asc_copy_l0c2ub【免费下载链接】asc-devkit本项目是CANN 推出的昇腾AI处理器专用的算子程序开发语言原生支持C和C标准规范主要由类库和语言扩展层构成提供多层级API满足多维场景算子开发诉求。项目地址: https://gitcode.com/cann/asc-devkit产品支持情况产品是否支持Ascend 950PR/Ascend 950DT√功能说明矩阵计算完成后对结果进行量化处理之后将处理结果搬运到Unified Buffer中。量化参数共有2个quant_pre和quant_post分别对应预处理和后处理阶段。quant_pre可选量化模式分别为NoQuant不使能量化功能。F322BF16float量化成bfloat16_t。量化结果不支持INF_NAN模式。F322F16float量化成half。量化结果支持INF_NAN模式。DEQF16int32_t量化成half。量化结果不支持INF_NAN模式。VDEQF16int32_t量化成half。量化结果不支持INF_NAN模式。QF322B8_PREfloat量化成uint8_t/int8_t。scalar量化。VQF322B8_PREfloat量化成uint8_t/int8_t。scalar量化。REQ8int32_t量化成uint8_t/int8_t。scalar量化。VREQ8int32_t量化成uint8_t/int8_t。矢量量化。QF322FP8_PREfloat量化成fp8_e4m3fn_tscalar量化。VQF322FP8_PREfloat量化成fp8_e4m3fn_t矢量量化。QF322HIF8_PREfloat量化成hifloat8_tHalf to Away Roundscalar量化。VQF322HIF8_PREfloat量化成hifloat8_tHalf to Away Round矢量量化。QF322HIF8_PRE_HYBRIDfloat量化成hifloat8_tHybrid Roundscalar量化。VQF322HIF8_PRE_HYBRIDfloat量化成hifloat8_tHybrid Round矢量量化。QS322BF16_PREint32_t量化成bfloat16_tscalar量化。VQS322BF16_PREint32_t量化成bfloat16_t矢量量化。QF322F16_PREfloat量化成halfscalar量化。VQF322F16_PREfloat量化成half矢量量化。QF322BF16_PREfloat量化成bfloat16_tscalar量化。VQF322BF16_PREfloat量化成bfloat16_t矢量量化。QF322BF16_PREfloat量化成floatscalar量化。该量化模式精度无法达到双万分之一可以达到双千分之一。VQF322BF16_PREfloat量化成float矢量量化。该量化模式精度无法达到双万分之一可以达到双千分之一。quant_post可选量化模式分别为NoConv不使能量化功能。QS162B8_POSTint16_t量化成bfloat8_tscalar量化。VQS162B8_POSTint16_t量化成bfloat8_t矢量量化。QF162B8_POSThalf量化成bfloat8_tscalar量化。VQF162B8_POSThalf量化成bfloat8_t矢量量化。QS162S4_POSTint16_t量化成int4_tscalar量化。VQS162S4_POSTint16_t量化成int4_t矢量量化。QF162S4_POSThalf量化成int4_t类型scalar量化。VQF162S4_POSThalf量化成int4_t类型矢量量化。QS162S16_POSTint16_t量化成int16_tscalar量化。VQS162S16_POSTint16_t量化成int16_t矢量量化。QF162S16_POSThalf量化成int16_tscalar量化。VQF162S16_POSThalf量化成int16_t矢量量化。函数原型常规搬运__aicore__ inline void asc_copy_l0c2ub(__ubuf__ bfloat16_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ half *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ hifloat8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ uint8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ bfloat16_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ half *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ hifloat8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ uint8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int32_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ void *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ void *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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)同步搬运__aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ bfloat16_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ half* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ hifloat8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ uint8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ bfloat16_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ half* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ hifloat8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ uint8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int32_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ void* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ void* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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)参数说明参数名输入/输出描述dst_addr输出目的操作数矢量的起始地址。src_addr输入源操作数矢量的起始地址。n_size输入源NZ矩阵在N方向上的大小。• 不使能NZ2ND功能取值范围[1, 4095]• 使能NZ2ND功能取值范围[1, 4095]。m_size输入源NZ矩阵在M方向上的大小。• 不使能NZ2ND功能取值范围[1, 65535]• 使能NZ2ND功能取值范围[1, 8192]。loop_dst_stride输入目的相邻ND矩阵起始地址之间的偏移。loop_src_stride输入源NZ矩阵中相邻Z排布的起始地址偏移取值范围[0, 65535]单位C0_Size16*sizeof(T)T为src_addr的数据类型。dual_dst_ctl输入双目的控制参数。sub_blockid输入子块ID。clip_relu_pre输入预处理阶段使能clip_relu需搭配normal relu归一化的relu函数一起使用且需要使能量化功能。unit_flag_ctl输入与unit_flag参数相关取值如下•0 保留值•2 使能unit_flag硬件执行完指令之后不会设置寄存器•3 使能unit_flag硬件执行完指令后会将unit_flag关闭。quant_pre输入预处理阶段量化参数。取值见功能说明。relu_pre输入预处理阶段使能relu。split_en输入是否使能通道拆分的功能默认false不使能该功能。仅在src_addr和dst_addr都为float时才能使能通道拆分且不能同时使能split_en和NZ2ND功能。NZ2ND_en输入使能NZ2ND开关。•false不使能•true使能。quant_post输入后处理阶段量化参数。取值见功能说明。relu_post输入后处理阶段使能relu。clip_relu_post输入后处理阶段使能clip_relu需搭配normal relu一起使用且需要使能量化功能。eltwise_op输入定义数据从l0c搬运至ub时的目的操作数地址和通道步长。eltwise_antq_en输入按位使能元素的反量化操作。C0_pad_en输入使能为C0配置填充位C0是通道循环的目标步长。broadcast_en输入是否使能广播能力。•false不使能•true使能在数据搬运时沿M轴方向进行数据广播。NZ2DN_en输入使能NZ2DN开关。•false不使能•true使能。返回值说明无流水类型PIPE_FIX约束说明src_addr的起始地址要求按照对应数据类型所占字节数对齐。dst_addr的起始地址要求32字节对齐。调用示例__ubuf__ bfloat16_t dst[256]; __cc__ float src[256]; uint16_t n_size 1; uint16_t m_size 1; uint32_t loop_dst_stride 0; uint16_t loop_src_stride 0; uint8_t dual_dst_ctl 5; bool sub_blockid true; uint8_t clip_relu_pre 0; uint8_t unit_flag_ctl 0; uint64_t quant_pre DEQF16; uint8_t relu_pre 0; bool split_en true; bool NZ2ND_en true; uint64_t quant_post VQS162B8_POST; uint8_t relu_post 0; bool clip_relu_post true; uint8_t eltwise_op 0; bool eltwise_antq_en true; bool C0_pad_en true; bool broadcast_en false; bool NZ2DN_en false; asc_copy_l0c2ub(dst, src, n_size, m_size, loop_dst_stride, loop_src_stride, dual_dst_ctl, sub_blockid, clip_relu_pre, unit_flag_ctl, quant_pre, relu_pre, split_en, NZ2ND_en, quant_post, relu_post, clip_relu_post, eltwise_op, eltwise_antq_en, C0_pad_en, broadcast_en, NZ2DN_en);【免费下载链接】asc-devkit本项目是CANN 推出的昇腾AI处理器专用的算子程序开发语言原生支持C和C标准规范主要由类库和语言扩展层构成提供多层级API满足多维场景算子开发诉求。项目地址: https://gitcode.com/cann/asc-devkit创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
CANN asc-devkit L0C到UB数据搬运API
asc_copy_l0c2ub【免费下载链接】asc-devkit本项目是CANN 推出的昇腾AI处理器专用的算子程序开发语言原生支持C和C标准规范主要由类库和语言扩展层构成提供多层级API满足多维场景算子开发诉求。项目地址: https://gitcode.com/cann/asc-devkit产品支持情况产品是否支持Ascend 950PR/Ascend 950DT√功能说明矩阵计算完成后对结果进行量化处理之后将处理结果搬运到Unified Buffer中。量化参数共有2个quant_pre和quant_post分别对应预处理和后处理阶段。quant_pre可选量化模式分别为NoQuant不使能量化功能。F322BF16float量化成bfloat16_t。量化结果不支持INF_NAN模式。F322F16float量化成half。量化结果支持INF_NAN模式。DEQF16int32_t量化成half。量化结果不支持INF_NAN模式。VDEQF16int32_t量化成half。量化结果不支持INF_NAN模式。QF322B8_PREfloat量化成uint8_t/int8_t。scalar量化。VQF322B8_PREfloat量化成uint8_t/int8_t。scalar量化。REQ8int32_t量化成uint8_t/int8_t。scalar量化。VREQ8int32_t量化成uint8_t/int8_t。矢量量化。QF322FP8_PREfloat量化成fp8_e4m3fn_tscalar量化。VQF322FP8_PREfloat量化成fp8_e4m3fn_t矢量量化。QF322HIF8_PREfloat量化成hifloat8_tHalf to Away Roundscalar量化。VQF322HIF8_PREfloat量化成hifloat8_tHalf to Away Round矢量量化。QF322HIF8_PRE_HYBRIDfloat量化成hifloat8_tHybrid Roundscalar量化。VQF322HIF8_PRE_HYBRIDfloat量化成hifloat8_tHybrid Round矢量量化。QS322BF16_PREint32_t量化成bfloat16_tscalar量化。VQS322BF16_PREint32_t量化成bfloat16_t矢量量化。QF322F16_PREfloat量化成halfscalar量化。VQF322F16_PREfloat量化成half矢量量化。QF322BF16_PREfloat量化成bfloat16_tscalar量化。VQF322BF16_PREfloat量化成bfloat16_t矢量量化。QF322BF16_PREfloat量化成floatscalar量化。该量化模式精度无法达到双万分之一可以达到双千分之一。VQF322BF16_PREfloat量化成float矢量量化。该量化模式精度无法达到双万分之一可以达到双千分之一。quant_post可选量化模式分别为NoConv不使能量化功能。QS162B8_POSTint16_t量化成bfloat8_tscalar量化。VQS162B8_POSTint16_t量化成bfloat8_t矢量量化。QF162B8_POSThalf量化成bfloat8_tscalar量化。VQF162B8_POSThalf量化成bfloat8_t矢量量化。QS162S4_POSTint16_t量化成int4_tscalar量化。VQS162S4_POSTint16_t量化成int4_t矢量量化。QF162S4_POSThalf量化成int4_t类型scalar量化。VQF162S4_POSThalf量化成int4_t类型矢量量化。QS162S16_POSTint16_t量化成int16_tscalar量化。VQS162S16_POSTint16_t量化成int16_t矢量量化。QF162S16_POSThalf量化成int16_tscalar量化。VQF162S16_POSThalf量化成int16_t矢量量化。函数原型常规搬运__aicore__ inline void asc_copy_l0c2ub(__ubuf__ bfloat16_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ half *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e4m3fn_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ fp8_e5m2_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ hifloat8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ uint8_t *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ float *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ bfloat16_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ half *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ hifloat8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ uint8_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ int32_t *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ void *dst_addr, __cc__ float *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub(__ubuf__ void *dst_addr, __cc__ int32_t *src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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)同步搬运__aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ bfloat16_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ half* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ hifloat8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ uint8_t* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ float* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ bfloat16_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ half* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __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, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ hifloat8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ uint8_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ int32_t* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ void* dst_addr, __cc__ float* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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) __aicore__ inline void asc_copy_l0c2ub_sync(__ubuf__ void* dst_addr, __cc__ int32_t* src_addr, uint16_t n_size, uint16_t m_size, uint32_t loop_dst_stride, uint16_t loop_src_stride, uint8_t dual_dst_ctl, bool sub_blockid, uint8_t clip_relu_pre, uint8_t unit_flag_ctl, uint64_t quant_pre, uint8_t relu_pre, bool split_en, 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)参数说明参数名输入/输出描述dst_addr输出目的操作数矢量的起始地址。src_addr输入源操作数矢量的起始地址。n_size输入源NZ矩阵在N方向上的大小。• 不使能NZ2ND功能取值范围[1, 4095]• 使能NZ2ND功能取值范围[1, 4095]。m_size输入源NZ矩阵在M方向上的大小。• 不使能NZ2ND功能取值范围[1, 65535]• 使能NZ2ND功能取值范围[1, 8192]。loop_dst_stride输入目的相邻ND矩阵起始地址之间的偏移。loop_src_stride输入源NZ矩阵中相邻Z排布的起始地址偏移取值范围[0, 65535]单位C0_Size16*sizeof(T)T为src_addr的数据类型。dual_dst_ctl输入双目的控制参数。sub_blockid输入子块ID。clip_relu_pre输入预处理阶段使能clip_relu需搭配normal relu归一化的relu函数一起使用且需要使能量化功能。unit_flag_ctl输入与unit_flag参数相关取值如下•0 保留值•2 使能unit_flag硬件执行完指令之后不会设置寄存器•3 使能unit_flag硬件执行完指令后会将unit_flag关闭。quant_pre输入预处理阶段量化参数。取值见功能说明。relu_pre输入预处理阶段使能relu。split_en输入是否使能通道拆分的功能默认false不使能该功能。仅在src_addr和dst_addr都为float时才能使能通道拆分且不能同时使能split_en和NZ2ND功能。NZ2ND_en输入使能NZ2ND开关。•false不使能•true使能。quant_post输入后处理阶段量化参数。取值见功能说明。relu_post输入后处理阶段使能relu。clip_relu_post输入后处理阶段使能clip_relu需搭配normal relu一起使用且需要使能量化功能。eltwise_op输入定义数据从l0c搬运至ub时的目的操作数地址和通道步长。eltwise_antq_en输入按位使能元素的反量化操作。C0_pad_en输入使能为C0配置填充位C0是通道循环的目标步长。broadcast_en输入是否使能广播能力。•false不使能•true使能在数据搬运时沿M轴方向进行数据广播。NZ2DN_en输入使能NZ2DN开关。•false不使能•true使能。返回值说明无流水类型PIPE_FIX约束说明src_addr的起始地址要求按照对应数据类型所占字节数对齐。dst_addr的起始地址要求32字节对齐。调用示例__ubuf__ bfloat16_t dst[256]; __cc__ float src[256]; uint16_t n_size 1; uint16_t m_size 1; uint32_t loop_dst_stride 0; uint16_t loop_src_stride 0; uint8_t dual_dst_ctl 5; bool sub_blockid true; uint8_t clip_relu_pre 0; uint8_t unit_flag_ctl 0; uint64_t quant_pre DEQF16; uint8_t relu_pre 0; bool split_en true; bool NZ2ND_en true; uint64_t quant_post VQS162B8_POST; uint8_t relu_post 0; bool clip_relu_post true; uint8_t eltwise_op 0; bool eltwise_antq_en true; bool C0_pad_en true; bool broadcast_en false; bool NZ2DN_en false; asc_copy_l0c2ub(dst, src, n_size, m_size, loop_dst_stride, loop_src_stride, dual_dst_ctl, sub_blockid, clip_relu_pre, unit_flag_ctl, quant_pre, relu_pre, split_en, NZ2ND_en, quant_post, relu_post, clip_relu_post, eltwise_op, eltwise_antq_en, C0_pad_en, broadcast_en, NZ2DN_en);【免费下载链接】asc-devkit本项目是CANN 推出的昇腾AI处理器专用的算子程序开发语言原生支持C和C标准规范主要由类库和语言扩展层构成提供多层级API满足多维场景算子开发诉求。项目地址: https://gitcode.com/cann/asc-devkit创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
