asc_loadalign

产 品 支 持 情 况

产 品	是 否 支 持
Ascend 950PR/Ascend 950DT	√

功 能 说 明

对 齐 数 据 搬 运 接 口，从UB连 续 对 齐 搬 入 目 的 操 作 数，支 持 多 种 搬 入 模 式，接 口 通 过int32_t传 入 偏 移，用 户 可 以 选 择 更 新 偏 移 或 者 更 新 源 操 作 数 的 地 址。

• NORM搬 入 模 式：正 常 模 式，搬 运VL数 据。
• BRC搬 入 模 式：搬 运 一 个b8/b16/b32类 型 的 数 据，并Broadcast到 所 有 元 素 位 置。
• UP2X搬 入 模 式：数 据2倍 上 采 样，加 载VL/2个 数 据，每 个 输 入 元 素 重 复 两 次，数 据 类 型 为b8/b16。
• DOWN2X搬 入 模 式：数 据2倍 下 采 样，加 载2倍VL的 数 据，数 据 每 隔 一 个 保 留，数 据 类 型 为b8/b16。
• UNPACK搬 入 模 式：解 压 缩 模 式，按 无 符 号 整 型u8/u16/u32加 载VL/2长 度 数 据，unpack到VL长 度u16/u32/u64类 型，中 间 位 置 补0。
• UNPACKV2搬 入 模 式：解 压 缩 模 式，按 无 符 号 整 型u8加 载VL/4长 度 数 据，unpack到VL长 度u32类 型，中 间 位 置 补0。
• BRCV2搬 入 模 式：读 取 一 个DataBlock（32B），并 广 播 到VL。
• BRCV3搬 入 模 式：加 载（VL/DataBlock）B的 数 据，并 将 每 个 元 素（16bit/32bit）广 播 到 一 个DataBlock（32B）中。
• DEINTLV搬 入 模 式：双 搬 入 模 式，基 于 元 素 的 交 错 搬 运，从src中 读 取2*VL长 度 数 据，将 偶 数 索 引 的 元 素 存 入dst0，将 奇 数 索 引 的 元 素 存 入dst1，数 据 类 型 为b8/b16/b32。

函 数 原 型

C++

// NORM搬 入 模 式
__simd_callee__ inline void asc_loadalign(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_int32_t& dst, __ubuf__ int32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_uint32_t& dst, __ubuf__ uint32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_float& dst, __ubuf__ float* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_int64_t& dst, __ubuf__ int64_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_uint64_t& dst, __ubuf__ uint64_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// BRC搬 入 模 式
__simd_callee__ inline void asc_loadalign_brc(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_int32_t& dst, __ubuf__ int32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_uint32_t& dst, __ubuf__ uint32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_float& dst, __ubuf__ float* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// UP2X搬 入 模 式
__simd_callee__ inline void asc_loadalign_upsample(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_upsample(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// DOWN2X搬 入 模 式
__simd_callee__ inline void asc_loadalign_downsample(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_downsample(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// UNPACK搬 入 模 式
__simd_callee__ inline void asc_loadalign_unpack(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_int32_t& dst, __ubuf__ int32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_uint32_t& dst, __ubuf__ uint32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_float& dst, __ubuf__ float* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// UNPACKV2搬 入 模 式
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_unpack_v2(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// BRCV2搬 入 模 式
__simd_callee__ inline void asc_loadalign_brc_v2(vector_int8_t& dst, __ubuf__ int8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_uint8_t& dst, __ubuf__ uint8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_fp4x2_e2m1_t& dst, __ubuf__ fp4x2_e2m1_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_fp4x2_e1m2_t& dst, __ubuf__ fp4x2_e1m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_fp8_e8m0_t& dst, __ubuf__ fp8_e8m0_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_fp8_e5m2_t& dst, __ubuf__ fp8_e5m2_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_fp8_e4m3fn_t& dst, __ubuf__ fp8_e4m3fn_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_hifloat8_t& dst, __ubuf__ hifloat8_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_int32_t& dst, __ubuf__ int32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_uint32_t& dst, __ubuf__ uint32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_float& dst, __ubuf__ float* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v2(vector_int4x2_t& dst, __ubuf__ int4b_t* src, int32_t offset)
// BRCV3搬 入 模 式
__simd_callee__ inline void asc_loadalign_brc_v3(vector_int16_t& dst, __ubuf__ int16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_uint16_t& dst, __ubuf__ uint16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_half& dst, __ubuf__ half* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_bfloat16_t& dst, __ubuf__ bfloat16_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_int32_t& dst, __ubuf__ int32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_uint32_t& dst, __ubuf__ uint32_t* src, int32_t offset)
__simd_callee__ inline void asc_loadalign_brc_v3(vector_float& dst, __ubuf__ float* src, int32_t offset)

参 数 说 明

参 数 名	输 入/输 出	描 述
dst	输 出	目 的 操 作 数（矢 量 数 据 寄 存 器）。
dst0	输 出	目 的 操 作 数（矢 量 数 据 寄 存 器）。
dst1	输 出	目 的 操 作 数（矢 量 数 据 寄 存 器）。
src	输 入	源 操 作 数（矢 量）的 起 始 地 址。
offset	输 入	地 址 偏 移 量。

矢 量 数 据 寄 存 器 的 详 细 说 明 请 参 见reg数 据 类 型 定 义.md。

返 回 值 说 明

无

流 水 类 型

PIPE_V

约 束 说 明

无

调 用 示 例

C++

vector_half dst;
__ubuf__ half* src;
int32_t offset;
asc_loadalign(dst, src, offset);