138 SPI FPU
138 : SPI FPU
- Author: Sebastian Pfeiler
- Description: A 32bit floating point adder accessible over spi
- GitHub repository
- Open in 3D viewer
- Clock: 50000000 Hz
How it works
This is a floating point unit accesible over SPI.
SCLK = ui[0]
NCS = ui[1]
COPI = ui[2]
CIPO = uo[0]
Constraint frequency(SCLK) < 4*frequency(clk)
It features 4 internal floating point registers, writeable and readable over SPI.
Every SPI transaction starts with a command byte followed by the arguments to the command. If the command outputs data, enough bytes need to be sent for the entire data to be received.
+-------+---------+--- ... -+
|Command|Arguments|... |
+-------+---------+--- ... -+
WRITE_TO_REG
Writes a floating point value (serialized on COPI) into an internal floating point register.
+----+----+----+----+----+----+
IN: |0x00|0x0r|b[0]|b[1]|b[2]|b[3]|
+----+----+----+----+----+----+
OUT:| und| und| und| und| und| und|
+----+----+----+----+----+----+
r is one of 0,1,2,3 corresponding to the 4 internal registers. b[0] is the lowest byte of the floating point value. b[3] is the highest byte.
PERFORM_ADD
Performs the computation
register[in_c] = fadd(register[in_a], register[in_b])
This computation takes a couple of cycles (6 I think right now).
These cycles are in reference to clk not to SCLK.
Nevertheless add another dummy byte after the command to ensure that the
add is being performed.
+----+----+----+----+----+
IN: |0x00|in_a|in_b|in_c|0x00|
+----+----+----+----+----+
OUT:| und| und| und| und| und|
+----+----+----+----+----+
in_a,in_b,in_c can be one of 0x00,0x01,0x02,0x03
READ_FROM_REG
Reads a float from the internal register and serialises it on CIPO.
+----+----+----+----+----+----+
IN: |0x00|0x0r|xxxx|xxxx|xxxx|xxxx|
+----+----+----+----+----+----+
OUT:| und| und|b[0]|b[1]|b[2]|b[3]|
+----+----+----+----+----+----+
r is one of 0,1,2,3 b[0] is the lowest byte of the float b[3] is the highest byte of the float
How to test
It is best to test with the provided arduino testbench in the repository
https://github.com/Qwendu/tt_float_adder
in the directory src/testbenches/arduino_integration_test.
I have not managed to get it to work with the provided SPI controllers in arduino.
Maybe this is because my spi_rx has some bugs.
External hardware
For integration test:
- Arduino
Known Bugs
-
[1] Denormalized numbers do not always add correctly.
-
[2] When testing on an fpga it was observed that it sometimes worked flawlessly and othertimes the output was always 0. to what extent that was a fault of the testsetup or the tester or the actual code has yet to be determined.
IO
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | SPI_CLK | SPI_OUT | Unused |
| 1 | SPI_NCS | Unused | Unused |
| 2 | SPI_IN | Unused | Unused |
| 3 | Unused | Unused | Unused |
| 4 | Unused | Unused | Unused |
| 5 | Unused | Unused | Unused |
| 6 | Unused | Unused | Unused |
| 7 | Unused | Unused | Unused |