Devlog 28 Execute
December 30, 2022
Log 28
In this session I’ll start by fixing the NEXT macro and then I’ll implement the execute function.
Fixing NEXT
Of course I had a feeling NEXT was wrong, because it was copied blindly from derzforth and has never been tested. However it may actually be correct afterall, but I’m not sure so let’s just move forward this these changes.
The previous NEXT macro looked like this:
.macro NEXT
lw a0, 0(s1) # load memory address from IP into W
addi s1, s1, CELL # increment IP by CELL size
lw t0, 0(a0) # load memory address from W into temporary
jr t0 # jump to the address in temporary
.endm
The new version looks like this:
.macro NEXT
mv a0, s1 # load memory address from IP into W
addi s1, s1, CELL # increment IP by CELL size
jr a0 # jump to the address in W
.endm
It’s a bit simpler, in fact what changed is that it now makes a direct jump to the memory address stored in the IP register (s1). I believe the approach used by derzforth is to perform a double-indirect jump, which is why it was loading an address from an address in the IP register. I’m not sure why it does that, since sectorforth performs a direct jump and jonesforth only a single indirect jump…
Execute
Now with NEXT “fixed” (I think?), we should be able to execute words (interpret them), so let’s define execute here:
execute:
la s1, ok # load the address of the interpreter into the IP register
addi a0, a1, 2*CELL # increment the address of the found word by 8 to get the codeword address
lw t0, 0(a0) # load memory address from W into temporary
execute_done:
jr t0 # jump to the address in temporary
The first thing is to load the ok function’s address into the IP register. Then we increment our found word from the X register (a1) by 2 CELLs, because that’s where we’ll find the codeword address.
Then we load that codeward memory address into a temporary t0 and jump to it.
I know this is probably wrong in many ways, but it seems to somewhat work for a single token. I think that’s because of a bug in my token processing algorithm, which I’ll look at in another session.
Fixing newlines
I discovered a small issue with newlines in the interpreter function. The first piece of code was checking if the character was a comment. The checkchar macro would perform a call uart_get immediately followed by a call uart_put, which was problematic because upon reading a newline it would.. send a newline to the REPL! It looked like this:
nope
?
state
ok
To fix this, i replaced that specific macro call with the expanded form, but added some code before the call uart_put. The code first checks if it’s a newline, if yes it skips the call uart_put and jumps to checking if it’s a comment. Otherwise it will call uart_put and then check if it’s a comment:
+ call uart_get # read a character from UART
+ li t4, CHAR_NEWLINE # load newline into temporary
+ beq a0, t4, skip_send # don't send the character if it's a newline
+ call uart_put # send the character to UART
+
+skip_send:
# validate the character which is located in the W (a0) register
- checkchar CHAR_COMMENT, skip_comment # check if character is a comment
+ li t0, CHAR_COMMENT # load comment character into temporary
+ beq a0, t0, skip_comment # skip the comment if it matches
Now when I type an invalid word in the REPL, I see a ?, or an ok if the word is valid - on the same line. Like this:
nope ?
state ok
Closing thoughts
Now I can execute ONE word (ex: state), but still can’t compile. I also can’t execute TWO ore more words (ex: tib tib +). In the next session I’ll focus on fixing the bug in code execution (hopefully it’s not related to the direct jump, but most likely it is haha), and then I’ll jump to compiling words.