This document provides details about the instructions used by ESP32 ULP coprocessor assembler.
ULP coprocessor has 4 16-bit general purpose registers, labeled R0, R1, R2, R3. It also has an 8-bit counter register (stage_cnt) which can be used to implement loops. Stage count regiter is accessed using special instructions.
ULP coprocessor can access 8k bytes of RTC_SLOW_MEM memory region. Memory is addressed in 32-bit word units. It can also access peripheral registers in RTC_CNTL, RTC_IO, and SENS peripherals.
All instructions are 32-bit. Jump instructions, ALU instructions, peripheral register and memory access instructions are executed in 1 cycle. Instructions which work with peripherals (TSENS, ADC, I2C) take variable number of cycles, depending on peripheral operation.
The instruction syntax is case insensitive. Upper and lower case letters can be used and intermixed arbitrarily. This is true both for register names and instruction names.
Note about addressing
---------------------
ESP32 ULP coprocessor's JUMP, ST, LD instructions which take register as an argument (jump address, store/load base address) expect the argument to be expressed in 32-bit words.
Consider the following example program::
entry:
NOP
NOP
NOP
NOP
loop:
MOVE R1, loop
JUMP R1
When this program is assembled and linked, address of label ``loop`` will be equal to 16 (expressed in bytes). However `JUMP` instruction expects the address stored in register to be expressed in 32-bit words. To account for this common use case, assembler will convert the address of label `loop` from bytes to words, when generating ``MOVE`` instruction, so the code generated code will be equivalent to::
0000 NOP
0004 NOP
0008 NOP
000c NOP
0010 MOVE R1, 4
0014 JUMP R1
The other case is when the argument of ``MOVE`` instruction is not a label but a constant. In this case assembler will use the value as is, without any conversion::
.set val, 0x10
MOVE R1, val
In this case, value loaded into R1 will be ``0x10``.
Similar considerations apply to ``LD`` and ``ST`` instructions. Consider the following code::
.global array
array: .long 0
.long 0
.long 0
.long 0
MOVE R1, array
MOVE R2, 0x1234
ST R2, R1, 0 // write value of R2 into the first array element,
// i.e. array[0]
ST R2, R1, 4 // write value of R2 into the second array element
// (4 byte offset), i.e. array[1]
ADD R1, R1, 2 // this increments address by 2 words (8 bytes)
ST R2, R1, 0 // write value of R2 into the third array element,
ULP coprocessor is clocked from RTC_FAST_CLK, which is normally derived from the internal 8MHz oscillator. Applications which need to know exact ULP clock frequency can calibrate it against the main XTAL clock::
#include "soc/rtc.h"
// calibrate 8M/256 clock against XTAL, get 8M/256 clock period
ULP coprocessor needs 2 clock cycle to fetch each instuction (fetching is not pipelined), plus certain number of cycles to execute, depending on the instruction. See description of each instruction for details on the execution time.
Note that when accessing RTC memories and RTC registers, ULP coprocessor has lower priority than the main CPUs. This means that ULP coprocessor execution may be suspended while the main CPUs access same memory region as the ULP.
The instruction subtracts the source register from another source register or subtracts 16-bit signed value from a source register, and stores result to the destination register.
The instruction does logical AND of a source register and another source register or 16-bit signed value and stores result to the destination register.
The instruction does logical shift to left of source register to number of bits from another source register or 16-bit signed value and store result to the destination register.
The instruction does logical shift to right of source register to number of bits from another source register or 16-bit signed value and store result to the destination register.
The instruction move to destination register value from source register or 16-bit signed value.
Note that when a label is used as an immediate, the address of the label will be converted from bytes to words. This is because LD, ST, and JUMP instructions expect the address register value to be expressed in words rather than bytes. To avoid using an extra instruction
**Examples**::
1: MOVE R1, R2 //R1 = R2 >> R3
2: MOVE R1, 0x03 //R1 = R2 >> 0x03
3: .set value1, 0x03 //constant value1=0x03
MOVE R1, value1 //R1 = value1
4: .global label //declaration of label
MOVE R1, label //R1 = address_of(label) / 4
...
label: nop //definition of label
**ST**– Store data to the memory
---------------------------------
**Syntax**
**ST***Rsrc, Rdst, offset*
**Operands**
-*Rsrc*– Register R[0..3], holds the 16-bit value to store
-*Rdst*– Register R[0..3], address of the destination, in 32-bit words
The instruction stores the 16-bit value of Rsrc to the lower half-word of memory with address Rdst+offset. The upper half-word is written with the current program counter (PC), expressed in words, shifted left by 5 bits::
The instruction makes a jump to a relative address if condition is true. Condition is the result of comparison of R0 register value and the threshold value.
**Examples**::
1:pos: JUMPR 16, 20, GE // Jump to address (position + 16 bytes) if value in R0 >= 20
2: // Down counting loop using R0 register
MOVE R0, 16 // load 16 into R0
label: SUB R0, R0, 1 // R0--
NOP // do something
JUMPR label, 1, GE // jump to label if R0 >= 1
**JUMPS**– Jump to a relative address (condition based on stage count)
The instruction makes a jump to a relative address if condition is true. Condition is the result of comparison of count register value and threshold value.
**Examples**::
1:pos: JUMPS 16, 20, EQ // Jump to (position + 16 bytes) if stage_cnt == 20
2: // Up counting loop using stage count register
STAGE_RST // set stage_cnt to 0
label: STAGE_INC 1 // stage_cnt++
NOP // do something
JUMPS label, 16, LT // jump to label if stage_cnt < 16
**STAGE_RST**– Reset stage count register
------------------------------------------
**Syntax**
**STAGE_RST**
**Operands**
No operands
**Description**
The instruction sets the stage count register to 0
The instruction sends an interrupt from ULP to RTC controller.
- If the SoC is in deep sleep mode, and ULP wakeup is enabled, this causes the SoC to wake up.
- If the SoC is not in deep sleep mode, and ULP interrupt bit (RTC_CNTL_ULP_CP_INT_ENA) is set in RTC_CNTL_INT_ENA_REG register, RTC interrupt will be triggered.
Note that before using WAKE instruction, ULP program may needs to wait until RTC controller is ready to wake up the main CPU. This is indicated using RTC_CNTL_RDY_FOR_WAKEUP bit of RTC_CNTL_LOW_POWER_ST_REG register. If WAKE instruction is executed while RTC_CNTL_RDY_FOR_WAKEUP is zero, it has no effect (wake up does not occur).
The instruction selects which of the ``SENS_ULP_CP_SLEEP_CYCx_REG`` (x = 0..4) register values is to be used by the ULP wakeup timer as wakeup period. By default, the value from ``SENS_ULP_CP_SLEEP_CYC0_REG`` is used.
**Examples**::
1: SLEEP 1 // Use period set in SENS_ULP_CP_SLEEP_CYC1_REG
2: .set sleep_reg, 4 // Set constant
SLEEP sleep_reg // Use period set in SENS_ULP_CP_SLEEP_CYC4_REG
``I2C_RD`` instruction reads one byte from I2C slave with index ``Slave_sel``. Slave address (in 7-bit format) has to be set in advance into `SENS_I2C_SLAVE_ADDRx` register field, where ``x == Slave_sel``.
8 bits of read result is stored into `R0` register.
**Examples**::
1: I2C_RD 0x10, 7, 0, 0 // Read byte from sub-address 0x10 of slave with address set in SENS_I2C_SLAVE_ADDR0
``I2C_WR`` instruction writes one byte to I2C slave with index ``Slave_sel``. Slave address (in 7-bit format) has to be set in advance into `SENS_I2C_SLAVE_ADDRx` register field, where ``x == Slave_sel``.
**Examples**::
1: I2C_WR 0x20, 0x33, 7, 0, 1 // Write byte 0x33 to sub-address 0x20 of slave with address set in SENS_I2C_SLAVE_ADDR1.
ULP source files are passed through C preprocessor before the assembler. This allows certain macros to be used to facilitate access to peripheral registers.
Some existing macros are defined in ``soc/soc_ulp.h`` header file. These macros allow access to the fields of peripheral registers by their names.
Peripheral registers names which can be used with these macros are the ones defined in ``soc/rtc_cntl_reg.h``, ``soc/rtc_io_reg.h``, and ``soc/sens_reg.h``.
READ_RTC_REG(rtc_reg, low_bit, bit_width)
Read up to 16 bits from rtc_reg[low_bit + bit_width - 1 : low_bit] into R0. For example::
#include "soc/soc_ulp.h"
#include "soc/rtc_cntl_reg.h"
/* Read 16 lower bits of RTC_CNTL_TIME0_REG into R0 */
READ_RTC_REG(RTC_CNTL_TIME0_REG, 0, 16)
READ_RTC_FIELD(rtc_reg, field)
Read from a field in rtc_reg into R0, up to 16 bits. For example::
#include "soc/soc_ulp.h"
#include "soc/sens_reg.h"
/* Read 8-bit SENS_TSENS_OUT field of SENS_SAR_SLAVE_ADDR3_REG into R0 */