Arduino library for AVR optimized shiftOut - e.g. 74HC595.

FastShiftOut is a class that has optimized code (AVR only) to shift out data faster than the default provided shiftOut() function. It speeds up the shift using low level ports and masks. These are predetermined in the constructor of the FastShiftOut object.

If not an ARDUINO_ARCH_AVR or ARDUINO_ARCH_MEGAAVR the class falls back to the default shiftOut() implementation.

The library allows to set (and get) the bitOrder and apply this to multiple write() calls. It also provide access to writeLSBFIRST() and writeMSBFIRST() which are the low level workers and most optimized code (so far).

The library provides wrapper functions to write multi-byte variables. These are write16(), write24(), write32() and write(array, size). The latter is used to shift out any size object.

The 0.4.0 version has a flag to unroll the inner loop in writeLSBFIRST() and writeMSBFIRST(). The AVR optimized code blocks the interrupts per byte.

Note: this optimization is new and thus experimental. Feedback, including improvements, is welcome.

The performance of write() is substantially faster for AVR than the default Arduino shiftOut(), but not as fast as HW SPI (need test?). Exact how large the performance gain is can be seen with the example FastShiftOut_test.ino. It does a measurement of different functions and shows how the class is to be used. See also table below.

Numbers may vary depending on bit-order flag.

Indicative time in microseconds, Arduino UNO, IDE 1.8.19, measured over 1000 calls.
(delta between 2 calls and 1 call to eliminate overhead)

• Note: 0.3.3 has improved the measurement, not the code sec.
• Note: 0.3.3 numbers fixed when implementing 0.4.0. (error in test sketch).
• Note: 0.4.0 measured with loop unroll flag disabled.
• Note: 0.4.0L measured with loop unrolled flag enabled.
• Note: 0.4.1 / 0.4.1L idem.
• Note: Loop unrolled is (8046 - 7818 = 128) bytes larger in size.

• https://github.com/RobTillaart/FastShiftIn
• https://github.com/RobTillaart/FastShiftInOut
• https://github.com/RobTillaart/FastShiftOut
• https://github.com/RobTillaart/ShiftInSlow
• https://github.com/RobTillaart/ShiftOutSlow
• https://github.com/RobTillaart/SWSPI (experimental)

#include "FastShiftOut.h"

bitOrder = { LSBFIRST, MSBFIRST };

• FastShiftOut(uint8_t dataOut, uint8_t clockPin, uint8_t bitOrder = LSBFIRST) Constructor.

• size_t write(uint8_t data) send a byte, also the workhorse of the Print interface.
• size_t write16(uint16_t data) send 2 bytes. Wrapper around 8 bit calls.
• size_t write24(uint32_t data) send 3 bytes. Wrapper around 8 bit calls.
• size_t write32(uint32_t data) send 4 bytes. Wrapper around 8 bit calls.
• size_t write(uint8_t *array, size_t size) send array of size bytes.
• uint8_t lastWritten() returns last byte written.
• size_t writeLSBFIRST(uint8_t data) lowest level function, optimized for LSB.
• size_t writeMSBFIRST(uint8_t data) lowest level function, optimized for MSB.

• bool setBitOrder(uint8_t bitOrder) set LSBFIRST or MSBFIRST. Returns false for other values ==> no change.
• uint8_t getBitOrder(void) returns LSBFIRST or MSBFIRST as set in constructor or latest set from setBitOrder().

As the FastShiftOut library implements the Print interface, one can also call

• size_t FSO.print(any type) or
• size_t FSO.println(any type)

to send e.g. a float with 4 digits over the line, or some text string.

Note: FSO.print() and FSO.println() return the number of characters printed, including an optional \r and \n.

The functions write16(), write24() and write32() of this library assume that the BIT-order is also the BYTE-order. This is not always the case as an n-byte element can have n! == factorial(n) distinct byte orders.

So write16() can have two, write24() can have six and write32() can even have (in theory) 24 distinct byte orders. Although LSB and MSB are the most common, other byte orders exist, and sometimes one explicitly wants to reorder the bytes.

If the BIT-order is not the BYTE-order, the user has two options

• call write() multiple times and merge the bytes in the order needed.
• call write32() (a.o) and reorder the bytes in a separate function.

The library will not support such functionality.

• The optimizations are AVR only for now, other platforms may follow.
• The 74HC595 needs 0.1uF caps and the data and clock lines may need
  pull up resistors, especially if wires are exceeding 10 cm (4").

• keep in sync with FastShiftIn()

• extend unit tests

• investigate ESP32 optimization readLSBFIRST readMSBFIRST
• performance ESP32
• example schema
• add invert flag?
  • if ((value & m) == 0) ==> if ((value & m) == invertFlag...
  • derived class?

• investigate separate BYTE-order,
  • only MSBFirst and LSBFirst
  • void setByteOrder() + uint8_t getByteOrder()

If you appreciate my libraries, you can support the development and maintenance. Improve the quality of the libraries by providing issues and Pull Requests, or donate through PayPal or GitHub sponsors.

Thank you,