Add direct CDMA matmul path

sdk/software/bsp/include/dma.h

@@ -10,21 +10,46 @@ typedef struct {
     volatile uint32_t TAG;       // 0x0C: 软件 Tag (硬件不处理，留给软件标记任务用)
     volatile uint32_t CTRL;      // 0x10: 控制寄存器 (配置 Burst 属性 + 启动)
     volatile uint32_t STATUS;    // 0x14: 状态寄存器 (Busy, Done)
-    volatile uint32_t RESV[10];  // 0x18 ~ 0x3C: 保留空间，凑齐 0x40 字节
+    volatile uint32_t MODE;      // 0x18: 0=mem2mem, 1=mem2ep, 2=ep2mem
+    volatile uint32_t EP_CFG;    // 0x1C: [3:0]=bank, [15:8]=word offset, [19:16]=endpoint id
+    volatile uint32_t RESV[8];   // 0x20 ~ 0x3C: 保留空间，凑齐 0x40 字节
 } dma_ch_regs_t;
 
 // DMA 控制器基地址
 #define CDMA_MUX_BASE 0xbf300000
 #define DMA_CH(i)     ((volatile dma_ch_regs_t*)(CDMA_MUX_BASE + (i) * 0x40))
 
+typedef struct {
+    volatile uint32_t CTRL;       // 0x200: bit0 start
+    volatile uint32_t STATUS;     // 0x204: bit0 busy, bit1 done (W1C)
+    volatile uint32_t SRC_BASE;   // 0x208
+    volatile uint32_t DST_BASE;   // 0x20C
+    volatile uint32_t GROUPS;     // 0x210
+    volatile uint32_t MAT_BASE;   // 0x214: 保留，兼容旧软件
+    volatile uint32_t SRC_STRIDE; // 0x218
+    volatile uint32_t DST_STRIDE; // 0x21C
+} dma_seq_regs_t;
+
+#define DMA_SEQ ((volatile dma_seq_regs_t*)(CDMA_MUX_BASE + 0x200))
+
 // CTRL 寄存器位定义
 #define CTRL_START_BIT       (1 << 0)
 #define CTRL_BURST_SIZE(x)   (((x) & 0x7) << 3)  // Bits [5:3]: AXI AxSIZE (0=1B, 1=2B, 2=4B, 3=8B)
 #define CTRL_BURST_LEN(x)    (((x) & 0xFF) << 6) // Bits [13:6]: AXI AxLEN (0=1 beat, 15=16 beats)
 
+#define DMA_MODE_MEM2MEM     0
+#define DMA_MODE_MEM2EP      1
+#define DMA_MODE_EP2MEM      2
+#define DMA_EP_CFG(id, bank, word_off) \
+    ((((id) & 0xF) << 16) | (((word_off) & 0xFF) << 8) | ((bank) & 0xF))
+
 // STATUS 寄存器位定义
 #define STATUS_BUSY_BIT      (1 << 0)
 #define STATUS_DONE_BIT      (1 << 1)
 
+#define SEQ_START_BIT        (1 << 0)
+#define SEQ_BUSY_BIT         (1 << 0)
+#define SEQ_DONE_BIT         (1 << 1)
+
 void dma_start_transfer(int ch_id, uint32_t src, uint32_t dst, uint32_t bytes, uint32_t tag);
-void dma_wait_polling(int ch_id);
+void dma_wait_polling(int ch_id);