usb_notes.txt

//ESP8266 instruction timing guide:
//
// nop : 1
// add, addi, or, extui : 2 (Sort of)
// movi: 1
// l32r: 7 (Sort of)
// l32i.n: 1 (Seems to be always)
//
// 1 Clock cycle:
//		add.n	a10, a10, a11
//
// 2 Clock Cycles:
//		add.n	a10, a10, a11
//		add.n	a10, a10, a11
//
// 2 Clock Cycles:
//		add.n	a10, a10, a11
//		extui	a10, a10, 24, 3
//
// 7 Clocks:
//		l32r	a10, 4010027c <_DoubleExceptionVector+0x20c>
//
// Also 7 Clocks:
//		addi.n	a10, a10, -1
//	    l32r	a10, 4010027c <_DoubleExceptionVector+0x20c>
//
// 9 Clocks:
//		addi.n	a10, a10, -1
//		addi.n	a11, a11, -1
//		l32r	a11, 4010027c <_DoubleExceptionVector+0x20c>
//
// 14 Clocks
//		l32r	a10, 4010027c <_DoubleExceptionVector+0x20c>
//		l32r	a11, 40100280 <_DoubleExceptionVector+0x210>
//
// 1 Clocks
//     	l32i.n	a10, a2, 0
//
// 2 Clocks 
//		add.n	a10, a10, a11
//		l32i.n	a10, a2, 0
//
// 3 Clocks (Second operation depends on first)
//		l32i.n	a10, a2, 0
//        (Nop or no nop, still 3 cycles)
//		add.n	a10, a10, a11
//
//
//	2 Clocks  (Second operation does not depend on first)
//		l32i.n	a10, a2, 0
//		add.n	a11, a11, a11
//
//  3 Clocks (@a2 == 0)  (Branch not taken)
//		l32i.n	a10, a2, 0
//		nop.n
//		bgeui	a10, 5, 40100b60 <end>
//
//  5 Clocks (@a2 == 6)  (Branch taken)  Branching takes 2 additional instructions
//		l32i.n	a10, a2, 0
//		nop.n
//		bgeui	a10, 5, 40100b60 <end>
//
//
//  Big note: with regular jumps and comparative jumps, if the target's address ends with  11 (binary), then it will take one extra cycle.
//
//
//  3 Clocks
//		j	end
//		...
//		end:
//
//  4 Clocks: (Branch not taken)
//		l32i	a10, a2, 0
//		bgeui	a10, 6, 40100b58 <end>
//		addi	a10, a10, 1
//
//	4 Clocks: (Branch not taken)
//		l32i	a10, a2, 0
//		extui	a10, a10, 23, 1
//	   	bgeui	a10, 6, 40100b58 <end>
//
//  6 Clocks: (Branch Taken)
//		l32i	a10, a2, 0
//			(Free to do unrelated instruction here)
//		extui	a10, a10, 1, 4
//		bgeui	a10, 6, 40100b58 <end>
//
//
// Function Calls:
//
//
//  5 Clocks: No function call.
//		nop.n
//		nop.n
//		j	40100b58 <end>
//	my_func:
//		ret.n
//	end:
//
// 10 Clocks: Include function call
//		call0	40100b54 <my_func>
//		j	40100b56 <end>
//	<my_func>:
//      ret.n
//  end:
//
// 10 Clocks: Include function call, and an instruction in the beginning
//        	call0	40100b54 <my_func>
//        	j	40100b58 <end>
// <my_func>:
//      	nop.n
//      	ret.n
// <end>:
//
// 12 Clocks:   (WHYYYYYYY WHY Jump from 10 to 12??)
// <intrs>:
//        	call0	40100b54 <my_func>
//        	j	40100b5a <end>
// <my_func>:
//      	nop.n
//      	nop.n
//      	ret.n
//
// 13 Clocks:
//<intrs>:
//        	call0	40100b54 <my_func>
//        	j	40100b5e <end>
// <my_func>:
//        	addi	a1, a1, -16
//      	nop.n
//        	addi	a1, a1, 16
//      	ret.n
//
// 15 Clocks:
// <intrs>:
//		call0	40100b54 <my_func>
//        	j	40100b62 <end>
// <my_func>:
//        	addi	a1, a1, -16
//      	s32i.n	a0, a1, 0
//      	nop.n
//      	l32i.n	a0, a1, 0
//        	addi	a1, a1, 16
//      	ret.n
// <end>:
//
//
// 21 Clocks:
// <intrs>:
//        	call0	40100b54 <my_func>
//        	j	40100b6e <end>
// <my_func>:
//        	addi	a1, a1, -16
//      	s32i.n	a0, a1, 0
//      	s32i.n	a2, a1, 4
//      	s32i.n	a3, a1, 8
//      	s32i.n	a4, a1, 12
//      	nop.n
//      	l32i.n	a4, a1, 12
//      	l32i.n	a3, a1, 8
//      	l32i.n	a2, a1, 4
//      	l32i.n	a0, a1, 0
//        	addi	a1, a1, 16
//      	ret.n
// <end>:
//
// For the following, we will be replacing 'nop' with the following:
//
// This takes 1 clock...
//      nop
//
// These take 20 Clocks... WHYYYY?Y??YY?
//		l32r	a2, 40100280 <_DoubleExceptionVector+0x210> (These are actually movi a2, (some constant))
//	   	l32r	a3, 40100270 <_DoubleExceptionVector+0x200>  
//     	l32r	a4, 40100284 <_DoubleExceptionVector+0x214>
//
// 14 Clocks:
//        	l32r	a2, 40100280 <_DoubleExceptionVector+0x210>
//        	l32r	a3, 40100270 <_DoubleExceptionVector+0x200>
//
// 7 Clocks:
//        	l32r	a3, 40100270 <_DoubleExceptionVector+0x200>
//
// These take 25 Clocks... Ok... this is getting weird.
//		l32r	a0, 40100280 <_DoubleExceptionVector+0x210> (These are actually movi a2, (some constant))
//		l32r	a2, 40100280 <_DoubleExceptionVector+0x210> 
//	   	l32r	a3, 40100270 <_DoubleExceptionVector+0x200>  
//     	l32r	a4, 40100284 <_DoubleExceptionVector+0x214>
//
// 33 Clock... WAT??
//        	l32r	a0, 40100280 <_DoubleExceptionVector+0x210>
//        	l32r	a2, 40100280 <_DoubleExceptionVector+0x210>
//        	l32r	a4, 40100280 <_DoubleExceptionVector+0x210>
//        	l32r	a0, 40100280 <_DoubleExceptionVector+0x210>
//        	l32r	a2, 40100280 <_DoubleExceptionVector+0x210>
//
// 9 Clocks:  (If you want to cleaverly pull data out of a table)
//        	l32r	a0, 40100160 <_DoubleExceptionVector+0xf0>
//			(ONE Optional nop here)
//      	sskipper32i.n	a1, a0, 0
//
//
//  BIG NOTE About jumping.  If the instruction you are jumping to ends in a 11 as the last two bits, it will take ONE extra cycle to complete.
//     JUMPS TAKE ONE EXTRA CYCLE WHEN 11 is last two bits: CONFIRMED.  This lets us place 3 instructions in and it will still take just as long!
//
// Whenever reading with any kind of load, the next instruction CANNOT use the data that was loaded.

//
//  mul16u a, b, c  <<1 cycle, always.
//
//


/*
	this is 5 cycles... Multiplies take two cycles to complete, but only one to issue.
	mul16u a7, a5, a6
	bbsi a7, 0, skipper
		_nop.n
		_nop
	skipper:

	...ALSO 5

	mul16u a7, a5, a6
	nop
	bbsi a7, 0, skipper
		_nop.n
		_nop
	skipper:

	... ALSO 5
	mul16u a7, a5, a6
	mul16u a5, a5, a6
	bbsi a7, 0, skipper
		_nop.n
		_nop.n
	skipper:

	...BUT if we use one multiply in another multiply it stalls. (6 cycles)
	mul16u a7, a5, a6
	mul16u a7, a5, a7
	bbsi a7, 0, skipper
		_nop.n
		_nop.n
	skipper:
*/


/*
	Calling...
	
.align 4
my_test_call:
	ret

elsewhere..
	_call0 my_test_call

	//IF the last two bits of the address the CALL line is on are:
	//  00 = 7 cycles
	//  01 = 6 cycles
	//	10 = 6 cycles
	//  11 = 7 cycles

*/


/*
	Revisiting l32
	//A4 = some place within iram, aligned.  Unaligned reads will crash regardless of if they're from iram or dram.


	//7, 8 cycles (Or +1 if there is a dependency)
	//
	// If l32r's address ends in:
	//		00: 7 cycles
	//		01: 7 cycles
	//		10: 8 cycles
	//		11: 8 cycles
	_l32r a6, my_test_data
	add a6, a7, a7	

	//l32i behaves exactly the same!!!  Btw! Can only load from an address that ends in 00.


	You can't seem to l8ui from iram.  It seems to always crash.

	//A4 = some place within dram

	//2 Cycles
	_l32i.n a6, a4, 0
	add a7, a7, a7

	//3 Cycles
	_l32i.n a7, a4, 0
	add a7, a7, a7

	
	//2 Cycles  (If a4 is aligned or not, regardless)
	_l8i.n a6, a4, 0
	add a7, a7, a7  (3 cycles if the dependency exists)

	
*/







////////////////////////////////////////////////////


#if 0


//NOTES: We are hooked up to USB by GPIO12: D-? (White), GPIO14: D-, GPIO5 EXTDEBUG

#define STDPLUS "12"
#define STDMINUS "14"

#define GPSET   __asm__ __volatile__ ("movi    a2, 0x60000304\nmovi    a4, 0x60000308\nmovi.n  a3, 32\n");  //GPIO5
#define LOW 	__asm__ __volatile__ ("s32i.n  a3, a4, 0");  //One instruction.
#define HIGH    __asm__ __volatile__ ("s32i.n  a3, a2, 0");

#define LOWS 	__asm__ __volatile__ ("s32i.n  a14, a13, 0":::"a3");
#define HIGHS   __asm__ __volatile__ ("s32i.n  a14, a12, 0":::"a3");

//Detailed analysis of some useful stuff and performance tweaking: http://naberius.de/2015/05/14/esp8266-gpio-output-performance/
//Reverse engineerd boot room can be helpful, too: http://cholla.mmto.org/esp8266/bootrom/boot.txt

//The folowing performs this code:
//   uint32_t gpio_status = GPIO_REG_READ(GPIO_STATUS_ADDRESS);
//   GPIO_REG_WRITE(GPIO_STATUS_W1TC_ADDRESS, gpio_status);      //clear interrupt status
//
//GPIO Base Register(PERIPHS_GPIO_BASEADDR): 0x60000300 + GPIO Status Register: 0x1c = 0x6000031c
//GPIO_STATUS_W1TC_ADDRESS  = 0x24
#define ACKNOWLEDGE_INTERRUPT __asm__ __volatile__ ("movi    a2, 0x6000031c \n l32i.n a3, a2, 0 \n movi a2, 0x60000324 \n s32i.n a3, a2, 0");

//    ETS_INTR_DISABLE(ETS_GPIO_INUM) ETS_GPIO_INUM=4, 
//#define ETS_INTR_ENABLE(inum) \
//    ets_isr_unmask((1<<inum))
//#define ETS_INTR_DISABLE(inum) \
//    ets_isr_mask((1<<inum))

//These still aren't very fast. I am sure more can be done.
#define FAST_ETS_INTR_DISABLE_GPIO _xtos_ints_off(1<<4);
#define FAST_ETS_INTR_ENABLE_GPIO _xtos_ints_on(1<<4);

/* Charles' useful assembly guide for the ESP:


* don't jump if bit in register set: bbci	a2, 14, 40100379 <gpio_intr+0x11>
*       jump if bit in register set: bbsi	a2, 14, 40100379 <gpio_intr+0x11>

* movi    a2, 0x60000304 <  move the address of 
* movi.n  a3, 4096 < set the value
* s32i.n  a3, a4, 0 < Update it in memory
*/

//XXX Careful: Note the "int" to let things wrap.
static inline int __attribute__((always_inline)) get_ccount(void) 
{
        unsigned r;
        asm volatile ("rsr %0, ccount" : "=r"(r));
        return r;
}

//static inline unsigned __attribute__((always_inline)) wait_bit() 
#define wait_bit \
{ \
	bittime+=53; \
  	while( bittime > get_ccount() ); \
}
	
#define WBV (((*pins)&_BV(DPLUS))?1:0)


//Plan of attack:  
//1: Spin at beginning waiting for a bit transition.
//2: Start "wait bit"ting every 53 cycles.


#define TIMEOUT 0xff

void gpio_intr(void)
{
	register unsigned to asm ("a9"); //timeout
	register int bittime asm ("a10");
	volatile register uint32_t * pins asm ("a11");

	//Variables used for debugging.
	volatile register uint32_t * pin_out_set asm ("a12");
	volatile register uint32_t * pin_out_clr asm ("a13");
	volatile register int psel asm("a14");

	uint32_t bits[16];
	unsigned i, j;

//Initialize common-use variables
__asm__ __volatile__("\n\
	movi a11, "STR_PIN_IN"\n\
	movi a12, "STR_PIN_OUT_SET"\n\
	movi a13, "STR_PIN_OUT_CLEAR"\n\
	movi a14, %0\n\
" : : "M"(1<<DEBUGPIN) : "a11", "a12", "a13", "a14" );
	FAST_ETS_INTR_DISABLE_GPIO;
	HIGHS
//#define SLOW_START

#ifdef SLOW_START
	//Wait for it to go low.
	__asm__ __volatile__("mstart:\n" );

	for( to = TIMEOUT; to; to-- )
	{
		if( !(PIN_IN & _BV(DPLUS)) ) break;
	}
	//Wait for it to go high.
	for( to = TIMEOUT; to; to-- )
	{
		if( (PIN_IN & _BV(DPLUS)) ) break;
	}
	__asm__ __volatile__("mend:\n" );

#else

	//First, we wait for the signal to go (or be) low.  Then we wait for the signal to go high.
	__asm__ __volatile__("\n\
		movi a9, 1023 #timeout \n\
find_low:\n\
        l32i.n a2, a11, 0\n\
		bbci a2, " STDPLUS ", done_low\n\
		addi.n a9, a9, -1\n\
		bnez a9, find_low\n\
		j end_gpio_intr\n\
done_low:\n\
		movi a9, 1023 #timeout \n\
find_high:\n\
        l32i.n a2, a11, 0\n\
		bbsi a2, " STDPLUS ", done_high\n\
		addi.n a9, a9, -1\n\
		bnez a9, find_high\n\
		j end_gpio_intr\n\
done_high:\n\
" : : : "a2" );
#endif
	LOWS


//	bittime = get_ccount() + 42 - 53;  //Each bit is 53 cycles, we are shooting for the middle of the coming bit, but give us back a few us..
										//GCC Seems pretty smart, this actually translates to moving the value into r10 and adding some values.
									//Scratch that.  GCC messes this up, too.
	//This does effectively take a snapshot of the start of the tracking.
	__asm__ __volatile("\n\
		rsr a10, ccount\n\
		addi a10, a10, -12\n\
	");


/*
	//Code for assembly below.
	//Wait for a double 0.
	for( to = TIMEOUT; to; to-- )
	{
		HIGHS
		wait_bit
		LOWS
		if( !(PIN_IN&_BV(DPLUS)) )
		{
			HIGHS
			wait_bit
			LOWS
			if( !(PIN_IN&_BV(DPLUS)) ) break;
		}
	}
*/
	//Wait for a double-low bit.
	//R10 contains when we should stop reading next.
	__asm__ __volatile__("\n\
		movi a9, 1023 #timeout \n\
find_double_low:\n\
        addi.n a9, a9, -1\n\
		beqz a9, end_gpio_intr\n\
		addi a10, a10, 53\n\
\n\
wait_double_loop_int:\n\
		rsr a3, ccount\n\
		bge a10, a3, wait_double_loop_int\n\
		# Time to read another bit.\n\
        l32i.n a3, a11, 0\n\
		memw\n\
		bbsi a3, " STDPLUS ", find_double_low\n\
		# Got a 0!  Let's see if that happens again.\n\
		addi a10, a10, 53\n\
wait_double_loop_int2:\n\
		rsr a3, ccount\n\
		bge a10, a3, wait_double_loop_int2\n\
		# Poll again.\n\
        l32i.n a3, a11, 0\n\
		memw\n\
		bbsi a3, " STDPLUS ", find_double_low\n\
" : : : "a3" );

	//Ok!  We now can use wait_bit to wait for the next bit then receive using WBV.

	for(i = 0; i < 16; i++ )
	{
		HIGHS
		wait_bit
		LOWS
		bits[i] = WBV;
//		GPSET; //Setup GPIO
//		LOW;
//		HIGH;

	}

	HIGHS
/*
	printf( "\n" );
	for(i = 0; i < 16; i++ )
	{
		printf( "%d", bits[i] );
	}
	LOWS
	printf( "\n" );*/


	__asm__ __volatile__("end_gpio_intr:");

	LOWS

	ACKNOWLEDGE_INTERRUPT;

	//ETS_GPIO_INTR_ENABLE();
	FAST_ETS_INTR_ENABLE_GPIO;

	//XT_CLI;
}


#endif