Robinus

Thanks for kind advice. It would take some time to comeback as I am working on CUDA device functions at the moment.

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

This is about loading a data into the memory, so it is [reading]. But what I thought was to ensure that the loaded data locates inside the physical memory, not in disk drives by operating system which has the virtual memory management. On CUDA programming, the data transfer between the host memory ( on a motherboard ) and the device memory ( memory on a graphics card ) happens. But if data is unluckily located in disk drive by the determination of an OS, then there will be a great delay at each time...

I wonder if matio offers loading a data ensuring it is allocated in the paged-locked memory so that I can access from CUDA with DMA. If there isn't, I'd like to modify the source code of matio for the method, EXTERN mat_t *Mat_Open(const char *matname,int mode). To this end, I'd like to know, which source code that I need to modify in matio. And I am open to any related suggestions for this purpose. Thanks a lot. cf. the attachment.pdf comes together with beautiful elaborations.