#include <nitro/mi.h>void MI_SendGXCommand( u32 dmaNo, const void* src, u32 commandLength );
void MI_SendGXCommandAsync( u32 dmaNo, const void* src, u32 commandLength,
MIDmaCallback callback, void* arg );
void MI_SendGXCommandAsyncFast( u32 dmaNo, const void* src, u32 commandLength,
MIDmaCallback callback, void* arg );| dmaNo | DMA channel used. |
| src | The transfer source address. |
| commandLength | Transfer command size |
| callback | Callback when DMA ends. |
| arg | Callback argument when DMA ends. |
None.
This function uses DMA to send a display list (GX command string) to command FIFO. The transfer source address must be 4-byte aligned.
The synchronous version of MI_SendGXCommand() waits for DMA to end inside the function. The asynchronous version of MI_SendGXCommandAsync() and MI_SendGXCommandAsFast() call callback when DMA ends. The callback is an MIDmaCallback type function (a void–type function that takes one void* argument). callback is called from the system DMA interrupt handler and therefore is called while interrupts are prohibited.
The difference between MI_SendGXCommandAsync() and MI_SendGXCommandAsyncFast():MI_SendGXCommandAsync() does not use the Geometry transfer DMA but uses only the usual synchronous DMA. A small amount of DMA is performed by using the interrupt that is generated when GXFIFO is less than half. MI_SendGXCommandAsyncFast() uses the Geometry transfer DMA, so that DMA starts automatically when GXFIFO becomes low. Note that multiple automatic DMAs cannot be started simultaneously because of hardware restrictions. However, the hardware allows the simultaneous startup of multipl automatic DMAs of the same type.
With TWL-SDK, exclusive processing is not performed with the MI_SendGXCommandAsync* functions and other G3_-based APIs because the execution speed of the graphics command has been prioritized. Therefore, we cannot guarantee correct operation if the GX command is written directly to command FIFO by a G3_ function while the GX command is being transferred by MI_SendGXCommandAsync*(). It is strongly recommended that you pay particular attention in avoiding such a situation on the API call side.
Although this function uses the old DMA, the function that performs the same operation using the new DMA is MI_SendNDmaGXCommand*. The MI_SendNDmaGXCommand* and MI_SendGXCommand* functions cannot be used at the same time.
See the section titled "DMA Controller Bugs" in the reference for the MI_DmaCopy function.
Uses the following IO registers: DMAn source register (0x40000B0 + 12n), DMAn destination register (0x40000B4 + 12n), and DMAn control register (0x40000B8 + 12n). (n is the DMA channel used.))The GXFIFO(0x4000400) IO register is also accessed.
MI_DmaCopy*
G3_BeginMakeDL
G3_EndMakeDL
MI_SendNDmaGXCommand*
2009/11/20 Added a description of bugs for the DMA controller.
2008/10/16 Added information about functions that use the new DMA.
2005/03/08 Standardized the notation used for the term "interrupt" in Japanese.
2004/12/22 Added a description about calling the callback.
2004/08/26 Added a description of MI_SendGXCommandAsyncFast.
2004/07/18 Added a description of operations when the callback is NULL.
2003/12/01 Initial version.
CONFIDENTIAL