Devlog 4 Validating The Implementation

November 18, 2022

Validating the implementation

I started this session by validating code that’s been written and ensuring things are where they should be. I’ve also created an TODO list to track my progress.

Log 4

As mentioned previously, I defined some constants for the RAM size etc, but I’m not sure if they are actually placed at the correct location in memory. Let’s have a look.

Constant locations

First I want to ensure the __stacktop address, i.e: the top of the stack, is correctly located at 0x20005000: 0x20000000 (start of RAM) + 0x5000 (size of RAM) and placed into the sp (stack pointer / x2) register:

grep __stacktop fiveforths.dump 
    la sp, __stacktop
 800004c:	fb810113          	addi	sp,sp,-72 # 20005000 <__stacktop>

Perfect! Next, let’s ensure our constants point to the correct location:

make readelf  | grep DSP_TOP
18: 20005000     0 NOTYPE  LOCAL  DEFAULT  ABS DSP_TOP
make readelf  | grep RSP_TOP
    19: 20004f00     0 NOTYPE  LOCAL  DEFAULT  ABS RSP_TOP
make readelf  | grep TIB
    20: 20004e00     0 NOTYPE  LOCAL  DEFAULT  ABS TIB_TOP
    21: 20004d00     0 NOTYPE  LOCAL  DEFAULT  ABS TIB

Great! The DSP_TOP should be at 0x20000000 (start of RAM) + 0x5000 (size of RAM) = 0x20005000, the RSP_TOP should be at 0x20005000 - 0x100 (256 Bytes) = 0x20004F00, and the TIB_TOP should be at 0x20004F00 - 0x100 (256 Bytes) = 0x20004E00. Additionally, the lower address of TIB should be at 0x20004E00 - 0x100 (256 Bytes) = 0x20004D00. So far so good, however..

Stack locations

I realized my stacks are placed at the wrong address, here’s a look:

make readelf  | grep _stack
    22: 20000100     0 NOTYPE  LOCAL  DEFAULT    3 data_stack
    23: 20000200     0 NOTYPE  LOCAL  DEFAULT    3 return_stack
    24: 20000300     0 NOTYPE  LOCAL  DEFAULT    3 tib_stack

Oops! They are starting at the bottom of the stack and growing upwards, when in fact they should start at the top and grow downwards. Let’s fix that:

-# reserve 3x 256 Bytes for stacks
-.bss
-.balign STACK_SIZE
-data_stack:
-    .space STACK_SIZE                   # reserve 256 Bytes for data stack
-return_stack:
-    .space STACK_SIZE                   # reserve 256 Bytes for return stack
-tib_stack:
-    .space STACK_SIZE                   # reserve 256 Bytes for terminal buffer

What? I realized that adding labels for these stacks and “reserving” zero-filled space in RAM was pointless. I also removed other reserved spaces for variables and indexes. The indexes will be added above the pad space, so let’s define that:

-indexes:
-    .space (CELL * 64)                  # reserve 64 CELLS zero-filled
+.equ INDEXES, NOOP - (CELL * 64)        # 64 CELLS between NOOP and INDEXES
+.equ PAD, INDEXES - (CELL * 64)         # 64 CELLS between INDEXES and PAD

Additional changes

I made some other minor changes such as creating labels for the ‘ok’, ‘?’, and ‘redefined ok’ strings, so we can easily jump to those when needed.

Since the Longan Nano Lite has 64K FLASH and 20K RAM, and both are located in the same MCU chip, I think it makes more sense to keep the entire program and constants in FLASH, and only use the RAM for variables, stacks, and user-defined dictionary words. This is just my assumption the Forth can actually be executed from FLASH instead of RAM.

Closing thoughts

I’ve made some progress in improving the memory layout and size of the program, but there’s still some work to do there. At least for now I can check off a few things from the TODO list. I planned on testing the current code this session but validating the memory locations/addresses was more work than expected, so I’ll get to that next time before implementing the missing words.