user5338487 said:

while(1)

{

HWREG(SOC_GPIO_0_REGS+0x14)|=0x00000002;//set gp[0][1]

HWREG(SOC_GPIO_0_REGS+0x14)&=0xfffffffd;// clear gp[0][1]

}

Your code above is performing a read-modify-write to the OUT_DATA01 register.  This is a very slow operation:

1. Reads will stall the processor.
2. Access to the GPIO peripheral requires the data to cross multiple interconnects, clock domains, etc.  It is slow.

The good news is that you can tremendously improve this code by appropriately leveraging the SET and CLEAR registers.  This will convert your read-modify-write operations into write-only operations.  For example:

while(1)

{

HWREG(SOC_GPIO_0_REGS+0x18) = 0x00000002;//set gp[0][1]

HWREG(SOC_GPIO_0_REGS+0x1C) = 0x00000002;// clear gp[0][1]

}

In the case above where you're doing nothing but these register writes, your write buffer will quickly fill up which will still stall the CPU (though for a lot less time than your read-modify-write).  So I expect your performance should double with this change.  And I think if you're doing a more "real world" example where you have other code mixed with this the performance will get even better, because those writes will be non-blocking as long as your write buffer doesn't fill up completely.  So if you're just doing one or two register writes here and there the performance improvement will be dramatic.

Best regards,
Brad