1920*1080大小输入图像帧拷贝到mpp frame buffer占用CPU高

[Iyiren]

我在使用XDMA从FPGA PCIE把数据放到了xdma.buffer里，xdma.buffer是posix_memalign()分配的内存，当我把192010803大小的数据用memcpy拷贝到mpp buffer(DRM或者DMA_HEAP类型)里，使用时间3---4ms，但是cpu占用率约40%。 我又使用librga的imcopy()，效果差不多，耗时3---4ms，cpu占用率仍然30%~40%。 请问有没有什么方法可以降低CPU使用率呢？有没有免拷贝或者DMA直接拷贝，不经过CPU的方法呢？

[Iyiren]

XDMA结构体：/ helper struct to remember the Xdma device names / typedef struct { char base_path; / path to first found Xdma device / char c2h0_path; / card to host DMA 0 / char h2c0_path; / host to card DMA 0 / char c2h1_path; / card to host DMA 1 / char h2c1_path; / host to card DMA 1 / char user_path; / XDMA AXI Lite interface / char event0_path; / XDMA interrupt0 interface / char buffer_c2h; / pointer to the allocated buffer card to host/ char buffer_h2c; / pointer to the allocated buffer host to card*/

char * buffer_c2h1;   /* pointer to the allocated buffer card to host*/
char * buffer_h2c1;   /* pointer to the allocated buffer host to card*/

void * user_map_base; /* XDMA AXI Lite interface map base address */
u_long buf_c2h_size; /* size of the buffer c2h in bytes */
u_long buf_h2c_size; /* size of the buffer h2c in bytes */
int c2h0;
int h2c0;

int c2h1;
int h2c1;

int user;
int event0;

} xdma_device; 内存分配：xdma.buf_c2h_size = 16* ONE_MB; // buffer card to host size posix_memalign((void **)&xdma.buffer_c2h, 1024, xdma.buf_c2h_size); //16 | 1024 读取数据：read(xdma.c2h0, xdma.buffer_c2h, bytes);

[HermanChen]

profile 看一下 cpu 占用是占用在哪里比较好