asc_ldcg
产品支持情况
功能说明
从L2 Cache加载缓存的数据,如果缓存命中,则直接返回数据。若未命中,则从Global Memory地址预加载数据缓存至L2 Cache,并返回数据。
函数原型
__simt_callee__ inline long int asc_ldcg(__gm__ long int* address)
__simt_callee__ inline unsigned long int asc_ldcg(__gm__ unsigned long int* address)
__simt_callee__ inline long long int asc_ldcg(__gm__ long long int* address)
__simt_callee__ inline unsigned long long int asc_ldcg(__gm__ unsigned long long int* address)
__simt_callee__ inline long2 asc_ldcg(__gm__ long2* address)
__simt_callee__ inline ulong2 asc_ldcg(__gm__ ulong2* address)
__simt_callee__ inline long4 asc_ldcg(__gm__ long4* address)
__simt_callee__ inline ulong4 asc_ldcg(__gm__ ulong4* address)
__simt_callee__ inline longlong2 asc_ldcg(__gm__ longlong2* address)
__simt_callee__ inline ulonglong2 asc_ldcg(__gm__ ulonglong2* address)
__simt_callee__ inline longlong4 asc_ldcg(__gm__ longlong4* address)
__simt_callee__ inline ulonglong4 asc_ldcg(__gm__ ulonglong4* address)
__simt_callee__ inline signed char asc_ldcg(__gm__ signed char* address)
__simt_callee__ inline unsigned char asc_ldcg(__gm__ unsigned char* address)
__simt_callee__ inline char2 asc_ldcg(__gm__ char2* address)
__simt_callee__ inline uchar2 asc_ldcg(__gm__ uchar2* address)
__simt_callee__ inline char4 asc_ldcg(__gm__ char4* address)
__simt_callee__ inline uchar4 asc_ldcg(__gm__ uchar4* address)
__simt_callee__ inline short asc_ldcg(__gm__ short* address)
__simt_callee__ inline unsigned short asc_ldcg(__gm__ unsigned short* address)
__simt_callee__ inline short2 asc_ldcg(__gm__ short2* address)
__simt_callee__ inline ushort2 asc_ldcg(__gm__ ushort2* address)
__simt_callee__ inline short4 asc_ldcg(__gm__ short4* address)
__simt_callee__ inline ushort4 asc_ldcg(__gm__ ushort4* address)
__simt_callee__ inline int asc_ldcg(__gm__ int* address)
__simt_callee__ inline unsigned int asc_ldcg(__gm__ unsigned int* address)
__simt_callee__ inline int2 asc_ldcg(__gm__ int2* address)
__simt_callee__ inline uint2 asc_ldcg(__gm__ uint2* address)
__simt_callee__ inline int4 asc_ldcg(__gm__ int4* address)
__simt_callee__ inline uint4 asc_ldcg(__gm__ uint4* address)
__simt_callee__ inline float asc_ldcg(__gm__ float* address)
__simt_callee__ inline float2 asc_ldcg(__gm__ float2* address)
__simt_callee__ inline float4 asc_ldcg(__gm__ float4* address)
__simt_callee__ inline bfloat16_t asc_ldcg(__gm__ bfloat16_t* address)
__simt_callee__ inline bfloat16x2_t asc_ldcg(__gm__ bfloat16x2_t* address)
__simt_callee__ inline half asc_ldcg(__gm__ half* address)
__simt_callee__ inline half2 asc_ldcg(__gm__ half2* address)
参数说明
表 1 参数说明
返回值说明
L2 Cache中的缓存数据或输入的Global Memory地址上的数据。
约束说明
无
需要包含的头文件
使用除half、half2、bfloat16_t、bfloat16x2_t类型之外的接口需要包含"simt_api/device_functions.h"头文件,使用half和half2类型接口需要包含"simt_api/asc_fp16.h"头文件,使用bfloat16_t和bfloat16x2_t类型接口需要包含"simt_api/asc_bf16.h"头文件。
#include "simt_api/device_functions.h"
#include "simt_api/asc_fp16.h"
#include "simt_api/asc_bf16.h"
调用示例
__simt_vf__ __launch_bounds__(1024) inline void kernel_asc_ldcg(__gm__ float* dst, __gm__ float* src)
{
int idx = threadIdx.x + blockIdx.x * blockDim.x;
dst[idx] = asc_ldcg(src + idx);
}