3 ms·

I wasn't able to find good information about how many instructions per subroutine call computers typically run these days. Probably I'm not using the right sea

by kragen 1y ago

I wasn't able to find good information about how many instructions per subroutine call computers typically run these days. Probably I'm not using the right search terms, but here's what I found. I just ran Emacs a few times under valgrind --tool=callgrind. Perhaps surprisingly, it was usably fast once it started up. I analyzed callgrind's output file with the following highly sophisticated methodology to determine the total number of subroutine calls (and presumably returns): print(sum(int(mo.group(1)) for line in sys.stdin for mo in [re.match(r'^calls=(\d+)', line)] if mo)) This, plus some division by numbers of executed instructions reported, revealed that it had a subroutine call for about every 55–59 instructions, and about a hundred and fifty thousand extra subroutine calls (and 7.7 million extra instructions) per character typed. This suggests that, at least in Emacs, the cost of subroutine call and return is probably higher than 15%, maybe 25%. Other programs I tested: - GNU ls: 79 instructions per call - PARI/GP: 360 instructions per call - CPython with Numpy: 88 instructions per call - Minetest (rather unplayable): 88 instructions per call

brucehoult 1y ago
Interesting results. Are you able to figure out how many of those are calls of leaf functions? On modern ISAs and ABIs calling a small leaf function is very cheap -- just the `jal` and `ret`, plus getting the arguments into the right registers first. In many cases memory is not touched at all (unless the code in the function necessarily touches memory), no stack frame is created etc.
- kragen 1y ago
  Hmm, I think that ought to be pretty doable with the callgrind output actually. I'm using amd64 rather than a modern ISA, so the calls do necessarily touch "memory", but leaf functions are still especially cheap—they can use temporary registers for local variables. In the case of Emacs most of the functions are compiled Lisp, and I doubt the Lisp compiler is smart enough to exploit leafness in its register allocator.
- kragen 1y ago
  Okay, so, I wrote a program http://canonical.org/~kragen/sw/dev3/leafcalls.py http://canonical.org/~kragen/sw/dev3/leafcalls.py to list the leaf functions with the number of calls to each, and totaled up the results. Assuming I didn't fuck up the analysis, which is never a good bet, here it is: <table> <tr><th>PID <th>Program <th>Instructions <th>Calls <th>Leaf calls <th>Functions <th>Leaf functions <tr><td>2603407 <th>Emacs <td>18,278,190,938 <td>328,734,692 <td>100,779,209 <td>16,875 <td>4,523 <tr><td>2611631 <th>Emacs <td>10,444,575,778 <td>176,576,759 <td> 59,334,998 <td>16,262 <td>4,470 <tr><td>2612835 <th>Emacs <td>10,453,593,569 <td>176,931,778 <td> 59,425,409 <td>16,422 <td>4,483 <tr><td>2622964 <th>ls -l <td> 28,770,079 <td> 362,597 <td> 154,386 <td> 520 <td> 218 <tr><td>2623998 <th>PARI/GP <td> 19,265,853 <td> 52,633 <td> 19,983 <td> 828 <td> 346 <tr><td>2624712 <th>Numpy <td> 356,605,939 <td> 4,040,096 <td> 952,628 <td> 4,366 <td>1,259 <tr><td>2627189 <th>Minetest<td>62,476,314,644 <td>708,735,071 <td>354,140,648 <td>22,318 <td>5,556 </table> I did look at the output file listing the leaf functions and it is at least not obviously wrong. So, if this is right, typically almost half of all calls are calls to leaf functions (functions that never call another function), although my Emacs and Numpy runs have only about a third and a fourth of their calls being leaf calls. All of these numbers are surprisingly low to me; I would have expected the answer to be more like two thirds to three fourths. There's an optimization I've applied sometimes the opportunity for which wouldn't show up here: on a leaf call that returns before it calls any other subroutines, you can sometimes avoid doing all the foofaraw for a non-leaf call by delaying it until after you know that you're in a non-leaf case. This isn't always a pure win, though, and I've never seen a compiler do it. The call graph output by callgrind only has subroutine granularity, so, absent any alternative, I'm only counting calls as leaf calls if they are calls to a leaf subroutine, that is, a subroutine that never calls other subroutines. I'm interested to hear if you have other stats!
  - brucehoult 1y ago
    > I've never seen a compiler do it. Oh I have. Often. https://godbolt.org/z/PM9herj3P https://godbolt.org/z/PM9herj3P Note: these could be tail-calls, and are if you use `-O2` or `-Os`, but that's not important. Make it `foo(n) + 13` and you still get the "don't make a stack frame unless needed" optimisation. Oh, here's an interesting one. Arm proactively pushes everything, x86 and RISC-V only if `err()` will be called. https://godbolt.org/z/93WEbc1q9 https://godbolt.org/z/93WEbc1q9 I also would expect the number of function executions which are dynamically leaf functions [1] to be greater than 50%. Except for the kind of perverse call chains you get in Java and perhaps some other OO code bases where there are huge numbers of layers of abstraction where each method simply calls one other lower level method. [1] i.e. that don't call another function on that execution, though in other circumstances they could.

by kragen 1y ago

brucehoult 1y ago

Interesting results. Are you able to figure out how many of those are calls of leaf functions? On modern ISAs and ABIs calling a small leaf function is very cheap -- just the `jal` and `ret`, plus getting the arguments into the right registers first. In many cases memory is not touched at all (unless the code in the function necessarily touches memory), no stack frame is created etc.

kragen 1y ago
Hmm, I think that ought to be pretty doable with the callgrind output actually. I'm using amd64 rather than a modern ISA, so the calls do necessarily touch "memory", but leaf functions are still especially cheap—they can use temporary registers for local variables. In the case of Emacs most of the functions are compiled Lisp, and I doubt the Lisp compiler is smart enough to exploit leafness in its register allocator.
kragen 1y ago
Okay, so, I wrote a program http://canonical.org/~kragen/sw/dev3/leafcalls.py http://canonical.org/~kragen/sw/dev3/leafcalls.py to list the leaf functions with the number of calls to each, and totaled up the results. Assuming I didn't fuck up the analysis, which is never a good bet, here it is: <table> <tr><th>PID <th>Program <th>Instructions <th>Calls <th>Leaf calls <th>Functions <th>Leaf functions <tr><td>2603407 <th>Emacs <td>18,278,190,938 <td>328,734,692 <td>100,779,209 <td>16,875 <td>4,523 <tr><td>2611631 <th>Emacs <td>10,444,575,778 <td>176,576,759 <td> 59,334,998 <td>16,262 <td>4,470 <tr><td>2612835 <th>Emacs <td>10,453,593,569 <td>176,931,778 <td> 59,425,409 <td>16,422 <td>4,483 <tr><td>2622964 <th>ls -l <td> 28,770,079 <td> 362,597 <td> 154,386 <td> 520 <td> 218 <tr><td>2623998 <th>PARI/GP <td> 19,265,853 <td> 52,633 <td> 19,983 <td> 828 <td> 346 <tr><td>2624712 <th>Numpy <td> 356,605,939 <td> 4,040,096 <td> 952,628 <td> 4,366 <td>1,259 <tr><td>2627189 <th>Minetest<td>62,476,314,644 <td>708,735,071 <td>354,140,648 <td>22,318 <td>5,556 </table> I did look at the output file listing the leaf functions and it is at least not obviously wrong. So, if this is right, typically almost half of all calls are calls to leaf functions (functions that never call another function), although my Emacs and Numpy runs have only about a third and a fourth of their calls being leaf calls. All of these numbers are surprisingly low to me; I would have expected the answer to be more like two thirds to three fourths. There's an optimization I've applied sometimes the opportunity for which wouldn't show up here: on a leaf call that returns before it calls any other subroutines, you can sometimes avoid doing all the foofaraw for a non-leaf call by delaying it until after you know that you're in a non-leaf case. This isn't always a pure win, though, and I've never seen a compiler do it. The call graph output by callgrind only has subroutine granularity, so, absent any alternative, I'm only counting calls as leaf calls if they are calls to a leaf subroutine, that is, a subroutine that never calls other subroutines. I'm interested to hear if you have other stats!
- brucehoult 1y ago
  > I've never seen a compiler do it. Oh I have. Often. https://godbolt.org/z/PM9herj3P https://godbolt.org/z/PM9herj3P Note: these could be tail-calls, and are if you use `-O2` or `-Os`, but that's not important. Make it `foo(n) + 13` and you still get the "don't make a stack frame unless needed" optimisation. Oh, here's an interesting one. Arm proactively pushes everything, x86 and RISC-V only if `err()` will be called. https://godbolt.org/z/93WEbc1q9 https://godbolt.org/z/93WEbc1q9 I also would expect the number of function executions which are dynamically leaf functions [1] to be greater than 50%. Except for the kind of perverse call chains you get in Java and perhaps some other OO code bases where there are huge numbers of layers of abstraction where each method simply calls one other lower level method. [1] i.e. that don't call another function on that execution, though in other circumstances they could.