Page 1 of 1
Simple Benchmarks
Posted: Thu Jan 17, 2019 5:28 am
by basicchip
I should have published something like this a long time ago. Of course you can find CoreMark tests for all these CPUs. But I thought I would give a simple test of BASIC with a floating point divide in the loop.
Code: Select all
dim x as single
x=1000000
while x>1
i=i+1
x=x/1.0001
loop
The variable i was there just to count iterations - 138140
Rooted around the office for various parts -- some never became products, or were specials versions for OEM customers
And the envelope please--
- GPD PC 31 msec Atom x7 z8750 1.6 GHz
- Dell PC 9 msec i5 3330S 2.76 GHz
Now the PC's are interpreting the BASIC, not compiled like the ARM
Re: Simple Benchmarks
Posted: Sun Jan 20, 2019 5:41 pm
by basicchip
Mike just walked me through generating code for SAMD21 -- known as Arduino Zero
and --
Check the blog at
https://www.coridium.us/coridium/blog/some-benchmarks for more complete numbers
Re: Simple Benchmarks
Posted: Mon Jan 21, 2019 5:44 am
by basicchip
Just for reference here is an equivalent program in C for another CPU. If you don't declare the variables volatile, gcc can optimize the whole loop out of existance.
Code: Select all
#include "mstimer.h"
volatile int i;
volatile float x;
int main() // Main function
{
printf("starting\n");
mstime_set(0);
mstime_start();
i = 1000000;
while (i) {
i=i-1; // same speed i--;
}
mstime_stop();
printf("integer time %d\n",mstime_get());
mstime_set(0);
mstime_start();
x = 1000000;
while (x>1) {
i=i+1;
x=x/1.0001;
}
mstime_stop();
printf("float time %d\n",mstime_get());
}
Re: Simple Benchmarks
Posted: Fri Feb 22, 2019 3:16 am
by basicchip
And for an older CPU, the float was 73994 msec, and integer time was 9399 msec. Yep chips have gotten faster over the years.
Re: Simple Benchmarks
Posted: Mon Aug 19, 2019 2:55 pm
by basicchip
I found a PC BASIC that is compiled rather than interpreted. It is QB64 and compiles to 64 bit op codes, it comes it about 10x faster than the PC-ARMbasic emulator.
Added that to the benchmarks blog table.
https://www.coridium.us/coridium/blog/some-benchmarks
Re: Simple Benchmarks
Posted: Thu Aug 29, 2019 11:29 pm
by basicchip
Got a couple micromites, as I was curious about performance on those interpreters.
48 MHz PIC32MX170F256B which according to Coremark are a bit faster than the same MHz ARM Cortex M0.
float test -- 23.606 seconds
integer test -- 81.992 seconds
While it a cute solution if you have a VT100 lying around, most people connect them to PCs, so not sure the Apple2 on a chip makes as much sense as a use case. Though I have talked to people who think it's a good fit for their application and they went that way vs. our compiler.
I applaud another BASIC designer, and tip my hat to his complete package. Yes his documentation is better than mine.
Also got a microbit so will test that too.
Re: Simple Benchmarks
Posted: Wed Sep 04, 2019 7:56 pm
by basicchip
Microbit numbers now posted at the list in the blog.
https://www.coridium.us/coridium/blog/some-benchmarks
Re: Simple Benchmarks
Posted: Tue Sep 17, 2019 12:49 am
by basicchip
As part of the Kickstarter for uChip, someone asked about performance of BASIC. So I did a quick comparison of my benchmarks in C vs BASIC.
Turns out the integer benchmark runs about the same for C and BASIC. Not real surprising as not much to optimize there.
The float benchmark threw me for a bit of a loss, as the C was running about 3 times slower. So from MakeItC I did a ASM listing to see what was going on. Turns out gcc was converting the numbers to double precision, then calculating and then converting back. I'm sure there is some setting I could do or some cast to keep it single precision. But it was surprising.
My benchmarks are very simple tests and not really meant as any kind of exhaustive comparison. Just a simple ballpark answer.
In general gcc will do better optimizations than BASIC, so it will be faster in many cases.
https://www.kickstarter.com/projects/70 ... ero-boards