC_IS_AIC {493 if ASC_IS_AIC {
@@ -495,14 +495,14 @@ __aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ float8_e4m3_t* dst,
495 }495 }
496}496}
497 497 
498-__aicore__ inline void asc_copy_l12l0b_transpose_sync_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t m_start_position, uint16_t k_start_position,498+__aicore__ inline void asc_copy_l12l0b_transpose_sync_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t m_start_position, uint16_t k_start_position,
499 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)499 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
500{500{
501 asc_copy_l12l0b_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);501 asc_copy_l12l0b_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
502 asc_sync_post_process();502 asc_sync_post_process();
503}503}
504 504 
505-__aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,505+__aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
506 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)506 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
507{507{
508 if ASC_IS_AIC {508 if ASC_IS_AIC {
@@ -510,7 +510,7 @@ __aicore__ inline void asc_copy_l12l0b_transpose_impl(__cb__ float8_e5m2_t* dst,
510 }510 }
511}511}
512 512 
513-__aicore__ inline void asc_copy_l12l0b_transpose_sync_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,513+__aicore__ inline void asc_copy_l12l0b_transpose_sync_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t m_start_position, uint16_t k_start_position,
514 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)514 uint8_t m_step, uint8_t k_step, int16_t src_stride, uint16_t dst_stride)
515{515{
516 asc_copy_l12l0b_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);516 asc_copy_l12l0b_transpose_impl(dst, src, m_start_position, k_start_position, m_step, k_step, src_stride, dst_stride);
Mimpl/c_api/instr_impl/npu_arch_3510/cube_datamove_impl/asc_copy_l12l0b_trans_impl.h+9-9
@@ -40,21 +40,21 @@ __aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ bfloat16_t* dst, _
40 asc_sync_post_process();40 asc_sync_post_process();
41}41}
42 42 
43-__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,43+__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
44 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)44 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
45{45{
46 if ASC_IS_AIC {46 if ASC_IS_AIC {
47 load_cbuf_to_cb_transpose(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);47 load_cbuf_to_cb_transpose(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);
48 }48 }
49}49}
50-__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ float8_e4m3_t* dst, __cbuf__ float8_e4m3_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,50+__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ fp8_e4m3fn_t* dst, __cbuf__ fp8_e4m3fn_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
51 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)51 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
52{52{
53 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);53 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);
54 asc_sync_post_process();54 asc_sync_post_process();
55}55}
56 56 
57-__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,57+__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
58 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)58 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
59{59{
60 if ASC_IS_AIC {60 if ASC_IS_AIC {
@@ -62,7 +62,7 @@ __aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ float8_e5m2_t* dst, __c
62 }62 }
63}63}
64 64 
65-__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ float8_e5m2_t* dst, __cbuf__ float8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,65+__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ fp8_e5m2_t* dst, __cbuf__ fp8_e5m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
66 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)66 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
67{67{
68 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);68 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);
@@ -173,28 +173,28 @@ __aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ uint8_t* dst, __cb
173 asc_sync_post_process();173 asc_sync_post_process();
174}174}
175 175 
176-__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ float4_e1m2x2_t* dst, __cbuf__ float4_e1m2x2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,176+__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
177 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)177 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
178{178{
179 if ASC_IS_AIC {179 if ASC_IS_AIC {
180 load_cbuf_to_cb_transpose_s4(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);180 load_cbuf_to_cb_transpose_s4(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);
181 }181 }
182}182}
183-__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ float4_e1m2x2_t* dst, __cbuf__ float4_e1m2x2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,183+__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ fp4x2_e1m2_t* dst, __cbuf__ fp4x2_e1m2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
184 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)184 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
185{185{
186 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);186 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);
187 asc_sync_post_process();187 asc_sync_post_process();
188}188}
189 189 
190-__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ float4_e2m1x2_t* dst, __cbuf__ float4_e2m1x2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,190+__aicore__ inline void asc_copy_l12l0b_trans_impl(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
191 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)191 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
192{192{
193 if ASC_IS_AIC {193 if ASC_IS_AIC {
194 load_cbuf_to_cb_transpose_s4(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);194 load_cbuf_to_cb_transpose_s4(dst, src, index_id, repeat, src_stride, dst_stride, inc, dst_frac_stride, src_frac_stride);
195 }195 }
196}196}
197-__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ float4_e2m1x2_t* dst, __cbuf__ float4_e2m1x2_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,197+__aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ fp4x2_e2m1_t* dst, __cbuf__ fp4x2_e2m1_t* src, uint16_t index_id, uint8_t repeat, uint16_t src_stride,
198 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)198 uint16_t dst_stride, uint16_t dst_frac_stride, uint16_t src_frac_stride)
199{199{
200 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);200 asc_copy_l12l0b_trans_impl(dst, src, index_id, repeat, src_stride, dst_stride, dst_frac_stride, src_frac_stride);
@@ -222,4 +222,4 @@ __aicore__ inline void asc_copy_l12l0b_trans_sync_impl(__cb__ int4b_t* dst, __cb
222#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)222#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)
223#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS223#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS
224#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC224#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC
225-#endif225+#endif
Mimpl/c_api/instr_impl/npu_arch_3510/vector_compute_impl.h+22-22
@@ -2887,7 +2887,7 @@ __simd_callee__ inline void asc_min_scalar(vector_float& dst, vector_float src,
2887 asc_min_scalar_impl(dst, src, value, mask);2887 asc_min_scalar_impl(dst, src, value, mask);
2888}2888}
2889 2889 
2890-__simd_callee__ inline void asc_min_scalar(vector_half& dst, vector_f16 src, half value, vector_bool mask)2890+__simd_callee__ inline void asc_min_scalar(vector_half& dst, vector_half src, half value, vector_bool mask)
2891{2891{
2892 asc_min_scalar_impl(dst, src, value, mask);2892 asc_min_scalar_impl(dst, src, value, mask);
2893}2893}
@@ -3026,103 +3026,103 @@ __simd_callee__ inline void asc_gt_scalar(vector_bool& dst, vector_float src, fl
3026 asc_gt_scalar_impl(dst, src, value, mask);3026 asc_gt_scalar_impl(dst, src, value, mask);
3027}3027}
3028 3028 
3029-//=================aschal2uint8===============3029+//=================asc_half2uint8===============
3030-__simd_callee__ inline void asc_half2uint8_rd(vector_u8& dst, vector_f16 src, vector_bool mask)3030+__simd_callee__ inline void asc_half2uint8_rd(vector_uint8_t& dst, vector_half src, vector_bool mask)
3031{3031{
3032 asc_half2uint8_rd_impl(dst, src, mask);3032 asc_half2uint8_rd_impl(dst, src, mask);
3033}3033}
3034 3034 
3035-__simd_callee__ inline void asc_half2uint8_rd_sat(vector_u8& dst, vector_f16 src, vector_bool mask)3035+__simd_callee__ inline void asc_half2uint8_rd_sat(vector_uint8_t& dst, vector_half src, vector_bool mask)
3036{3036{
3037 asc_half2uint8_rd_sat_impl(dst, src, mask);3037 asc_half2uint8_rd_sat_impl(dst, src, mask);
3038}3038}
3039 3039 
3040-__simd_callee__ inline void asc_half2uint8_rd_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3040+__simd_callee__ inline void asc_half2uint8_rd_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3041{3041{
3042 asc_half2uint8_rd_v2_impl(dst, src, mask);3042 asc_half2uint8_rd_v2_impl(dst, src, mask);
3043}3043}
3044 3044 
3045-__simd_callee__ inline void asc_half2uint8_rd_sat_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3045+__simd_callee__ inline void asc_half2uint8_rd_sat_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3046{3046{
3047 asc_half2uint8_rd_sat_v2_impl(dst, src, mask);3047 asc_half2uint8_rd_sat_v2_impl(dst, src, mask);
3048}3048}
3049 3049 
3050-__simd_callee__ inline void asc_half2uint8_rn(vector_u8& dst, vector_f16 src, vector_bool mask)3050+__simd_callee__ inline void asc_half2uint8_rn(vector_uint8_t& dst, vector_half src, vector_bool mask)
3051{3051{
3052 asc_half2uint8_rn_impl(dst, src, mask);3052 asc_half2uint8_rn_impl(dst, src, mask);
3053}3053}
3054 3054 
3055-__simd_callee__ inline void asc_half2uint8_rn_sat(vector_u8& dst, vector_f16 src, vector_bool mask)3055+__simd_callee__ inline void asc_half2uint8_rn_sat(vector_uint8_t& dst, vector_half src, vector_bool mask)
3056{3056{
3057 asc_half2uint8_rn_sat_impl(dst, src, mask);3057 asc_half2uint8_rn_sat_impl(dst, src, mask);
3058}3058}
3059 3059 
3060-__simd_callee__ inline void asc_half2uint8_rn_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3060+__simd_callee__ inline void asc_half2uint8_rn_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3061{3061{
3062 asc_half2uint8_rn_v2_impl(dst, src, mask);3062 asc_half2uint8_rn_v2_impl(dst, src, mask);
3063}3063}
3064 3064 
3065-__simd_callee__ inline void asc_half2uint8_rn_sat_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3065+__simd_callee__ inline void asc_half2uint8_rn_sat_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3066{3066{
3067 asc_half2uint8_rn_sat_v2_impl(dst, src, mask);3067 asc_half2uint8_rn_sat_v2_impl(dst, src, mask);
3068}3068}
3069 3069 
3070-__simd_callee__ inline void asc_half2uint8_rna(vector_u8& dst, vector_f16 src, vector_bool mask)3070+__simd_callee__ inline void asc_half2uint8_rna(vector_uint8_t& dst, vector_half src, vector_bool mask)
3071{3071{
3072 asc_half2uint8_rna_impl(dst, src, mask);3072 asc_half2uint8_rna_impl(dst, src, mask);
3073}3073}
3074 3074 
3075-__simd_callee__ inline void asc_half2uint8_rna_sat(vector_u8& dst, vector_f16 src, vector_bool mask)3075+__simd_callee__ inline void asc_half2uint8_rna_sat(vector_uint8_t& dst, vector_half src, vector_bool mask)
3076{3076{
3077 asc_half2uint8_rna_sat_impl(dst, src, mask);3077 asc_half2uint8_rna_sat_impl(dst, src, mask);
3078}3078}
3079 3079 
3080-__simd_callee__ inline void asc_half2uint8_rna_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3080+__simd_callee__ inline void asc_half2uint8_rna_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3081{3081{
3082 asc_half2uint8_rna_v2_impl(dst, src, mask);3082 asc_half2uint8_rna_v2_impl(dst, src, mask);
3083}3083}
3084 3084 
3085-__simd_callee__ inline void asc_half2uint8_rna_sat_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3085+__simd_callee__ inline void asc_half2uint8_rna_sat_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3086{3086{
3087 asc_half2uint8_rna_sat_v2_impl(dst, src, mask);3087 asc_half2uint8_rna_sat_v2_impl(dst, src, mask);
3088}3088}
3089 3089 
3090-__simd_callee__ inline void asc_half2uint8_ru(vector_u8& dst, vector_f16 src, vector_bool mask)3090+__simd_callee__ inline void asc_half2uint8_ru(vector_uint8_t& dst, vector_half src, vector_bool mask)
3091{3091{
3092 asc_half2uint8_ru_impl(dst, src, mask);3092 asc_half2uint8_ru_impl(dst, src, mask);
3093}3093}
3094 3094 
3095-__simd_callee__ inline void asc_half2uint8_ru_sat(vector_u8& dst, vector_f16 src, vector_bool mask)3095+__simd_callee__ inline void asc_half2uint8_ru_sat(vector_uint8_t& dst, vector_half src, vector_bool mask)
3096{3096{
3097 asc_half2uint8_ru_sat_impl(dst, src, mask);3097 asc_half2uint8_ru_sat_impl(dst, src, mask);
3098}3098}
3099 3099 
3100-__simd_callee__ inline void asc_half2uint8_ru_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3100+__simd_callee__ inline void asc_half2uint8_ru_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3101{3101{
3102 asc_half2uint8_ru_v2_impl(dst, src, mask);3102 asc_half2uint8_ru_v2_impl(dst, src, mask);
3103}3103}
3104 3104 
3105-__simd_callee__ inline void asc_half2uint8_ru_sat_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3105+__simd_callee__ inline void asc_half2uint8_ru_sat_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3106{3106{
3107 asc_half2uint8_ru_sat_v2_impl(dst, src, mask);3107 asc_half2uint8_ru_sat_v2_impl(dst, src, mask);
3108}3108}
3109 3109 
3110-__simd_callee__ inline void asc_half2uint8_rz(vector_u8& dst, vector_f16 src, vector_bool mask)3110+__simd_callee__ inline void asc_half2uint8_rz(vector_uint8_t& dst, vector_half src, vector_bool mask)
3111{3111{
3112 asc_half2uint8_rz_impl(dst, src, mask);3112 asc_half2uint8_rz_impl(dst, src, mask);
3113}3113}
3114 3114 
3115-__simd_callee__ inline void asc_half2uint8_rz_sat(vector_u8& dst, vector_f16 src, vector_bool mask)3115+__simd_callee__ inline void asc_half2uint8_rz_sat(vector_uint8_t& dst, vector_half src, vector_bool mask)
3116{3116{
3117 asc_half2uint8_rz_sat_impl(dst, src, mask);3117 asc_half2uint8_rz_sat_impl(dst, src, mask);
3118}3118}
3119 3119 
3120-__simd_callee__ inline void asc_half2uint8_rz_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3120+__simd_callee__ inline void asc_half2uint8_rz_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3121{3121{
3122 asc_half2uint8_rz_v2_impl(dst, src, mask);3122 asc_half2uint8_rz_v2_impl(dst, src, mask);
3123}3123}
3124 3124 
3125-__simd_callee__ inline void asc_half2uint8_rz_sat_v2(vector_u8& dst, vector_f16 src, vector_bool mask)3125+__simd_callee__ inline void asc_half2uint8_rz_sat_v2(vector_uint8_t& dst, vector_half src, vector_bool mask)
3126{3126{
3127 asc_half2uint8_rz_sat_v2_impl(dst, src, mask);3127 asc_half2uint8_rz_sat_v2_impl(dst, src, mask);
3128}3128}
Mimpl/c_api/instr_impl/npu_arch_3510/vector_datamove_impl.h+1-1
@@ -6277,7 +6277,7 @@ __simd_callee__ inline void asc_scatter(vector_uint32_t& dst, __ubuf__ uint32_t*
6277 asc_scatter_impl(dst, src, index, mask);6277 asc_scatter_impl(dst, src, index, mask);
6278}6278}
6279 6279 
6280-__simd_callee__ inline void asc_scatter(vector_bf16& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)6280+__simd_callee__ inline void asc_scatter(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)
6281{6281{
6282 asc_scatter_impl(dst, src, index, mask);6282 asc_scatter_impl(dst, src, index, mask);
6283}6283}
Mimpl/c_api/instr_impl/npu_arch_3510/vector_datamove_impl/asc_scatter_impl.h+2-2
@@ -74,7 +74,7 @@ __simd_callee__ inline void asc_scatter_impl(vector_uint32_t& dst, __ubuf__ uint
74}74}
75 75 
76//asc_scatter bfloat16_t76//asc_scatter bfloat16_t
77-__simd_callee__ inline void asc_scatter_impl(vector_bf16& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)77+__simd_callee__ inline void asc_scatter_impl(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask)
78{78{
79 if ASC_IS_AIV{79 if ASC_IS_AIV{
80 vscatter(dst, src, index, mask);80 vscatter(dst, src, index, mask);
@@ -102,4 +102,4 @@ __simd_callee__ inline void asc_scatter_impl(vector_float& dst, __ubuf__ float*
102#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)102#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)
103#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS103#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS
104#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC104#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC
105-#endif105+#endif
Mimpl/c_api/instr_impl/npu_arch_3510/vector_datamove_impl/reg_store/asc_storealign_impl.h+2-2
@@ -1823,7 +1823,7 @@ union asc_store_align_config{
1823 };1823 };
1824};1824};
1825 1825 
1826-// ==========asc_storealign_impl(int8_t/uint8_t/int16_t/uint16_t/int32_t/uint32_t/float8_e4m3_t/float8_e5m2_t/float8_e8m0_t/bfloat16_t/half/float)==========1826+// ==========asc_storealign_impl(int8_t/uint8_t/int16_t/uint16_t/int32_t/uint32_t/fp8_e4m3fn_t/fp8_e5m2_t/fp8_e8m0_t/bfloat16_t/half/float)==========
1827__simd_callee__ inline void asc_storealign_impl(__ubuf__ int8_t* dst_align32b, vector_int8_t src, uint16_t block_stride, uint16_t repeat_stride, vector_bool mask)1827__simd_callee__ inline void asc_storealign_impl(__ubuf__ int8_t* dst_align32b, vector_int8_t src, uint16_t block_stride, uint16_t repeat_stride, vector_bool mask)
1828{1828{
1829 if ASC_IS_AIV {1829 if ASC_IS_AIV {
@@ -1966,7 +1966,7 @@ union asc_store_align_config_post{
1966 };1966 };
1967};1967};
1968 1968 
1969-// ==========asc_storealign_postupdate_impl(int8_t/uint8_t/int16_t/uint16_t/int32_t/uint32_t/float8_e4m3_t/float8_e5m2_t/float8_e8m0_t/bfloat16_t/half/float)==========1969+// ==========asc_storealign_postupdate_impl(int8_t/uint8_t/int16_t/uint16_t/int32_t/uint32_t/fp8_e4m3fn_t/fp8_e5m2_t/fp8_e8m0_t/bfloat16_t/half/float)==========
1970__simd_callee__ inline void asc_storealign_postupdate_impl(__ubuf__ int8_t* dst_align32b, vector_int8_t src, uint16_t block_stride, uint16_t repeat_stride, vector_bool mask)1970__simd_callee__ inline void asc_storealign_postupdate_impl(__ubuf__ int8_t* dst_align32b, vector_int8_t src, uint16_t block_stride, uint16_t repeat_stride, vector_bool mask)
1971{1971{
1972 if ASC_IS_AIV {1972 if ASC_IS_AIV {
Minclude/c_api/cube_compute/cube_compute.h+12-12
@@ -204,39 +204,39 @@ __aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ bfloat16_t*
204 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,204 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
205 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);205 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
206 206 
207-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e4m3_t* b_matrix,207+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix,
208 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,208 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
209 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);209 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
210 210 
211-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,211+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
212- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,212+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
213 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,213 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
214 bool c_matrix_init_val);214 bool c_matrix_init_val);
215 215 
216-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix, __cb__ float8_e5m2_t* b_matrix,216+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix,
217 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,217 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
218 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);218 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
219 219 
220-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e4m3_t* a_matrix,220+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e4m3fn_t* a_matrix,
221- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,221+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
222 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,222 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
223 bool c_matrix_init_val);223 bool c_matrix_init_val);
224 224 
225-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e4m3_t* b_matrix,225+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e4m3fn_t* b_matrix,
226 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,226 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
227 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);227 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
228 228 
229-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,229+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
230- __cb__ float8_e4m3_t* b_matrix, uint16_t left_height, uint16_t n_dim,230+ __cb__ fp8_e4m3fn_t* b_matrix, uint16_t left_height, uint16_t n_dim,
231 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,231 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
232 bool c_matrix_init_val);232 bool c_matrix_init_val);
233 233 
234-__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix, __cb__ float8_e5m2_t* b_matrix,234+__aicore__ inline void asc_mmad(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix, __cb__ fp8_e5m2_t* b_matrix,
235 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,235 uint16_t left_height, uint16_t n_dim, uint16_t right_width, uint8_t unit_flag,
236 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);236 bool disable_gemv, bool c_matrix_source, bool c_matrix_init_val);
237 237 
238-__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ float8_e5m2_t* a_matrix,238+__aicore__ inline void asc_mmad_sync(__cc__ float* c_matrix, __ca__ fp8_e5m2_t* a_matrix,
239- __cb__ float8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,239+ __cb__ fp8_e5m2_t* b_matrix, uint16_t left_height, uint16_t n_dim,
240 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,240 uint16_t right_width, uint8_t unit_flag, bool disable_gemv, bool c_matrix_source,
241 bool c_matrix_init_val);241 bool c_matrix_init_val);
242 242 
Minclude/c_api/reg_compute/reg_convert.h+21-21
@@ -487,45 +487,45 @@ __simd_callee__ inline void asc_hif82float_v3(vector_float& dst, vector_hifloat8
487__simd_callee__ inline void asc_hif82float_v4(vector_float& dst, vector_hifloat8_t src, vector_bool mask);487__simd_callee__ inline void asc_hif82float_v4(vector_float& dst, vector_hifloat8_t src, vector_bool mask);
488 488 
489//=================asc_half2uint8===============489//=================asc_half2uint8===============
490-__simd_callee__ inline void asc_half2uint8_rd(vector_u8& dst, vector_f16 src, vector_bool mask);490+__simd_callee__ inline void asc_half2uint8_rd(vector_uint8_t& dst, vector_half src, vector_bool mask);
491 491 
492-__simd_callee__ inline void asc_half2uint8_rd_sat(vector_u8& dst, vector_f16 src, vector_bool mask);492+__simd_callee__ inline void asc_half2uint8_rd_sat(vector_uint8_t& dst, vector_half src, vector_bool mask);
493 493 
494-__simd_callee__ inline void asc_half2uint8_rd_v2(vector_u8& dst, vector_f16 src, vector_bool mask);494+__simd_callee__ inline void asc_half2uint8_rd_v2(vector_uint8_t& dst, vector_half src, vector_bool mask);
495 495 
496-__simd_callee__ inline void asc_half2uint8_rd_sat_v2_impl(vector_u8& dst, vector_f16 src, vector_bool mask);496+__simd_callee__ inline void asc_half2uint8_rd_sat_v2_impl(vector_uint8_t& dst, vector_half src, vector_bool mask);
497 497 
498-__simd_callee__ inline void asc_half2uint8_rn(vector_u8& dst, vector_f16 src, vector_bool mask);498+__simd_callee__ inline void asc_half2uint8_rn(vector_uint8_t& dst, vector_half src, vector_bool mask);
499 499 
500-__simd_callee__ inline void asc_half2uint8_rn_sat(vector_u8& dst, vector_f16 src, vector_bool mask);500+__simd_callee__ inline void asc_half2uint8_rn_sat(vector_uint8_t& dst, vector_half src, vector_bool mask);
501 501 
502-__simd_callee__ inline void asc_half2uint8_rn_v2(vector_u8& dst, vector_f16 src, vector_bool mask);502+__simd_callee__ inline void asc_half2uint8_rn_v2(vector_uint8_t& dst, vector_half src, vector_bool mask);
503 503 
504-__simd_callee__ inline void asc_half2uint8_rn_sat_v2_impl(vector_u8& dst, vector_f16 src, vector_bool mask);504+__simd_callee__ inline void asc_half2uint8_rn_sat_v2_impl(vector_uint8_t& dst, vector_half src, vector_bool mask);
505 505 
506-__simd_callee__ inline void asc_half2uint8_rna(vector_u8& dst, vector_f16 src, vector_bool mask);506+__simd_callee__ inline void asc_half2uint8_rna(vector_uint8_t& dst, vector_half src, vector_bool mask);
507 507 
508-__simd_callee__ inline void asc_half2uint8_rna_sat(vector_u8& dst, vector_f16 src, vector_bool mask);508+__simd_callee__ inline void asc_half2uint8_rna_sat(vector_uint8_t& dst, vector_half src, vector_bool mask);
509 509 
510-__simd_callee__ inline void asc_half2uint8_rna_v2(vector_u8& dst, vector_f16 src, vector_bool mask);510+__simd_callee__ inline void asc_half2uint8_rna_v2(vector_uint8_t& dst, vector_half src, vector_bool mask);
511 511 
512-__simd_callee__ inline void asc_half2uint8_rna_sat_v2_impl(vector_u8& dst, vector_f16 src, vector_bool mask);512+__simd_callee__ inline void asc_half2uint8_rna_sat_v2_impl(vector_uint8_t& dst, vector_half src, vector_bool mask);
513 513 
514-__simd_callee__ inline void asc_half2uint8_ru(vector_u8& dst, vector_f16 src, vector_bool mask);514+__simd_callee__ inline void asc_half2uint8_ru(vector_uint8_t& dst, vector_half src, vector_bool mask);
515 515 
516-__simd_callee__ inline void asc_half2uint8_ru_sat(vector_u8& dst, vector_f16 src, vector_bool mask);516+__simd_callee__ inline void asc_half2uint8_ru_sat(vector_uint8_t& dst, vector_half src, vector_bool mask);
517 517 
518-__simd_callee__ inline void asc_half2uint8_ru_v2(vector_u8& dst, vector_f16 src, vector_bool mask);518+__simd_callee__ inline void asc_half2uint8_ru_v2(vector_uint8_t& dst, vector_half src, vector_bool mask);
519 519 
520-__simd_callee__ inline void asc_half2uint8_ru_sat_v2_impl(vector_u8& dst, vector_f16 src, vector_bool mask);520+__simd_callee__ inline void asc_half2uint8_ru_sat_v2_impl(vector_uint8_t& dst, vector_half src, vector_bool mask);
521 521 
522-__simd_callee__ inline void asc_half2uint8_rz(vector_u8& dst, vector_f16 src, vector_bool mask);522+__simd_callee__ inline void asc_half2uint8_rz(vector_uint8_t& dst, vector_half src, vector_bool mask);
523 523 
524-__simd_callee__ inline void asc_half2uint8_rz_sat(vector_u8& dst, vector_f16 src, vector_bool mask);524+__simd_callee__ inline void asc_half2uint8_rz_sat(vector_uint8_t& dst, vector_half src, vector_bool mask);
525 525 
526-__simd_callee__ inline void asc_half2uint8_rz_v2(vector_u8& dst, vector_f16 src, vector_bool mask);526+__simd_callee__ inline void asc_half2uint8_rz_v2(vector_uint8_t& dst, vector_half src, vector_bool mask);
527 527 
528-__simd_callee__ inline void asc_half2uint8_rz_sat_v2_impl(vector_u8& dst, vector_f16 src, vector_bool mask);528+__simd_callee__ inline void asc_half2uint8_rz_sat_v2_impl(vector_uint8_t& dst, vector_half src, vector_bool mask);
529 529 
530//===================asc_int162half_impl=================530//===================asc_int162half_impl=================
531__simd_callee__ inline void asc_int162half_rd(vector_half& dst, vector_int16_t src, vector_bool mask);531__simd_callee__ inline void asc_int162half_rd(vector_half& dst, vector_int16_t src, vector_bool mask);
@@ -1002,4 +1002,4 @@ __simd_callee__ inline void asc_half2float_v2(vector_float& dst, vector_half src
1002#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H) 1002#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H)
1003#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS 1003#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS
1004#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H 1004#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H
1005-#endif 1005+#endif
Minclude/c_api/reg_compute/reg_load.h+1-1
@@ -1593,7 +1593,7 @@ __simd_callee__ inline void asc_scatter(vector_int32_t& dst, __ubuf__ int32_t* s
1593 1593 
1594__simd_callee__ inline void asc_scatter(vector_uint32_t& dst, __ubuf__ uint32_t* src, vector_uint32_t index, vector_bool mask);1594__simd_callee__ inline void asc_scatter(vector_uint32_t& dst, __ubuf__ uint32_t* src, vector_uint32_t index, vector_bool mask);
1595 1595 
1596-__simd_callee__ inline void asc_scatter(vector_bf16& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask);1596+__simd_callee__ inline void asc_scatter(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, vector_uint16_t index, vector_bool mask);
1597 1597 
1598__simd_callee__ inline void asc_scatter(vector_half& dst, __ubuf__ half* src, vector_uint16_t index, vector_bool mask);1598__simd_callee__ inline void asc_scatter(vector_half& dst, __ubuf__ half* src, vector_uint16_t index, vector_bool mask);
1599 1599 
Minclude/c_api/reg_compute/reg_vector.h+2-2
@@ -969,7 +969,7 @@ __simd_callee__ inline void asc_min_scalar(vector_uint8_t& dst, vector_uint8_t s
969 969 
970__simd_callee__ inline void asc_min_scalar(vector_float& dst, vector_float src, float value, vector_bool mask);970__simd_callee__ inline void asc_min_scalar(vector_float& dst, vector_float src, float value, vector_bool mask);
971 971 
972-__simd_callee__ inline void asc_min_scalar(vector_half& dst, vector_f16 src, half value, vector_bool mask);972+__simd_callee__ inline void asc_min_scalar(vector_half& dst, vector_half src, half value, vector_bool mask);
973 973 
974//==================asc_reduce_max_datablock====================974//==================asc_reduce_max_datablock====================
975__simd_callee__ inline void asc_reduce_max_datablock(vector_half& dst, vector_half src, vector_bool mask);975__simd_callee__ inline void asc_reduce_max_datablock(vector_half& dst, vector_half src, vector_bool mask);
@@ -1254,4 +1254,4 @@ __aicore__ inline void asc_set_va_reg(ub_addr8_t addr, __ubuf__ float** src_arra
1254#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H) 1254#if defined(UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H)
1255#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS 1255#undef ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS
1256#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H 1256#undef UNDEF_ASCENDC_C_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC_C_API_H
1257-#endif 1257+#endif