New to profiling: dmd -profile; ldc2 --fdmd-trace-functions
Thanks in advance for any assistance. As the subject line states I'm just now trying to learn profiling. I have a very small program with 1/2 dozen functions and would like to see where the cpu is spending the most time. I've tried both of these lines with identical results: **ldc2 --fdmd-trace-functions myprog.d** **dmd -profile myprog.d** The result is that no 'trace' files are generatedas far as I can tell. When I run myprog.exe it generates no errors, runs very quickly, and does not do what it previously was doing. **ldc2 --version** **LDC - the LLVM D compiler (1.30.0-git-5fd86e5):** **based on DMD v2.100.1 and LLVM 14.0.3** **dmd --version** **DMD64 D Compiler v2.100.2-dirty**
Re: does dmd --build=profile-gc work with core.stdc.stdlib.exit()?
On Sunday, 13 November 2022 at 19:02:29 UTC, mw wrote: BTW, can --build=profile-gc can intercept "Ctrl+C" and generate *partial* report file? And what's the suggested proper way to do Is there a profile-gc plugin function I can call in the middle of my program to generate *partial* report file?
Re: does dmd --build=profile-gc work with core.stdc.stdlib.exit()?
On Sunday, 13 November 2022 at 18:51:17 UTC, mw wrote: On Sunday, 13 November 2022 at 18:48:42 UTC, mw wrote: BTW, can --build=profile-gc can intercept "Ctrl+C" and generate *partial* report file? And what's the suggested proper way to do early exit, and still let --build=profile-gc generate reports? I tried presss "Ctrl+C", and that cannot stop the program, it just hangs there. I have to `kill -9 ` it to get it stopped. My build command is: ``` /dmd2/linux/bin64/dub build --build=profile-gc --config=... --compiler=dmd ```
Re: does dmd --build=profile-gc work with core.stdc.stdlib.exit()?
On Sunday, 13 November 2022 at 18:48:42 UTC, mw wrote: BTW, can --build=profile-gc can intercept "Ctrl+C" and generate *partial* report file? And what's the suggested proper way to do early exit, and still let --build=profile-gc generate reports? I tried presss "Ctrl+C", and that cannot stop the program, it just hangs there. I have to `kill -9 ` it to get it stopped.
does dmd --build=profile-gc work with core.stdc.stdlib.exit()?
Hi, I'm mem-profiling a multi-threaded program, and want it to exit early, so I added a call ``` core.stdc.stdlib.exit(-1); ``` in a loop in one of the thread. However when the program reached this point, it seems hang: it's not exiting, and CPU usage dropped to 0%. I'm wondering does dmd --build=profile-gc work with core.stdc.stdlib.exit()? And where is the output report file, and the filename? I didn't see any report file generated in the current working dir. BTW, can --build=profile-gc can intercept "Ctrl+C" and generate *partial* report file? And what's the suggested proper way to do early exit, and still let --build=profile-gc generate reports? Thanks!
Re: How to profile compile times of a source code?
On Saturday, 30 January 2021 at 22:47:39 UTC, Ahmet Sait wrote: I'm looking for ways to figure out what parts of the code slows down the compiler other than brute force trial. You could try some of the tools listed on the wiki for build time profiling: https://wiki.dlang.org/Development_tools#Build_time_profiling (intentional bump to aid search results, as apparently this list is not very well known)
Re: How to profile compile times of a source code?
On Sunday, 31 January 2021 at 16:39:12 UTC, Boris Carvajal wrote: On Sunday, 31 January 2021 at 16:13:24 UTC, Anonymouse wrote: Does ldc produce traces in a format that Tracy supports? I can't seem to open the generated *.time-trace files with it. (tracy 0.7.5-1 installed from Arch Linux AUR.) Use the 'import-chrome' tool bundled in Tracy to convert it. Indeed. It's a little annoying that you will see many horizontal 'traces' in tracy and probably will have to scroll all the way down to see the actual LDC trace. It's a tradeoff I made between nicely displayed trace block names, vs. using only a few categories... Maybe other people can chime in a figure out a better way of outputting the trace item names (we are using LLVM functionality here btw). -Johan
Re: How to profile compile times of a source code?
On Sunday, 31 January 2021 at 16:13:24 UTC, Anonymouse wrote: On Sunday, 31 January 2021 at 12:31:59 UTC, Johan Engelen wrote: Try LDC 1.25 (now in beta testing) with --ftime-trace. Clang has the same option, so you can read more about it online in that context. Be sure to check out the related commandline flags. I recommend the Tracy UI to look at traces, because it is by far the fastest viewer of large traces. https://github.com/wolfpld/tracy -Johan Does ldc produce traces in a format that Tracy supports? I can't seem to open the generated *.time-trace files with it. (tracy 0.7.5-1 installed from Arch Linux AUR.) Use the 'import-chrome' tool bundled in Tracy to convert it.
Re: How to profile compile times of a source code?
On Sunday, 31 January 2021 at 12:31:59 UTC, Johan Engelen wrote: Try LDC 1.25 (now in beta testing) with --ftime-trace. Clang has the same option, so you can read more about it online in that context. Be sure to check out the related commandline flags. I recommend the Tracy UI to look at traces, because it is by far the fastest viewer of large traces. https://github.com/wolfpld/tracy -Johan Does ldc produce traces in a format that Tracy supports? I can't seem to open the generated *.time-trace files with it. (tracy 0.7.5-1 installed from Arch Linux AUR.)
Re: How to profile compile times of a source code?
On Sunday, 31 January 2021 at 12:31:59 UTC, Johan Engelen wrote: On Sunday, 31 January 2021 at 12:16:50 UTC, Imperatorn wrote: On Saturday, 30 January 2021 at 23:34:50 UTC, Stefan Koch wrote: [...] Interesting. Is this something that we could get into dmd with a switch? 🤔 Try LDC 1.25 (now in beta testing) with --ftime-trace. Clang has the same option, so you can read more about it online in that context. Be sure to check out the related commandline flags. I recommend the Tracy UI to look at traces, because it is by far the fastest viewer of large traces. https://github.com/wolfpld/tracy -Johan Cool, will take a look 👍
Re: How to profile compile times of a source code?
On Sunday, 31 January 2021 at 12:16:50 UTC, Imperatorn wrote: On Saturday, 30 January 2021 at 23:34:50 UTC, Stefan Koch wrote: On Saturday, 30 January 2021 at 22:47:39 UTC, Ahmet Sait wrote: [...] I have a way of getting the profile data your are after. Get the dmd_tracing_20942 branch from https://github.com/UplinkCoder/dmd Compile that version of dmd. this special version of dmd will generate a trace file which can be read with the included printTraceHeader tool Interesting. Is this something that we could get into dmd with a switch? 🤔 Try LDC 1.25 (now in beta testing) with --ftime-trace. Clang has the same option, so you can read more about it online in that context. Be sure to check out the related commandline flags. I recommend the Tracy UI to look at traces, because it is by far the fastest viewer of large traces. https://github.com/wolfpld/tracy -Johan
Re: How to profile compile times of a source code?
On Saturday, 30 January 2021 at 23:34:50 UTC, Stefan Koch wrote: On Saturday, 30 January 2021 at 22:47:39 UTC, Ahmet Sait wrote: [...] I have a way of getting the profile data your are after. Get the dmd_tracing_20942 branch from https://github.com/UplinkCoder/dmd Compile that version of dmd. this special version of dmd will generate a trace file which can be read with the included printTraceHeader tool Interesting. Is this something that we could get into dmd with a switch? 🤔
Re: How to profile compile times of a source code?
On Saturday, 30 January 2021 at 23:34:50 UTC, Stefan Koch wrote: this special version of dmd will generate a trace file which can be read with the included printTraceHeader tool you will want to take a look at the PhaseHist command which shows you the compiler phase that took the most time. Alternative I recommend using callgrind to profile where dmd spents most of it's time. For that to be useful you need a debug build of dmd though.
Re: How to profile compile times of a source code?
On Saturday, 30 January 2021 at 22:47:39 UTC, Ahmet Sait wrote: I'm looking for ways to figure out what parts of the code slows down the compiler other than brute force trial. Can I use -vtemplates switch for this? Would -v (verbose) switch helpful in some way? How would I know if my bottleneck is ctfe or templates? How do the compiler devs approach this issue? I'm interested in all kinds of tricks to the point of debugging the compiler itself although anything less complicated would be appreciated. I have a way of getting the profile data your are after. Get the dmd_tracing_20942 branch from https://github.com/UplinkCoder/dmd Compile that version of dmd. this special version of dmd will generate a trace file which can be read with the included printTraceHeader tool
How to profile compile times of a source code?
I'm looking for ways to figure out what parts of the code slows down the compiler other than brute force trial. Can I use -vtemplates switch for this? Would -v (verbose) switch helpful in some way? How would I know if my bottleneck is ctfe or templates? How do the compiler devs approach this issue? I'm interested in all kinds of tricks to the point of debugging the compiler itself although anything less complicated would be appreciated.
Re: Is there a way to benchmark/profile portably?
On Thursday, 7 May 2020 at 10:51:27 UTC, Simen Kjærås wrote: If I understand correctly, you want to measure how many cycles pass, rather than clock time? Something like that. Well, I would also like to eliminate differences based on different memory caches between machines. In addition, if the profiler could eliminate the randomness of the benchamrks that results from memory aligment, context switches and dunno, that would obviously be a big plus for automatic testing. If so, it seems perf can do that: https://perf.wiki.kernel.org/index.php/Tutorial Sounds worth a look. Thanks!
Re: Is there a way to benchmark/profile portably?
On Thursday, 7 May 2020 at 11:06:17 UTC, Dennis wrote: You can make a reference program that you use to get a measure for how fast the computer is that you run the benchmark on. Then you can use that to scale your actual benchmark results. When testing regressions there's a fairly obvious choice for this reference program: the old version. You can compare those results with the new version and report the relative difference. I have to keep that in mind, but I'd prefer something that does not require keeping old sources working, or floating their binaries around.
Re: Is there a way to benchmark/profile portably?
On Thursday, 7 May 2020 at 10:21:07 UTC, Dukc wrote: Is there some way to measure the performance of a function so that the results will be same in different computers (all x86, but otherwise different processors)? I'm thinking of making a test suite that could find performance regressions automatically. You can make a reference program that you use to get a measure for how fast the computer is that you run the benchmark on. Then you can use that to scale your actual benchmark results. When testing regressions there's a fairly obvious choice for this reference program: the old version. You can compare those results with the new version and report the relative difference.
Re: Is there a way to benchmark/profile portably?
On Thursday, 7 May 2020 at 10:21:07 UTC, Dukc wrote: Is there some way to measure the performance of a function so that the results will be same in different computers (all x86, but otherwise different processors)? I'm thinking of making a test suite that could find performance regressions automatically. I figured out Bochs[1] could be used for that, but it seems an overkill for me to install a whole virtual operating system just to benchmark functions. Does anybody know more lightweight way? [1] http://bochs.sourceforge.net/ If I understand correctly, you want to measure how many cycles pass, rather than clock time? If so, it seems perf can do that: https://perf.wiki.kernel.org/index.php/Tutorial -- Simen
Is there a way to benchmark/profile portably?
Is there some way to measure the performance of a function so that the results will be same in different computers (all x86, but otherwise different processors)? I'm thinking of making a test suite that could find performance regressions automatically. I figured out Bochs[1] could be used for that, but it seems an overkill for me to install a whole virtual operating system just to benchmark functions. Does anybody know more lightweight way? [1] http://bochs.sourceforge.net/
How to use -profile when I initialize D-runtime myself?
Hi, I had a related question in the "Does -profile need the D runtime?" thread. Since D runtime is required for profile to work, the question is how can I use profile when initializing it myself? -- Robert M. Münch http://www.saphirion.com smarter | better | faster
Issue with --build=profile
I'm working on a rather big project which I'm compiling with DMD/LDC. The thing is when I'm trying to profile it, using `dub build --build=profile`, weird issues keep popping up, coming from the project's dependencies... eg. "Warning: statement is not reachable" from the dyaml package. And then it crashes with "/Library/D/dmd/bin/dmd failed with exit code 1." How could I deal with this? (In case it matters, I could also point you to my project sources, although I don't think it has anything to do with my project per-se)
Re: Does -profile need the D runtime?
On 2019-05-18 17:46:52 +, Stefan Koch said:
On Saturday, 18 May 2019 at 16:35:44 UTC, Robert M. Münch wrote:
I want to profile my windows app which has a WinMain(). One of the
first statements in WinMain() within a try{} is:
Runtime.initialize();
But when I compile my app with -profile, it crashes on entry of
WinMain(). Looks like this function is instrumented with code, that
might need the D runtime. Is this the case?
Is there a way to flag specific functions as "don't profile"?
yes profiling does relay on druntime being present.
There is currently no such thing as "don't profile"
Ok, thanks.
So, how to best do it when compiling a windows program? I read that
with WinMain() it's necessary to start the druntime manually. Which is
to late when using the profiler. Or am I missing something?
--
Robert M. Münch
http://www.saphirion.com
smarter | better | faster
Re: Does -profile need the D runtime?
On Saturday, 18 May 2019 at 16:35:44 UTC, Robert M. Münch wrote:
I want to profile my windows app which has a WinMain(). One of
the first statements in WinMain() within a try{} is:
Runtime.initialize();
But when I compile my app with -profile, it crashes on entry of
WinMain(). Looks like this function is instrumented with code,
that might need the D runtime. Is this the case?
Is there a way to flag specific functions as "don't profile"?
yes profiling does relay on druntime being present.
There is currently no such thing as "don't profile"
Does -profile need the D runtime?
I want to profile my windows app which has a WinMain(). One of the
first statements in WinMain() within a try{} is:
Runtime.initialize();
But when I compile my app with -profile, it crashes on entry of
WinMain(). Looks like this function is instrumented with code, that
might need the D runtime. Is this the case?
Is there a way to flag specific functions as "don't profile"?
--
Robert M. Münch
http://www.saphirion.com
smarter | better | faster
Re: Aliasing a mixin (or alternative ways to profile a scope)
On Saturday, 9 March 2019 at 09:12:13 UTC, Dennis wrote: On Friday, 8 March 2019 at 11:42:11 UTC, Simon wrote: Thanks, this works flawlessly. Out of interest: what is the "enum" doing there? I had the exact same behaviour in a function before, that I only called at compile-time, so why did it complain then? Can I somehow tell the compiler that a function should only be available at compile-time? The enum (which in D is not only for enumerations, but also for "manifest constants") ensures it is evaluated at compile-time, since things are only evaluated at compile-time if they have to. See also Adam D. Ruppe's post in this thread: https://forum.dlang.org/post/blaawtdhljjantvga...@forum.dlang.org Thanks!
Re: Aliasing a mixin (or alternative ways to profile a scope)
On Friday, 8 March 2019 at 11:42:11 UTC, Simon wrote: Thanks, this works flawlessly. Out of interest: what is the "enum" doing there? I had the exact same behaviour in a function before, that I only called at compile-time, so why did it complain then? Can I somehow tell the compiler that a function should only be available at compile-time? The enum (which in D is not only for enumerations, but also for "manifest constants") ensures it is evaluated at compile-time, since things are only evaluated at compile-time if they have to. See also Adam D. Ruppe's post in this thread: https://forum.dlang.org/post/blaawtdhljjantvga...@forum.dlang.org
Re: Aliasing a mixin (or alternative ways to profile a scope)
On Thursday, 7 March 2019 at 21:50:17 UTC, Johannes Loher wrote:
```
enum profile_scope(string name) = "import core.stdc.stdio :
printf;
printf(\""
~ name ~ "\n\"); scope(exit) printf(\"" ~ name ~ "\n\");";
extern (C) void main()
{
mixin(profile_scope!"func1");
}
```
This uses string concatenation only at compile time and not
during run
time, so it does not require the garbage collector and is
compatible
with betterC :)
Thanks, this works flawlessly. Out of interest: what is the
"enum" doing there? I had the exact same behaviour in a function
before, that I only called at compile-time, so why did it
complain then? Can I somehow tell the compiler that a function
should only be available at compile-time?
Re: Aliasing a mixin (or alternative ways to profile a scope)
Am 07.03.19 um 22:50 schrieb Johannes Loher:
> [...]
As a small addition, if you always want to pass the function name as a
parameter, you can simplify this to the following:
```
enum profile_scope(string name = __FUNCTION__) = "import core.stdc.stdio
: printf; printf(\""
~ name ~ "\n\"); scope(exit) printf(\"" ~ name ~ "\n\");";
extern (C) void main()
{
mixin(profile_scope!());
foo();
}
extern (C) void foo() {
mixin(profile_scope!());
}
```
This will print
```
main.main
main.foo
main.foo
main.main
```
to the commandline.
Re: Aliasing a mixin (or alternative ways to profile a scope)
Am 07.03.19 um 22:21 schrieb Simon:
>
> Is there a way to achieve this while compiling with -betterC? I use a
> custom string struct right now, and your version needs TypeInfo,
> concatening using ~ needs the garbage collector. I have the feeling D is
> really not agreeing with the way I want to do things. If this is not
> possible, I will just roll with the Adam's struct solution.
>
Using Adams struct solution is perfectly fine. I believe it is probably
the cleaner solution overall.
My solution depends on TypeInfo because format does. I used format
because it makes the string more readable. You can avoid this and use
string concatenation in combination with providing the function name as
template parameter instead:
```
enum profile_scope(string name) = "import core.stdc.stdio : printf;
printf(\""
~ name ~ "\n\"); scope(exit) printf(\"" ~ name ~ "\n\");";
extern (C) void main()
{
mixin(profile_scope!"func1");
}
```
This uses string concatenation only at compile time and not during run
time, so it does not require the garbage collector and is compatible
with betterC :)
Re: Aliasing a mixin (or alternative ways to profile a scope)
On Thursday, 7 March 2019 at 20:34:48 UTC, Johannes Loher wrote:
auto profile_scope(string name)
{
import std.format : format;
return q{import std.stdio : writeln; writeln("%1$s");
scope(exit)
writeln("%1$s");}.format(name);
}
void main()
{
mixin(profile_scope("func1"));
}
Is there a way to achieve this while compiling with -betterC? I
use a custom string struct right now, and your version needs
TypeInfo, concatening using ~ needs the garbage collector. I have
the feeling D is really not agreeing with the way I want to do
things. If this is not possible, I will just roll with the Adam's
struct solution.
Re: Aliasing a mixin (or alternative ways to profile a scope)
Am 07.03.19 um 21:07 schrieb Simon:
> Hello,
>
> I am currently porting the frontend of my instrumenting profiler to D.
> It features a C++-makro that profiles the time between entering and
> leaving a scope (achieved with con-/destructors in C++). Since D has
> scopeguards, I hoped to achieve this by creating a mixin that generates
> the following code:
>
> measure("func1");
> scope(exit) measure("func1");
>
> Since I of course don't want to type a mixin for that everytime I want
> to use it, I tried to alias it:
>
> import std.meta : Alias;
> alias profile_scope(string name) = Alias!(mixin("measure(\"" ~ name ~
> "\"); scope(exit) measure(\"" ~ name ~ "\");"));
>
> I expected this to generate the code above by typing
> profile_scope("func1") in my programm. However the compiler (dmd) gives
> me the following error:
>
> source\main.d(24): Error: template plattform.profile_scope cannot deduce
> function from argument types !()(string), candidates are:
> source\plattform.d(262): plattform.profile_scope(string name)
>
> This looks like a deduction/overload kind of error, which has me
> completly confused since there is no other identifier of the name
> "profile_scope" in my programm and the error message shows only one
> candidate.
>
> Is there any way I can achive generating those two statements using only
> something that effectively looks like a function call/C-macro with an
> argument?
The error you are seeing is due to the fact that you pass "func1" as
runtime parameter instead of as a teamplate parameter. You would need to
do `profile_scope!("func1");`. However, this will still not work and
you'll get an error message similar to this:
```
main.d(4): Error: incomplete mixin expression writeln("func1");
scope(exit) writeln("func1");
main.d(8): Error: template instance `main.profile_scope!"func1"` error
instantiating
```
The underlying problem is that you can only alias expressions, i.e. you
cannot use alias to generate macro like functionality.
`measure("func1"); scope(exit) measure("func1");` is not an expression,
it is simply a statement, so the alias cannot work.
If you want to inject code directly at some place, basically your only
option is to use a string mixin. I understand that you do not want to
write `mixin` all the time, but to me, this does not look that bad:
```
auto profile_scope(string name)
{
import std.format : format;
return q{import std.stdio : writeln; writeln("%1$s"); scope(exit)
writeln("%1$s");}.format(name);
}
void main()
{
mixin(profile_scope("func1"));
}
```
(replace writeln with your appropriate function call, "measure" in your
example).
Re: Aliasing a mixin (or alternative ways to profile a scope)
On Thursday, 7 March 2019 at 20:07:27 UTC, Simon wrote:
measure("func1");
scope(exit) measure("func1");
I would suggest just using a struct. Make its constructor do the
first measure, and its destructor do the second measure.
I betcha you can avoid mixin entirely.
Aliasing a mixin (or alternative ways to profile a scope)
Hello,
I am currently porting the frontend of my instrumenting profiler
to D. It features a C++-makro that profiles the time between
entering and leaving a scope (achieved with con-/destructors in
C++). Since D has scopeguards, I hoped to achieve this by
creating a mixin that generates the following code:
measure("func1");
scope(exit) measure("func1");
Since I of course don't want to type a mixin for that everytime I
want to use it, I tried to alias it:
import std.meta : Alias;
alias profile_scope(string name) = Alias!(mixin("measure(\"" ~
name ~ "\"); scope(exit) measure(\"" ~ name ~ "\");"));
I expected this to generate the code above by typing
profile_scope("func1") in my programm. However the compiler (dmd)
gives me the following error:
source\main.d(24): Error: template plattform.profile_scope cannot
deduce function from argument types !()(string), candidates are:
source\plattform.d(262):plattform.profile_scope(string
name)
This looks like a deduction/overload kind of error, which has me
completly confused since there is no other identifier of the name
"profile_scope" in my programm and the error message shows only
one candidate.
Is there any way I can achive generating those two statements
using only something that effectively looks like a function
call/C-macro with an argument?
Re: Understanding gc memory profile report
Although I responded below, I'm curious on what others think about this
question.
On 09/08/2017 02:13 AM, John Burton wrote:
> I wrote this simple program to test out my understanding of memory
> allocation :-
I changed it to display the location and capacity:
import std.stdio;
void info(A)(A array) {
writefln("ptr: %s, capacity: %s, length: %s",
array.ptr, array.capacity, array.length);
}
void main()
{
int [] array = new int[250];
info(array);
// Append one value to the array
array ~= 123;
info(array);
}
Sample output:
ptr: 7FD9B63CD000, capacity: 255, length: 250
ptr: 7FD9B63CD000, capacity: 255, length: 251
> I compiled it with 'dmd test.d -profile=gc'
>
> After running it, the profile report was :-
>
> bytes allocated, allocations, type, function, file:line
>2000 1 int[] D main test.d:5
> 4 1 int[] D main test.d:10
Here's my output with v2.076.0-dirty (that really is the version :o)):
4000 1 int[] D main deneme.d:412
72 1 std.format.FormatException
std.exception.enforceEx!(FormatException).enforceEx!bool.enforceEx
/usr/include/dmd/phobos/std/exception.d:615
64 2 std.array.Appender!string.Appender.Data
std.array.Appender!string.Appender.this
/usr/include/dmd/phobos/std/array.d:2817
32 1
std.array.Appender!(char[]).Appender.Data
std.array.Appender!(char[]).Appender.this
/usr/include/dmd/phobos/std/array.d:2817
4 1 int[] D main deneme.d:417
And why is mine 4000 bytes? I guess it's the size of the underlying buffer.
> This is not how I expected it to be. I would have expected that the
> runtime either did not have to allocate at all at line 10 to add a new
> element because there was already space or it would have to allocate
> space for the new enlarged array and copy the array to it, in which case
> I'd expect it to allocate 2004 bytes (or more) to copy the enlarged
> array in to.
>
> I would not expect that it could have allocated 4 bytes to add an item
> separably from the original 2000. Is there some way that the runtime can
> grown the original allocation by 4 bytes and that's what I'm seeing? If
> so, is there a limit to how much it can do this?
>
> Can anyone help me understand what is going on here?
I think you're right. Perhaps more accurate information is not possible
with the current implementation. Perhaps a higher-layer function counts
"potential" allocations without being sure whether the space could be
used for this slice or not. The "D Slices" article has related information:
https://dlang.org/d-array-article.html
Ali
Understanding gc memory profile report
I wrote this simple program to test out my understanding of
memory allocation :-
import std.stdio;
void main()
{
int [] array = new int[250];
writeln(array.length, " elements ", array);
// Append one value to the array
array ~= 123;
writeln(array.length, " elements ", array);
}
I compiled it with 'dmd test.d -profile=gc'
After running it, the profile report was :-
bytes allocated, allocations, type, function, file:line
2000 1 int[] D main test.d:5
4 1 int[] D main test.d:10
This is not how I expected it to be. I would have expected that
the runtime either did not have to allocate at all at line 10 to
add a new element because there was already space or it would
have to allocate space for the new enlarged array and copy the
array to it, in which case I'd expect it to allocate 2004 bytes
(or more) to copy the enlarged array in to.
I would not expect that it could have allocated 4 bytes to add an
item separably from the original 2000. Is there some way that the
runtime can grown the original allocation by 4 bytes and that's
what I'm seeing? If so, is there a limit to how much it can do
this?
Can anyone help me understand what is going on here?
Re: Can compiler profile itself?
On Wednesday, 25 January 2017 at 12:49:05 UTC, Stefan Koch wrote: On Tuesday, 24 January 2017 at 23:25:12 UTC, Profile Anaysis wrote: I am trying to compile some code and it takes around 6 seconds. Even if I change one line in one module, it takes the same time. There are about 20 different d modules. [...] yes the compiler can be used to profile itself. build it with make ENABLE_PROFILE=1 If your code is overly slow you are probably using recursive variaidic templates. Would it not be easier to have this feature as a command line argument that enables it in the compiler or include a compiler version that does this? Not all everyone is good at building from sources...
Re: Can compiler profile itself?
On Tuesday, 24 January 2017 at 23:25:12 UTC, Profile Anaysis wrote: I am trying to compile some code and it takes around 6 seconds. Even if I change one line in one module, it takes the same time. There are about 20 different d modules. [...] yes the compiler can be used to profile itself. build it with make ENABLE_PROFILE=1 If your code is overly slow you are probably using recursive variaidic templates.
Can compiler profile itself?
I am trying to compile some code and it takes around 6 seconds.
Even if I change one line in one module, it takes the same time.
There are about 20 different d modules.
I used to get around 1-2 second compile times several months ago
on different projects.
I did upgrade a few things and it seems that something has made
things slow.
Can we ask the compiler what seems to be taking the most time(a
module, a template, etc...)?
If the compiler can profile itself then it would be very easy to
figure out what is slowing things down in compilation.
I feel that I should be able to compile all the modules
separately and only have to recompile the module that I
changed... and that it should be quite fast(e.g., if the module
contains nothing but the main and a write statement, regardless
of all the other stuff in the project, I'd expect that to be
fast).
For example, If I do the
import std.stdio;
int main(string[] argv)
{
writeln("Hello D-World!");
return 0;
}
it takes <1s.
Yet if I import a bunch of modules it takes 6+s and the time
never reduces on subsequent builds.
I am using Visual D.
Visual D has a separate compile and link feature but when I try
it for dmdx64 I get
Internal error: backend\mscoffobj.c 2016
It does work for x32 but the speed does not change.
I'm convinved that dmd is recompiling everything each time... and
this doesn't seem very fast IMO when less than 0.1% of the
project has changed.
Profile ldc2
I like how dmd has the -profile switch, but I'm not sure how to go about profiling ldc2. I don't see any compiler flags in the -help. I'm also using Windows, if that matters, but if the best options are on Linux, I can dual-boot.
Re: Profile Ouput
On Tuesday, 14 July 2015 at 13:35:40 UTC, Mike Parker wrote: -- 4000 _Dmain _D4main6selectFZk 40002736471 4000 _D3std6random27__T7uniformVAyaa2_5b29TkTkZ7uniformFNfkkZk -- Got it now. The 2736 and 471 are the call tree time and function call time, respectively, in *ticks*. They are converted to seconds then displayed as milliseconds int the table at the bottom of the file.
Re: Profile Ouput
On Wednesday, 15 July 2015 at 18:02:11 UTC, jmh530 wrote: I've been confused by this too. The only thing I can find is this http://www.digitalmars.com/ctg/trace.html I think it would be cool to write something that takes the output and puts it in a prettier format. Yeah, I eventually stumbled on to that, but it unfortunately tells me nothing I don't already know. I suppose I'll just email Walter and ask him.
Re: Profile Ouput
On Wednesday, 15 July 2015 at 11:47:53 UTC, Mike Parker wrote: On Tuesday, 14 July 2015 at 13:35:40 UTC, Mike Parker wrote: -- 4000 _Dmain _D4main6selectFZk 40002736471 4000 _D3std6random27__T7uniformVAyaa2_5b29TkTkZ7uniformFNfkkZk -- OK, I've finally realized that the top part of trace.log is an inverted (with main at the bottom) call tree. I've got everything now except that 2736 and 471 on the second line. Anyone? I've been confused by this too. The only thing I can find is this http://www.digitalmars.com/ctg/trace.html I think it would be cool to write something that takes the output and puts it in a prettier format.
Re: Profile Ouput
On Tuesday, 14 July 2015 at 13:35:40 UTC, Mike Parker wrote: -- 4000 _Dmain _D4main6selectFZk 40002736471 4000 _D3std6random27__T7uniformVAyaa2_5b29TkTkZ7uniformFNfkkZk -- OK, I've finally realized that the top part of trace.log is an inverted (with main at the bottom) call tree. I've got everything now except that 2736 and 471 on the second line. Anyone?
Profile Ouput
Perhaps my googlefu has failed me, but I can find anything describing the output of DMD's -profile switch. I get that trace.def is a list of all functions called and I understand the table at the bottom of trace.log. What I'm not sure about is everything above the table in trace.log, stuff like this: -- 4000 _Dmain _D4main6selectFZk 40002736471 4000 _D3std6random27__T7uniformVAyaa2_5b29TkTkZ7uniformFNfkkZk -- I get that it's telling me that the function main.select is called 4000 times and that it, in turn, calls std.random.uniform 4000 times. What I not clear about is the 2736 and the 471. They match none of the times int the table below, where main.select has a Tree Time of 823 and a Func Time of 141. I also don't understand the 4000 next to _Dmain, as it's only called once. If anyone could clarify, or point me somewhere that explains it, I'd appreciate it.
Re: core.atomic and -profile switch
Saurabh Das: I see. I wasn't sure - hence I asked. https://d.puremagic.com/issues/show_bug.cgi?id=11471 In general, how do I check if this is a known issue? Search in Bugzilla and/or ask to people. Bye, bearophile
Re: core.atomic and -profile switch
I see. I wasn't sure - hence I asked. In general, how do I check if this is a known issue? Thanks, Saurabh On Wednesday, 2 April 2014 at 11:19:16 UTC, bearophile wrote: Saurabh Das: With error: /usr/include/dmd/druntime/import/core/atomic.d(910): Error: asm statements are assumed to throw Is there a workaround for this? I have a decent sized codebase which I wish to profile for hotspots - but it won't compile with '-profile'. Isn't this a recent regression? Bye, bearophile
Re: core.atomic and -profile switch
Saurabh Das: With error: /usr/include/dmd/druntime/import/core/atomic.d(910): Error: asm statements are assumed to throw Is there a workaround for this? I have a decent sized codebase which I wish to profile for hotspots - but it won't compile with '-profile'. Isn't this a recent regression? Bye, bearophile
core.atomic and -profile switch
Hello
For this test program ('test.d'):
import core.atomic;
int func1(shared int a)
{
return atomicLoad(a);
}
These invocations of dmd succeed:
1. dmd -main test.d
2. dmd -main -debug test.d
3. dmd -main -release test.d
But this one fails:
dmd -main -profile test.d
With error:
/usr/include/dmd/druntime/import/core/atomic.d(910): Error: asm
statements are assumed to throw
Is there a workaround for this? I have a decent sized codebase
which I wish to profile for hotspots - but it won't compile with
'-profile'.
Warm Regards,
Saurabh
Re: Strange behaviour with -profile
On Sunday, 23 February 2014 at 20:46:47 UTC, Jesse Phillips wrote: On Sunday, 23 February 2014 at 08:32:50 UTC, Danny Arends wrote: I have a medium sized project at: https://github.com/DannyArends/DaNode it compiles fine with the normal build switches I used: $ rdmd --build-only -O -gc -release -w danode/server.d if I compile with the -profile switch, I get a weird error: $ rdmd --build-only -O -gc -release -profile -w danode/server.d "Warning: statement is not reachable" I thought DMD provided an error if you tried to optimize with debug symbols. -O should not be used with -gc, -g, or -profile Maybe that will address the problem. While I think my advice is still correct, I can't seem to find the combination which caused a compiler message: dmd -g -O -profile -release -debug -inline test.d Also found that dub will do the same odd mix of switches: ["profile", "optimize", "inline", "debugInfo"] Debug and optimize just don't make sense together, and profile needs debug symbols I thought.
Re: Strange behaviour with -profile
On Sunday, 23 February 2014 at 08:32:50 UTC, Danny Arends wrote: I have a medium sized project at: https://github.com/DannyArends/DaNode it compiles fine with the normal build switches I used: $ rdmd --build-only -O -gc -release -w danode/server.d if I compile with the -profile switch, I get a weird error: $ rdmd --build-only -O -gc -release -profile -w danode/server.d "Warning: statement is not reachable" I thought DMD provided an error if you tried to optimize with debug symbols. -O should not be used with -gc, -g, or -profile Maybe that will address the problem.
Strange behaviour with -profile
I have a medium sized project at: https://github.com/DannyArends/DaNode it compiles fine with the normal build switches I used: $ rdmd --build-only -O -gc -release -w danode/server.d if I compile with the -profile switch, I get a weird error: $ rdmd --build-only -O -gc -release -profile -w danode/server.d "Warning: statement is not reachable" No line number or anything to point me to a solution The weird thing is, I'm compiling using also with the -w switch, so in theory this should break compilation at the first warning, but the executable is created, and runs. I am using DMD64 D Compiler v2.064, so it could be that the newest betas would fix this issue. But I haven gotten around to testing those, and this current build 2.064 works fine as a semi production build, so updating might also 'degrade' stability, and I have to find time to test it. So I rather fix this, any suggestions ? Danny Arends http://www.dannyarends.nl
Re: Error when trying to compile with --profile.
31-Oct-2013 13:28, bearophile пишет: Wolftein: Using --debug or --release works fine. But when i use --profile i get: c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2187): Error: balancedParens is not nothrow c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2188): Error: balancedParens is not nothrow Error: this.__invariant is not nothrow Error: this.__invariant is not nothrow Is this bug in Bugzilla? DMD 2.064 should not come out with this problem. It would unless the above is a regression. The release can't be stalled forever. Do you have a small test case? Bye, bearophile -- Dmitry Olshansky
Re: Error when trying to compile with --profile.
Wolftein: Using --debug or --release works fine. But when i use --profile i get: c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2187): Error: balancedParens is not nothrow c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2188): Error: balancedParens is not nothrow Error: this.__invariant is not nothrow Error: this.__invariant is not nothrow Is this bug in Bugzilla? DMD 2.064 should not come out with this problem. Do you have a small test case? Bye, bearophile
Error when trying to compile with --profile.
Using --debug or --release works fine. But when i use --profile i get: c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2187): Error: balancedParens is not nothrow c:\Program Files (x86)\DLang\dmd2\windows\bin\..\..\src\phobos\std\path.d(2188): Error: balancedParens is not nothrow Error: this.__invariant is not nothrow Error: this.__invariant is not nothrow
Re: -profile and nothrow
On Monday, 28 October 2013 at 18:34:51 UTC, Dicebot wrote: Any difference in semantics between -profile and normal flow is a bug, so, yes, please report to bugzilla. Done: http://d.puremagic.com/issues/show_bug.cgi?id=11375
Re: -profile and nothrow
Any difference in semantics between -profile and normal flow is a bug, so, yes, please report to bugzilla.
Re: -profile and nothrow
Reduced:
module b;
class B {
this() {}
}
module a;
import b;
class C : B {}
class D(X) : X {}
alias D!(C) HtmlPanel;
dmd -profile a.d
Error: 'a.C.this' is not no throw
Everything is ok without the profile... may I post a bug report
for this?
/Paolo
On Monday, 28 October 2013 at 15:20:59 UTC, Paolo Invernizzi
wrote:
Hi all,
Compiling my application with 2.064.beta.4 and 2.063.2 with the
"-profile" switch, I have an error like:
Error: 'foo.ui.Widgets.this' is not no throw
I've read something similar involving contracts, but this seems
something different.
Of course everything compiles and run fine without the
"-profile" switch.
Any clue, or may I try to reduce (it is a BIG codebase)?
Thanks,
Paolo
-profile and nothrow
Hi all, Compiling my application with 2.064.beta.4 and 2.063.2 with the "-profile" switch, I have an error like: Error: 'foo.ui.Widgets.this' is not no throw I've read something similar involving contracts, but this seems something different. Of course everything compiles and run fine without the "-profile" switch. Any clue, or may I try to reduce (it is a BIG codebase)? Thanks, Paolo
Re: How do you profile your apps under windows?
10.08.2013 01:30, Gary Willoughby пишет: On Friday, 9 August 2013 at 16:39:41 UTC, Alexandr Druzhinin wrote: -profile switch doesn't work for me (nothing happens), so I'm curious how to profile? I had the same problem on Linux where -profile didn't seem to produce anything. I had a program which ran an infinite loop running as a daemon. When compiled and run using -profile i'd let it run for a bit then kill it and nothing was produced. After much head scratching i implemented a timer in the program that only ran in debug mode and after the timer expired i raised an assert fail: assert(false); This exited the program and produced the trace file which i could then examine. I guess killing the process also killed the profiler and so never gave me the report. Tried this approach but the app hangs up somewhere in druntime code after my code execution finished.
Re: How do you profile your apps under windows?
10.08.2013 00:59, H. S. Teoh пишет:
On Fri, Aug 09, 2013 at 11:39:41PM +0700, Alexandr Druzhinin wrote:
-profile switch doesn't work for me (nothing happens), so I'm
curious how to profile?
You've to run your program first. It will create a bunch of extra files
in your current working directory that contains the profile results.
T
Of course I run the program first. When the program was simpler it
worked fine. But at the some moment I tried to profile it again but
there were no profile results generated. More over now my app just hangs
up after finishing (my code) execution somewhere in druntime kingdom.
To try gprof I installed ubuntu 12.04 (newer versions didn't install
from flushdisk) and also installed gdc-4.8 ("native" gdc-4.6 doesn't
compile my code) from a package (compiling gdc from source fails,
playing with configure options just wasted several hours of my spare
time) and even build my app and became exciting that now I would get
opportunity to profile using gprof but my app just crashed with dumping
without any informative message. I tried to use gdb (I'm a noob with it)
and found that it crashed somewhere about exit from static ctors - I'm
not sure I interpreted gdb output correctly.
So now I have a choice - to wait when good and smart guys solve these
problems either with dmd -profile or gdc code generating and try to work
around myself - probably gdc codegen isn't the real reason. Here I'm
sitting and making decision what to do further...
But nevertheless thanks to all who made the universe more beautiful and
handy by means of D language and its infrastructure! Thank you guys!
Re: How do you profile your apps under windows?
The Very Sleepy profiler works ok (if you run the apps through cv2pdb to get the debug symbols in the right format at least)
Re: How do you profile your apps under windows?
On Friday, 9 August 2013 at 16:39:41 UTC, Alexandr Druzhinin wrote: -profile switch doesn't work for me (nothing happens), so I'm curious how to profile? I had the same problem on Linux where -profile didn't seem to produce anything. I had a program which ran an infinite loop running as a daemon. When compiled and run using -profile i'd let it run for a bit then kill it and nothing was produced. After much head scratching i implemented a timer in the program that only ran in debug mode and after the timer expired i raised an assert fail: assert(false); This exited the program and produced the trace file which i could then examine. I guess killing the process also killed the profiler and so never gave me the report.
Re: How do you profile your apps under windows?
09-Aug-2013 20:39, Alexandr Druzhinin пишет: -profile switch doesn't work for me (nothing happens), so I'm curious how to profile? My take is to compile with -g and use say AMD CodeAnalyst (or Intel's analog). Works quite well even though all symbols are mangled. With -profile the app usually outputs trace.log somewhere in the folder launched. -- Dmitry Olshansky
Re: How do you profile your apps under windows?
On Fri, Aug 09, 2013 at 11:39:41PM +0700, Alexandr Druzhinin wrote: > -profile switch doesn't work for me (nothing happens), so I'm > curious how to profile? You've to run your program first. It will create a bunch of extra files in your current working directory that contains the profile results. T -- A bend in the road is not the end of the road unless you fail to make the turn. -- Brian White
How do you profile your apps under windows?
-profile switch doesn't work for me (nothing happens), so I'm curious how to profile?
how to profile compilation steps?
I'd like to at least get timestamps for the various steps appearing the the dmd -v output. Is there an easy way to modify dmd to output that? That could be fed to a D function that parses it and generates a nice profile report (ie the compiler would do not much more than just adding timestamps, leaving most of the details to a D function)
Re: -profile
Anthony Goins: What is the error here :) 1. dmd profiletest.d should fail because balancedParens is not nothrow even though it is called from a contract. 2. dmd -profile profiletest.d should succeed because nothrow, @safe, pure don't apply to contracts I think -profile and pre-conditions should not be correlated. That's why I have reported this bug as a dmd bug instead just of a Phobos bug. I am not sure what's the right answer. I think the right answer is 1. Bye, bearophile
Re: -profile
Anthony Goins: Is this even worth a bug report? http://d.puremagic.com/issues/show_bug.cgi?id=10295 Bye, bearophile
Re: -profile
On Friday, 7 June 2013 at 23:57:37 UTC, bearophile wrote: Anthony Goins: Is this even worth a bug report? Yes, it's worth a bug report in Bugzilla, it's a (small) Phobos Bug. I confirmed its presence. Bye, bearophile Thank you for all your help. Not just here but throughout the forums. One more question though... What is the error here :) 1. dmd profiletest.d should fail because balancedParens is not nothrow even though it is called from a contract. 2. dmd -profile profiletest.d should succeed because nothrow, @safe, pure don't apply to contracts
Re: -profile
Anthony Goins: Is this even worth a bug report? Yes, it's worth a bug report in Bugzilla, it's a (small) Phobos Bug. I confirmed its presence. Bye, bearophile
Re: -profile
On Friday, 7 June 2013 at 23:42:58 UTC, Anthony Goins wrote:
On Friday, 7 June 2013 at 21:39:26 UTC, bearophile wrote:
Anthony Goins:
Simple question is why?
I am compiling this code with -profile, and I don't see errors:
import std.string: balancedParens;
void main() {
assert(balancedParens("[[]]", '[', ']'));
}
Bye,
bearophile
okay found it
module profiletest;
import std.path;
void main()
{
assert(globMatch("foo.bar", "*"));
}
dmd -profile profiletest.d
/usr/include/dmd/phobos/std/path.d(2187): Error: balancedParens
is not nothrow
/usr/include/dmd/phobos/std/path.d(2188): Error: balancedParens
is not nothrow
globMatch is @safe pure nothrow
balancedParens is not nothrow
-release ignores contracts, right?
oops forgot to say
balancedParens is called from the 'in' contract of globMatch.
Is this even worth a bug report?
Re: -profile
On Friday, 7 June 2013 at 21:39:26 UTC, bearophile wrote:
Anthony Goins:
Simple question is why?
I am compiling this code with -profile, and I don't see errors:
import std.string: balancedParens;
void main() {
assert(balancedParens("[[]]", '[', ']'));
}
Bye,
bearophile
okay found it
module profiletest;
import std.path;
void main()
{
assert(globMatch("foo.bar", "*"));
}
dmd -profile profiletest.d
/usr/include/dmd/phobos/std/path.d(2187): Error: balancedParens
is not nothrow
/usr/include/dmd/phobos/std/path.d(2188): Error: balancedParens
is not nothrow
globMatch is @safe pure nothrow
balancedParens is not nothrow
-release ignores contracts, right?
Re: -profile
Anthony Goins:
Simple question is why?
I am compiling this code with -profile, and I don't see errors:
import std.string: balancedParens;
void main() {
assert(balancedParens("[[]]", '[', ']'));
}
Bye,
bearophile
Re: -profile
Anthony Goins: Simple question is why? Looks like a little compiler bug. Please create a minimal example that shows the problem. If it's a real problem then it's for Bugzilla. Bye, bearophile
-profile
Compiling with -profile and without -release Gives the following error (not exactly but the best I can recall) balancedParens is not nothrow safe pure With -release and -profile all is fine. Builds fine without -profile. Simple question is why?
Reading results from dmd -profile
I am trying to read the text file (trace.log) created by running the dmd profiler on some code, so that I can use demangle to make the output a bit more readable. I can read it in as a char[], but the whenever I try any string ops I get an exception because there are invalid UTF-8 sequences in the text. So two questions: why does the output contain invalid UTF-8 sequences, and what would be a good way to handle the invalid sequences, since I guess I need them for proper demangling? Cheers
profiling with -profile
I have this code:
import std.stdio;
void main()
{
f();
g();
}
void f()
{
writeln("f()");
}
void g()
{
writeln("g()");
}
I compile with the command:
dmd test.d -profile
Then I execute it. It prints:
f()
g()
as expected. Shouldn't it print profiling information as well?
Re: profiling with -profile
Sorry, I just saw the generated file... :p
Re: GUI or more human readable -profile data?
On Thursday, March 01, 2012 00:17:55 Robert Clipsham wrote: > Just as a side note, -profile doesn't work with multi-threaded > applications, so using some other profiler would probably be a better > bet anyway. It doesn't work with 64-bit programs either. - Jonathan M Davis
Re: GUI or more human readable -profile data?
On 29/02/2012 21:30, simendsjo wrote: On Wed, 29 Feb 2012 22:28:29 +0100, Robert Clipsham wrote: On 29/02/2012 19:41, simendsjo wrote: http://www.digitalmars.com/ctg/trace.html Has someone made some GUI/pretty printing/dump to database or other tools to make the profile data a bit simpler to digest? If you're on Windows you could try: http://h3.gd/code/xfProf/ Although I don't believe it uses the data from -profile. On linux/OS X you can use valgrind/kcachegrind, though it doesn't have demangling for D symbols (at least, last time I checked). Nope, linux. But I bet the feature is used, so somewhere out there, there's some tools for making it simpler to make sense of :) Just as a side note, -profile doesn't work with multi-threaded applications, so using some other profiler would probably be a better bet anyway. -- Robert http://octarineparrot.com/
Re: GUI or more human readable -profile data?
Boom: https://bitbucket.org/stqn/profiled I've never used it though.
Re: GUI or more human readable -profile data?
On Wed, 29 Feb 2012 22:28:29 +0100, Robert Clipsham wrote: On 29/02/2012 19:41, simendsjo wrote: http://www.digitalmars.com/ctg/trace.html Has someone made some GUI/pretty printing/dump to database or other tools to make the profile data a bit simpler to digest? If you're on Windows you could try: http://h3.gd/code/xfProf/ Although I don't believe it uses the data from -profile. On linux/OS X you can use valgrind/kcachegrind, though it doesn't have demangling for D symbols (at least, last time I checked). Nope, linux. But I bet the feature is used, so somewhere out there, there's some tools for making it simpler to make sense of :)
Re: GUI or more human readable -profile data?
On 29/02/2012 19:41, simendsjo wrote: http://www.digitalmars.com/ctg/trace.html Has someone made some GUI/pretty printing/dump to database or other tools to make the profile data a bit simpler to digest? If you're on Windows you could try: http://h3.gd/code/xfProf/ Although I don't believe it uses the data from -profile. On linux/OS X you can use valgrind/kcachegrind, though it doesn't have demangling for D symbols (at least, last time I checked). -- Robert http://octarineparrot.com/
GUI or more human readable -profile data?
http://www.digitalmars.com/ctg/trace.html Has someone made some GUI/pretty printing/dump to database or other tools to make the profile data a bit simpler to digest?
Re: Profile DMD
"Timon Gehr" wrote in message news:jb8iak$2ucr$1...@digitalmars.com... > On 12/01/2011 07:42 PM, Nick Sabalausky wrote: >> Is there an easy way to profile DMD similar to how DMD itself has >> the -profile switch? >> >> > > Use DMC, I guess. http://www.digitalmars.com/ctg/trace.html Ah, ok, and it looks like DMD's makefile (at least on windows) has a target "trace" that includes the appropriate switches. So... make -f win32.mak trace ...seems to do the trick.
Re: Profile DMD
On 12/01/2011 07:42 PM, Nick Sabalausky wrote: Is there an easy way to profile DMD similar to how DMD itself has the -profile switch? Use DMC, I guess. http://www.digitalmars.com/ctg/trace.html
Profile DMD
Is there an easy way to profile DMD similar to how DMD itself has the -profile switch?
Benchmarking with -profile
How do you guys benchmark your D code? I tried using the -profile option, but was surprised to see wildly varying results (greater than an order of magnitude difference in Func Time) from one run to the next.
Re: -profile and threaded code
BCS wrote: > Hello BCS, > >> I have a program that runs an "easily" parallelizable loop. When I run >> it as a single thread it only takes about 10% longer than 2 threads >> (on a dual-core). I'm trying to track down the lossed time and am >> wondering if turning on -profile is even worth looking at. The concern >> is that it might not be thread safe or might just skew the result so >> much as to be useless. >> > > Well it seems that runing threded code with -profile gives a seg-v. So I > guess that answers that. See last comment of Walter: http://d.puremagic.com/issues/show_bug.cgi?id=971
Re: -profile and threaded code
BCS wrote: > Hello BCS, > >> I have a program that runs an "easily" parallelizable loop. When I run >> it as a single thread it only takes about 10% longer than 2 threads >> (on a dual-core). I'm trying to track down the lossed time and am >> wondering if turning on -profile is even worth looking at. The concern >> is that it might not be thread safe or might just skew the result so >> much as to be useless. >> > > Well it seems that runing threded code with -profile gives a seg-v. So I > guess that answers that. > > Try GDC. :)
Re: -profile and threaded code
Hello BCS, I have a program that runs an "easily" parallelizable loop. When I run it as a single thread it only takes about 10% longer than 2 threads (on a dual-core). I'm trying to track down the lossed time and am wondering if turning on -profile is even worth looking at. The concern is that it might not be thread safe or might just skew the result so much as to be useless. Well it seems that runing threded code with -profile gives a seg-v. So I guess that answers that.
-profile and threaded code
I have a program that runs an "easily" parallelizable loop. When I run it as a single thread it only takes about 10% longer than 2 threads (on a dual-core). I'm trying to track down the lossed time and am wondering if turning on -profile is even worth looking at. The concern is that it might not be thread safe or might just skew the result so much as to be useless.
