aclprofAicoreMetrics

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
typedef enum {
    ACL_AICORE_ARITHMETIC_UTILIZATION = 0,     // Arithmetic utilization ratio
    ACL_AICORE_PIPE_UTILIZATION = 1,           // Ratio of the time consumed by compute units to that of move units
    ACL_AICORE_MEMORY_BANDWIDTH = 2,           // Ratio of external memory read/write instructions
    ACL_AICORE_L0B_AND_WIDTH = 3,              // Ratio of internal memory read/write instructions
    ACL_AICORE_RESOURCE_CONFLICT_RATIO = 4,    // Ratio of pipeline queue instructions
    ACL_AICORE_MEMORY_UB = 5,                  // Ratio of internal memory read/write instructions
    ACL_AICORE_L2_CACHE = 6,                   // Read/write cache hits and re-allocations upon cache misses
    ACL_AICORE_PIPE_EXECUTE_UTILIZATION = 7,   // Ratio of the time consumed by compute units to that of move units
    ACL_AICORE_MEMORY_ACCESS = 8,              // Bandwidth of the operator's memory access on cores
    ACL_AICORE_NONE = 0xFF
}aclprofAicoreMetrics;

Atlas 200I/500 A2 inference product: ACL_AICORE_MEMORY_ACCESS is not supported.

Atlas inference product: ACL_AICORE_L2_CACHE, ACL_AICORE_PIPE_EXECUTE_UTILIZATION, and ACL_AICORE_MEMORY_ACCESS are not supported.

Atlas training product: ACL_AICORE_L2_CACHE, ACL_AICORE_PIPE_EXECUTE_UTILIZATION, and ACL_AICORE_MEMORY_ACCESS are not supported.

Atlas A2 training product/Atlas A2 inference product: ACL_AICORE_PIPE_EXECUTE_UTILIZATION is not supported.

Atlas A3 training product/Atlas A3 inference product: ACL_AICORE_PIPE_EXECUTE_UTILIZATION is not supported.

Atlas 350 Accelerator Card: ACL_AICORE_L2_CACHE, ACL_AICORE_PIPE_EXECUTE_UTILIZATION, and ACL_AICORE_MEMORY_ACCESS are not supported.