User Profile

Why is using control type transfer on the MAC m2 architecture slower than other platforms? The following is my test code, which was tested using the same USB device in Mac m2, Mac x64, and Ubuntu x64 environments. #include <stdio.h> #include <string.h> #include <stdlib.h> #include <libusb.h> #include <sys/time.h> uint64_t get_us() { uint64_t time = 0; struct timeval tv; gettimeofday(&tv, NULL); time = tv.tv_sec * 1000 * 1000 + tv.tv_usec; return time; } int main() { // open usb handle int ret; ret = libusb_init(NULL); libusb_device_handle *usb_handle; usb_handle = libusb_open_device_with_vid_pid(NULL, 0x248a, 0x8266); if (usb_handle == NULL) { exit(1); } ret = libusb_set_auto_detach_kernel_driver(usb_handle, 1); ret = libusb_claim_interface(usb_handle, 0); uint64_t start = get_us(); unsigned char buf[256]; ret = libusb_control_transfer(usb_handle, 0xC1, 0x02, 0x8014, 0, buf, 12, 1000); ret = libusb_control_transfer(usb_handle, 0xC1, 0x02, 0x8014, 0, buf, 12, 1000); ret = libusb_control_transfer(usb_handle, 0xC1, 0x02, 0x8014, 0, buf, 12, 1000); ret = libusb_control_transfer(usb_handle, 0xC1, 0x02, 0x8014, 0, buf, 12, 1000); memset(buf, 0, 256); ret = libusb_control_transfer(usb_handle, 0x41, 0x02, 0x9548, 0, buf, 16, 1000); ret = libusb_control_transfer(usb_handle, 0x41, 0x02, 0x9548, 0, buf, 16, 1000); ret = libusb_control_transfer(usb_handle, 0x41, 0x02, 0x9548, 0, buf, 16, 1000); ret = libusb_control_transfer(usb_handle, 0x41, 0x02, 0x9548, 0, buf, 16, 1000); uint64_t end = get_us(); printf("spent time %ld(us)\n", end - start); libusb_close(usb_handle); libusb_exit(NULL); return 0; } Finally, my conclusion is to call libusb_control_transfer() function on the MAC m2 architecture takes about 10 times slower than the average in x64 architecture. The following are the software running records and wireshark software packet capture records on the MAC m2 architecture. MacBook-Pro % ./libusb_test spent time 22949(us) The following are the running records and wireshark software packet capture records on Ubuntu20 X64. ubuntu20:~$ ./libusb_test spent time 1246(us) From the data captured by Wireshark software, it can be seen that on the MAC m2 architecture, a control transmission takes about 3ms from submission to reply. On the x64 architecture, only around 200us is required. I also tested bulk type transmission, and the performance on the MAC m2 architecture is consistent with that on the x64 architecture. Is this the reason for the MAC m2 architecture USB driver? Is there a solution to this problem?

I am writing a C program on Mac that calls the libUSB API. However, I found some performance differences between the same programs on MAC x64 and MAC M2 systems. Program on Mac m2 call libusb_control_transfer() function is much slower than programs on Mac x64. For example, running a program on Mac x64 takes 3 seconds to complete a function, while on Mac m2 it takes 10 seconds. I try to count the number of times a program calls libusb_control_transfer() and the time it takes each time, and finally calculate the average time it takes to call libusb_control_transfer(). The following figure shows the performance of the program on Mac m2.(Unit is us per time) The following figure shows the performance of the program on Mac x64. (Unit is us per time) The difference in average time spent calling a function has reached over 10 times. Is this performance gap caused by the USB driver or the libUSB library? Thanks.