Champ - N. Harold Cham's 65C02 Profiler
This is a 6502/65C02 emulator / profiler that enables you to really get to know your APPLE ][ HiRes Graphics Mode Demo.
- full, cycle-accurate 65C02 emulation
- screen output as animated GIF with exact frame timing
- calculation of average frame rate
- see how much time is spent in which subroutine
- watch variables (single variables or pairs)
- no required dependencies except Ruby, gcc and Merlin32
- optional: GraphViz if you want a call graph
First, make sure you have gcc, ruby and Merlin32 installed. You need to prepare a YAML file to tell champ about all source and object files and their memory locations.
load: 0x6000: plot3d/plot3d242.s 0x1200: plot3d/multtab.s 0x8400: plot3d/ldrwtab.s 0x8900: plot3d/SINETABLE 0x8b00: plot3d/object2.s 0x9000: plot3d/object1.s 0x9500: plot3d/FONT 0xb600: plot3d/projtab.s entry: ENTRY instant_rts: - LOAD1
We specified some source files (which will get compiled automatically) and some object files along with their locations in memory (
load). We also specified the entry point for our program (
entry), this can be a label or an address.
Furthermore, we can disable subroutines by replacing the first opcode with a RTS (
instant_rts). This is necessary in some cases because Champ does not emulate hardware and thus can not load data from disk, for example.
Running the profiler
To start champ, type:
$ ./champ.rb --max-frames 100 plot3d.yaml
This will run the emulator and write the HTML report to
report.html. If you do not specify the maximum number of frames, you can still cancel the emulator by pressing Ctrl+C at any time. If you need fast results and don't need the animated GIF of all frames, specify the
--no-animation flag, which will still give you all the information but without the animation.
You can watch registers or variables at certain program counter addresses by inserting a champ directive in a comment after the respective code line in your assembler source code. All champ directives start with an at sign (
@). Here's an example (example01.yaml / example01.s):
DSK test MX %11 ORG $6000 FOO EQU $8000 JSR TEST BRK TEST LDA #64 ; load 64 into accumulator ASL ; multiply by two @Au STA FOO ; store result @Au RTS
Running this example results in the following watch graphs:
Champ generates a graph for every watch. You can see the watched variable plotted against the cycles, and also the PC address, file name, and source line number of the watch as well as the subroutine in which the watch was defined. At the right border you can see a histogram of the variable, which is pretty minimal in this example but may look more interesting in other cases.
@Au directive, we tell champ to monitor the A register and interpret it as an unsigned 8 bit integer (likewise,
@As would treat the value as a signed 8 bit integer).
By default, all champ values get recorded before the operation has been executed. To get the value after the operation, you can write:
@Au(post). Feel free to add as many champ directives as you need in a single line, each starting with a new at sign.
DSK test MX %11 ORG $6000 FOO EQU $8000 ; @u16 JSR TEST BRK TEST LDA #0 STA FOO LDA #$40 STA FOO+1 ; @FOO(post) RTS
Here, we declare the type of the global variable in the same place the variable itself is declared, using
s16 to declare the type. Later, we can just watch the variable by issuing a champ directive like
@FOO. In this example, we use the
(post) option to see the variable contents after the
DSK test MX %11 ORG $6000 FOO EQU $8000 ; @u8 JSR TEST BRK TEST LDX #$FF LDA #1 LOOP TAY TXA EOR #$FF LSR LSR STA FOO TYA CLC ADC FOO ; @Au(post) DEX ; @Xu(post) @Au,FOO(post) BNE LOOP RTS
This is a small program which lets the accumulator grow exponentially while X decreases linearly:
We can also watch pairs of variables by separating them with a comma in the champ directive:
DEX ; @Xu(post) @Au,FOO(post)
This will plot FOO against A:
Subroutine cycle watches
DSK test MX %11 ORG $6000 LDX #$20 JSR COUNT LDX #$30 JSR COUNT LDX #$40 JSR COUNT BRK COUNT DEX ; @Xu(post) @cycles BNE COUNT RTS
Please note that this will only yield data for labels which actually get called via
JSR at some point in the program.
This programm calls the
COUNT subroutine three times with different X arguments, and we get both X and the number of cycles spent in
We see the three incantations of
X decreasing to 0 each time, and at the end of every loop, the amount of cycles spent in
To disable a watch, add a
; right behind the
ADC FOO ; @;Au(post)
Should your program run into an error, champ shows you where in the source code the error occured and an execution log of the previous 20 CPU steps (you can increase the size of the log with
Look at this example (example05.yaml / example05.s):
DSK test MX %11 ORG $6000 LDX #$ff COUNT PHA PHA DEX BNE COUNT
This is a program which repeatedly pushes the A register onto the stack until the stack is full. Stack overflow, mission accomplished! Champ will generate the following output:
On the left hand side you can see that the error occured in example05.s, line 7, while attempting a
PHA operation when
SP is already down to zero from previous
PHA operations (as can be seen on the right).
Did you know?
By the way, there's a full-fledged, incremental, standalone, no-dependencies GIF encoder in pgif.c that writes animated GIFs and uses some optimizations to further minimize space. It's stream-friendly and as you feed pixels in via
stdin, it dutifully writes GIF data to
stdin gets closed.