Scanned from source published in 80-BUS NEWS from Vol 2, Issue 3 (May-June 1983) to Vol 3, Issue 3 (May-June 1984).
Adapted for the freeware Zilog Macro Assembler 2.10 to produce the original ROM code (checksum A934H). PA
http://www.nascomhomepage.com/
==============================================================================
The updates to the original BASIC within this file are copyright (C) Grant Searle
You have permission to use this for NON COMMERCIAL USE ONLY. If you wish to use it elsewhere, please include an acknowledgement to myself.
==============================================================================
The rework to support MS Basic HLOAD, RESET, and the 8085 and Z80 instruction tuning are copyright (C) 2020-2021 Phillip Stevens
This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
@feilipu, August 2020
==============================================================================
ACIA 6850 interrupt driven serial I/O to run modified NASCOM Basic 4.7.
Full input and output buffering with incoming data hardware handshaking. Handshake shows full before the buffer is totally filled to allow run-on from the sender. Transmit and receive are interrupt driven, and are fast. Use 115200 baud with 8n2 setting.
Receive buffer is 255 bytes, to allow efficient pasting of Basic into the editor. The Transmit buffer is 63 bytes.
These ROMs provides both Intel HLOAD function and a RST, INT0, and NMI RAM JumP Table, starting at 0x8000. This allows you to upload Assembly or compiled C programs, and then run them as described.
This ROM works with the most basic default versions of the RC2014, with 32k of RAM. This is the ROM to choose if you want fast I/O from a standard Z80 based RC2014.
This version requires a 64k/56k RAM module. The 56k version utilises the full 56k RAM memory space of the RC2014, starting at 0x2000.
This version provides both Intel HLOAD function and a RST, INT0, and NMI RAM JumP Table, starting at 0x2000. This allows you to upload Assembly or compiled C programs, and then run them as described.
This ROM works with the most basic default versions of the RC2014, with 32k of RAM. This is the ROM to choose if you want fast I/O from a standard RC2014, and you have installed an Am9511A APU Module.
This version works with the most basic default version of the RC2014 running with an 8085 CPU Module, with 32k of RAM. This is the ROM to choose if you want fast I/O from a standard RC2014, and you have installed an 8085 CPU Module.
This version works with the most basic default version of the RC2014 running with an 8085 CPU Module, with 32k of RAM. This is the ROM to choose if you want fast I/O from a standard RC2014, and you have installed both an 8085 CPU Module and Am9511A APU Module.
==============================================================================
There are a number of important Z80 addresses or origins that need to be managed within your assembly program.
For convenience, because we can't easily change the ROM code interrupt routines already present in the RC2014, the ACIA serial Tx and Rx routines are reachable by calling RST instructions from your program.
- Tx:
RST 08Hexpects a byte to transmit in thearegister. - Rx:
RST 10Hreturns a received byte in thearegister, and will block (loop) until it has a byte to return. - Rx Check:
RST 18Hwill immediately return the number of bytes in the Rx buffer (0 if buffer empty) in thearegister. - ACIA Interrupt:
RST 38His used by the ACIA 68B50 Serial Device.
All RST xxH targets can be rewritten in a JP table originating at 0x8000 in RAM. This allows the use of debugging tools and reorganising the efficient RST call instructions as needed.
For the RC2014 with 32k Basic the location for USR(x) loaded user program address is 0x8204, and with 56k Basic the location for USR(x) is 0x2204.
-
Select the preferred origin
.ORGfor your arbitrary program, and assemble a HEX file using your preferred assembler, or compile a C program using z88dk. For the RC2014 32kB, suitable origins commence from0x8400, and the default z88dk origin for RC2014 is0x9000. -
Give the
HLOADcommand within Basic. -
Using a serial terminal, upload the HEX file for your arbitrary program that you prepared in Step 1, using the Linux
catutility or similar. If desired the pythonslowprint.pyprogram can also be used for this purpose.python slowprint.py > /dev/ttyUSB0 < myprogram.hexorcat > /dev/ttyUSB0 < myprogram.hex. The RC2014 interface can absorb full rate uploads, so usingslowprint.pyis an unnecessary precaution. -
Start your program by typing
PRINT USR(0), or? USR(0), or other variant if you have a parameter to pass to your program. -
Profit.
Note that your program and the USR(x) jump address setting will remain in place through a RC2014 Warm Reset, provided you prevent Basic from initialising the RAM locations you have used. Also, you can reload your assembly program to the same RAM location through multiple Warm Resets, without reprogramming the USR(x) jump.
Any Basic programs loaded will also remain in place during a Warm Reset.
Issuing the RESET keyword will clear the RC2014 RAM, and return the original memory contents.
MS Basic uses 4 Byte values extensively as floating point numbers in Microsoft Binary Format, and as pointers to strings. Many of the improvements are in handling these values as they are shifted around in memory, and to BCDE registers and the stack.
- 4
LDIinstructions are used to move values from one location (the Floating Point RegisterFPREG) to another location in memory, and these are in-lined to also save the call-return cycles. - The
LD (x),DELD(x+2),BCinstruction pair is used to grab values into registers and save from registers, avoiding the need to preserveHLand often saving push-pop cycles and of course the call-return cycles. - There is a 16_16x16 multiply
MLDEBCused to calculate table offsets, which was optimised to use shift instructions available to the Z80. I experimented with different zero multiplier checks, and with removing the checks, but Microsoft had already done the right optimisation there, so it was left as it was. - The extensions that Grant Searle had inserted into the operand evaluation chain to check for Hex and Binary numbers were moved to the end of the operand checks, so as not to slow down the normal operand or function evaluation.
Doing these changes got about 6% improvement in the benchmarks.
The next step was to use the z88dk-ticks tool to evaluate hotspots and try to remediate them. Using the debug mode it is possible to capture exactly how many iterations (visits) and how many cycles are consumed by each instruction.
The testing revealed that the comparison function CPDEHL was a very heavily used function. As it is quite small, and through removing the call-return overhead, it adds only a few bytes per instance to in-line it. There is plenty of space in the 8kB ROM to allow this change so it was made.
Then, the paths taken by the JR and JP conditional instructions were examined, by checking which path was taken most frequently across the benchmarks. This resulted in changing a few JR instructions for JP instructions, when the conditional path was mostly true, and one replacement of a JP instruction where the conditional was most often false.
Looking further at z88dk-ticks hotspot results, the next most used function is GETCHR used to collect input from code strings. GETCHR is a larger function and is used about 50 times throughout the code base, so there is little point to in-line it. However I do note the new JR conditional is used in checking for spaces in token strings, which does save a few cycles. Microsoft warns in the Nascom Basic Manual to optimise performance by removing spaces in code. Now it is even more true than before.
As the Z80 and 8085 have better shift instructions than the 8080, these instructions have been used where possible. Specifically for the 8085 the rl de and the sra hl undocumented instructions have been used where appropriate.
So with these changes we are now at 12% improvement over the original Microsoft code.
So at this point I'll call it done. It seems that without rewriting the code substantially that's about all that I can squeeze out. The result is that with no change in function, MS Basic is now simply 12% faster.
==============================================================================
YAZ180 (deprecated, see yabios)
ASCI0 interrupt driven serial I/O to run modified NASCOM Basic 4.7.
If you're using the YAZ180 with 32kB Nascom Basic, then all of the RAM between 0x3000 and 0x7FFF is available for your assembly programs, without limitation. In the YAZ180 the area between 0x2000 and 0x2FFF is reserved for system calls, buffers, and stack space. For the RC2014 the area from 0x8000 is reserved for these uses.
In the YAZ180 32kB Basic, the area from 0x4000 to 0x7FFF is the Banked memory area, and this RAM can be managed by the HexLoadr program to write to all of the physical RAM space using ESA Records.
HexLoadr supports the Extended Segment Address Record Type, and will store the MSB of the ESA in the Z180 BBR Register. The LSB of the ESA is silently abandoned. When HexLoadr terminates the BBR is returned to the original value.
Two versions of initialisation routines NASCOM Basic are provided.
The 56k version utilises the full 56k RAM memory space of the YAZ180, starting at 0x2000.
Full input and output ASCI0 buffering. Transmit and receive are interrupt driven.
Receive buffer is 255 bytes, to allow efficient pasting of Basic into the editor. Receive buffer overflows are silently discarded.
Transmit buffer is 255 bytes, because the YAZ180 is 36.864MHz CPU. Transmit function busy waits when buffer is full. No Tx characters lost.
The 32k version uses the CA0 space for buffers and the CA1 space for Basic.
This leaves the Bank RAM / Flash space in 0x4000 to 0x7FFF available for other usage.
The rationale is to allow in-circuit programming, and an exit to another system. An integrated HexLoadr program is provided for this purpose.
Full input and output ASCI0 buffering. Transmit and receive are interrupt driven.
Receive buffer is 255 bytes, to allow efficient pasting of Basic into the editor. Receive buffer overflows are silently discarded.
Transmit buffer is 255 bytes, because the YAZ180 is 36.864MHz CPU. Transmit function busy waits when buffer is full. No Tx characters lost.