将内存段“EPI_CS2_EXEC_1”添加到工程中。调整“长度”以适应客户的 AI 模型大小。下面提供了 CMD 文件的更改：

MEMORY
{
    RAM_S        : ORIGIN = 0x20000000, LENGTH = 0x18000   // 96KB internal SRAM
    RAM_C        : ORIGIN = 0x00C18000, LENGTH = 0x8000    // 32KB code SRAM
    FLASH_BANK0  : ORIGIN = 0x00000000, LENGTH = 0x40000   // 256KB Flash
    FLASH_BANK1  : ORIGIN = 0x00040000, LENGTH = 0x40000   // 256KB Flash

    // External SDRAM through EPI — AI model code + weights execute from here
    // Customer with 300KB AI model: increase to 0x0080000 (512KB) or 0x0100000 (1MB)
    EPI_CS2_EXEC_1(RWX) : origin = 0x80000000, length = 0x0010000  // 64KB (demo size)
}
```

添加 .CS1_EXECUTE 段并将 `libCMSISDSP_m33_ti_arm_clang.a` 替换为“customer_ai_model.a”文件名。下面提供了 CMD 文件的更改：

.CS1_EXECUTE : LOAD = FLASH_BANK0, palign(8),
               RUN  = EPI_CS2_EXEC_1,
               LOAD_START(emif1_cs2_sec1_loadstart),
               LOAD_SIZE(emif1_cs2_sec1_loadsize),
               RUN_START(emif1_cs2_sec1_runstart),
               RUN_SIZE(emif1_cs2_sec1_runsize)
{
    // Replace with the customer's AI .a filename.
    // The <*> wildcard grabs ALL objects in the archive automatically.
    customer_ai_model.a<*>(.text .text.* .rodata .rodata*)
}

每个关键字的含义在表 2-2 中进行了说明：