Re: LRU cache for ~=
On 2009-10-20 03:41:59 +0200, Walter Bright newshou...@digitalmars.com said: Denis Koroskin wrote: Safe as in SafeD (i.e. no memory corruption) :) Right. The problems with other definitions of safe is they are too ill-defined. no (or as little as possible) undefined behaviour comes to mind. Initialization of values in D follows that. Slice appending does not (and will remain so also with LRU cache). LRU cache improves the current status quo, but is still a cludge: undefined behavious in some corner cases, undefined optimization behaviour... It improves things, but cannot be trusted in general, thus a clean library type is still needed imho. Fawzi
Re: LRU cache for ~=
Andrei Alexandrescu wrote:
Rainer Deyke wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
What's yours?
SafeD is easy to learn and it keeps the programmers away from undefined
behaviors. -- safed.html.
The behaviour you quoted is undefined behaviour, therefore it's not safe
according to the only SafeD definition in the spec.
Re: LRU cache for ~=
On Mon, 19 Oct 2009 14:51:32 -0400, Andrei Alexandrescu seewebsiteforem...@erdani.org wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? This is a very good idea. Incidentally, you only need the upper bound location, the beginning location is irrelevant, since you don't grow down. What do you do in the case where the memory was recycled? Does a GC collection cycle clean out the cache as well? This is better than my two previous ideas. The only drawback I see is if you have many threads doing appending, or you are appending more than 8 arrays at once in a round-robin fashion, you would lose all the benefit (although it shouldn't affect correctness). At that point however, you'd have to ask yourself why you aren't using a specialized appender type or function. In response to other's queries about how many LRUs to use, you'd probably want one per heap, and you'd want to lock/not lock based on whether the heap is thread local or not. -Steve
Re: LRU cache for ~=
On Mon, 19 Oct 2009 22:37:26 -0400, dsimcha dsim...@yahoo.com wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s
article
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s
article
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s
article
dsimcha wrote:
Started playing w/ the implementation a little and I see a
problem. What
about
the garbage collector? There are two possibilities:
[snip]
The only possible solutions I see would be to have the GC know
everything
about
the LRU cache and evict stale entries (probably slows down GC a
lot, a
huge PITA
to implement, couples things that shouldn't be tightly coupled),
or clear the
cache every time GC is run (probably would make appending so slow
as to
defeat the
purpose of having the cache).
I think GC.collect may simply evict the entire cache. The
collection
cycle costs so much, the marginal cost of losing cached
information is
lost in the noise.
Andrei
But then you have to copy the whole array again, likely triggering
another GC if
the array is large. Then things really get ugly as, for all
practical purposes,
you've completely done away with the cache.
This happens whether or not a cache is in use.
Andrei
But the array isn't guaranteed to get reallocated immediately after
*every* GC
run. If you're appending to a huge array, the GC will likely run
several times
while you're appending, leading to several unnecessary reallocations.
I don't think I understand this.
1. Request for an append comes that runs out of memory
2. GC runs and clears memory
3. Array is reallocated and the capacity cached.
No?
This is entirely correct.
Each of
those unnecessary reallocations will increase the memory footprint of
your
program, possibly triggering another GC run and wiping out your cache
again in
short order, until, for sufficiently large arrays,
a ~= b;
is almost equivalent to
a = a ~ b;
I don't understand how the cache makes that all worse.
Andrei
The cache doesn't make anything *worse* than with no cache. The only
point I'm
trying to make is that, for large arrays, if the GC clears the cache
every time it
runs, things would start to get *almost as bad as* having no cache
because the
copy operations become expensive and the GC may run frequently.
The cache can't be cleared every time, or else you might as well only
keep one LRU entry:
int[] twos, threes;
for(int i = 1; i 1; i++)
{
twos ~= i * 2;
threes ~= i * 3;
}
At some point, twos or threes needs an allocation triggering a collection,
and that clears the cache, making the other array need an allocation,
clearing the cache, etc.
I'd think you only want to clear the entries affected by the collection.
-Steve
Re: LRU cache for ~=
Steven Schveighoffer wrote: On Mon, 19 Oct 2009 14:51:32 -0400, Andrei Alexandrescu seewebsiteforem...@erdani.org wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? This is a very good idea. Incidentally, you only need the upper bound location, the beginning location is irrelevant, since you don't grow down. Awesome, didn't think of that. So now more cases are caught: auto a = new int[100]; a ~= 42; a = a[50 .. $]; a ~= 52; That wouldn't have worked with my original suggestion, but it does work safely with yours. What do you do in the case where the memory was recycled? Does a GC collection cycle clean out the cache as well? As you saw, there was some discussion about that as well. This is better than my two previous ideas. The only drawback I see is if you have many threads doing appending, or you are appending more than 8 arrays at once in a round-robin fashion, you would lose all the benefit (although it shouldn't affect correctness). At that point however, you'd have to ask yourself why you aren't using a specialized appender type or function. Yah. As I suspect a lot of code is actually doing round-robin naturally, I'm considering using a random eviction strategy. That way performance will degrade smoother. A more advanced algorithm would use introspection to choose dynamically between LRU and random. Andrei
Re: LRU cache for ~=
On Tue, 20 Oct 2009 10:14:52 -0400, Andrei Alexandrescu seewebsiteforem...@erdani.org wrote: Steven Schveighoffer wrote: On Mon, 19 Oct 2009 14:51:32 -0400, Andrei Alexandrescu seewebsiteforem...@erdani.org wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? This is a very good idea. Incidentally, you only need the upper bound location, the beginning location is irrelevant, since you don't grow down. Awesome, didn't think of that. So now more cases are caught: auto a = new int[100]; a ~= 42; a = a[50 .. $]; a ~= 52; That wouldn't have worked with my original suggestion, but it does work safely with yours. What do you do in the case where the memory was recycled? Does a GC collection cycle clean out the cache as well? As you saw, there was some discussion about that as well. This is better than my two previous ideas. The only drawback I see is if you have many threads doing appending, or you are appending more than 8 arrays at once in a round-robin fashion, you would lose all the benefit (although it shouldn't affect correctness). At that point however, you'd have to ask yourself why you aren't using a specialized appender type or function. Yah. As I suspect a lot of code is actually doing round-robin naturally, I'm considering using a random eviction strategy. That way performance will degrade smoother. A more advanced algorithm would use introspection to choose dynamically between LRU and random. Andrei So you want to synchronize the ~= function? I thought the LRU would be thread local and therefore independent of these issues, as well as being faster. And if the LRU isn't thread-local, then why not make it part of the GC? It would both be more general and much simpler/cleaner to implement.
Re: LRU cache for ~=
On Tue, 20 Oct 2009 10:05:42 -0400, Steven Schveighoffer
schvei...@yahoo.com wrote:
On Mon, 19 Oct 2009 22:37:26 -0400, dsimcha dsim...@yahoo.com wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s
article
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s
article
dsimcha wrote:
== Quote from Andrei Alexandrescu
(seewebsiteforem...@erdani.org)'s article
dsimcha wrote:
Started playing w/ the implementation a little and I see a
problem. What
about
the garbage collector? There are two possibilities:
[snip]
The only possible solutions I see would be to have the GC know
everything
about
the LRU cache and evict stale entries (probably slows down GC a
lot, a
huge PITA
to implement, couples things that shouldn't be tightly coupled),
or clear the
cache every time GC is run (probably would make appending so
slow as to
defeat the
purpose of having the cache).
I think GC.collect may simply evict the entire cache. The
collection
cycle costs so much, the marginal cost of losing cached
information is
lost in the noise.
Andrei
But then you have to copy the whole array again, likely triggering
another GC if
the array is large. Then things really get ugly as, for all
practical purposes,
you've completely done away with the cache.
This happens whether or not a cache is in use.
Andrei
But the array isn't guaranteed to get reallocated immediately after
*every* GC
run. If you're appending to a huge array, the GC will likely run
several times
while you're appending, leading to several unnecessary reallocations.
I don't think I understand this.
1. Request for an append comes that runs out of memory
2. GC runs and clears memory
3. Array is reallocated and the capacity cached.
No?
This is entirely correct.
Each of
those unnecessary reallocations will increase the memory footprint
of your
program, possibly triggering another GC run and wiping out your
cache again in
short order, until, for sufficiently large arrays,
a ~= b;
is almost equivalent to
a = a ~ b;
I don't understand how the cache makes that all worse.
Andrei
The cache doesn't make anything *worse* than with no cache. The only
point I'm
trying to make is that, for large arrays, if the GC clears the cache
every time it
runs, things would start to get *almost as bad as* having no cache
because the
copy operations become expensive and the GC may run frequently.
The cache can't be cleared every time, or else you might as well only
keep one LRU entry:
int[] twos, threes;
for(int i = 1; i 1; i++)
{
twos ~= i * 2;
threes ~= i * 3;
}
At some point, twos or threes needs an allocation triggering a
collection, and that clears the cache, making the other array need an
allocation, clearing the cache, etc.
I'd think you only want to clear the entries affected by the collection.
-Steve
If it was free and simple to only clear the affected entries, sure. But
doing so requires (very heavy?) modification of the GC in order to track
and check changes. It also reduces collection performance. I think, that
if GC allocations added entries to the LRU, and therefore the information
in the LRU is never stale, you could avoid clearing the LRU. But this
requires the LRU to be part of the GC.
Re: LRU cache for ~=
On Tue, 20 Oct 2009 10:37:40 -0400, Robert Jacques sandf...@jhu.edu wrote: So you want to synchronize the ~= function? I thought the LRU would be thread local and therefore independent of these issues, as well as being faster. And if the LRU isn't thread-local, then why not make it part of the GC? It would both be more general and much simpler/cleaner to implement. quoting myself earlier: On Tue, 20 Oct 2009 09:58:01 -0400, Steven Schveighoffer schvei...@yahoo.com wrote: In response to other's queries about how many LRUs to use, you'd probably want one per heap, and you'd want to lock/not lock based on whether the heap is thread local or not. You need a locked operation in the case where the heap is shared, otherwise, you lose safety. At the moment all we *have* is a shared heap. So ~= is a synchronized operation until thread-local heaps are available. I think the only logical place for the LRU is the GC, it makes no sense to have a a shared LRU for an unshared GC or vice versa. -Steve
Re: LRU cache for ~=
On Tue, 20 Oct 2009 10:14:52 -0400, Andrei Alexandrescu seewebsiteforem...@erdani.org wrote: Steven Schveighoffer wrote: This is a very good idea. Incidentally, you only need the upper bound location, the beginning location is irrelevant, since you don't grow down. Awesome, didn't think of that. So now more cases are caught: auto a = new int[100]; a ~= 42; a = a[50 .. $]; a ~= 52; That wouldn't have worked with my original suggestion, but it does work safely with yours. It was one of the coolest parts of my original proposal :) http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.Darticle_id=63146 But using a cache solves a lot of the problems I didn't. What do you do in the case where the memory was recycled? Does a GC collection cycle clean out the cache as well? As you saw, there was some discussion about that as well. Yeah, I'm reading in thread order :) Still got 91 unread messages, so maybe I'll read all before replying again... -Steve
Re: LRU cache for ~=
On Tue, 20 Oct 2009 10:48:31 -0400, Robert Jacques sandf...@jhu.edu wrote: On Tue, 20 Oct 2009 10:05:42 -0400, Steven Schveighoffer schvei...@yahoo.com wrote: I'd think you only want to clear the entries affected by the collection. If it was free and simple to only clear the affected entries, sure. But doing so requires (very heavy?) modification of the GC in order to track and check changes. Why? All you have to do is check whether a block is referenced in the LRU while freeing the block. I don't even think it would be that performance critical. Using my vastly novice assumptions about how the GC collection cycle works: step 1, mark all blocks that are not referenced by any roots. step 2, check which blocks are referenced by the LRU, if they are, then remove them from the LRU. step 3, recycle free blocks. But this requires the LRU to be part of the GC. I think we're already in that boat. If the LRU isn't attached to the GC, then ~= becomes a locking operation even if the GC is thread-local, which makes no sense. -Steve
Re: LRU cache for ~=
On Tue, 20 Oct 2009 11:24:21 -0400, Steven Schveighoffer schvei...@yahoo.com wrote: On Tue, 20 Oct 2009 10:48:31 -0400, Robert Jacques sandf...@jhu.edu wrote: On Tue, 20 Oct 2009 10:05:42 -0400, Steven Schveighoffer schvei...@yahoo.com wrote: I'd think you only want to clear the entries affected by the collection. If it was free and simple to only clear the affected entries, sure. But doing so requires (very heavy?) modification of the GC in order to track and check changes. Why? All you have to do is check whether a block is referenced in the LRU while freeing the block. I don't even think it would be that performance critical. Using my vastly novice assumptions about how the GC collection cycle works: step 1, mark all blocks that are not referenced by any roots. step 2, check which blocks are referenced by the LRU, if they are, then remove them from the LRU. step 3, recycle free blocks. I agree, but my mind hadn't gotten there yet. (It was thinking of the overhead of generational/concurrent collections, for some strange reason) But this requires the LRU to be part of the GC. I think we're already in that boat. If the LRU isn't attached to the GC, then ~= becomes a locking operation even if the GC is thread-local, which makes no sense. -Steve Of course, Andrei just stated the cache should be thread-local (and probably in the function, not the GC) which throws a spanner into the works.
Re: LRU cache for ~=
Brad Roberts wrote: On Mon, 19 Oct 2009, Walter Bright wrote: Denis Koroskin wrote: Safe as in SafeD (i.e. no memory corruption) :) Right. The problems with other definitions of safe is they are too ill-defined. There's SafeD, which has a fairly formal definition. But a fairly generic name, which confuses people repeatedly. I'm not the first to recommend that the name be changed. It does more harm than good.
LRU cache for ~=
I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? Andrei
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? Andrei 1. I assume the LRU cache would be thread-local, so no lock would be necessary? 2. I don't understand how this solves the safety problem: // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. 3. I'm pretty familiar with these parts of druntime and would probably be willing to do the implementation after I understand a few details of it a little better.
Re: LRU cache for ~=
dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? Andrei 1. I assume the LRU cache would be thread-local, so no lock would be necessary? Yes. 2. I don't understand how this solves the safety problem: It's rather subtle. I'd figured it out in the morning and forgot it by the time I explained to Walter. // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. This is actually one of the less subtle cases. Upon the call to ~=, bar is not in the LRU cache so ~= conservatively reallocates it. As a rule of thumb (which takes care of the subtler issues): if a growing slice is not found in the LRU cache, it will always be conservatively reallocated. This is exactly because you don't know whether that slice was reduced from a larger slice. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. 3. I'm pretty familiar with these parts of druntime and would probably be willing to do the implementation after I understand a few details of it a little better. That would be awesome!!! Andrei
Re: LRU cache for ~=
Andrei Alexandrescu, el 19 de octubre a las 13:51 me escribiste: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? I think this is moving in the wrong direction. It's a patch, not a solution. And this makes dynamic arrays more coupled with the way the GC works. I think the GC shouldn't even provide a sizeOf(void*), we should move in that direction, removing restrictions to the GC instead of enforcing the current ones; at lest if the goal is to eventually have a better GC (a copying GC doesn't even have to store the size of the cell, for example). I hope you eventually realize that you are complicating things much more than necessary. -- Leandro Lucarella (AKA luca) http://llucax.com.ar/ -- GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145 104C 949E BFB6 5F5A 8D05) -- ACCIDENTE FATAL EN FLORES, MUEREN DOS PERSONAS Y UN BOLIVIANO -- Crónica TV
Re: LRU cache for ~=
Leandro Lucarella wrote: Andrei Alexandrescu, el 19 de octubre a las 13:51 me escribiste: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? I think this is moving in the wrong direction. It's a patch, not a solution. And this makes dynamic arrays more coupled with the way the GC works. I think the GC shouldn't even provide a sizeOf(void*), we should move in that direction, removing restrictions to the GC instead of enforcing the current ones; at lest if the goal is to eventually have a better GC (a copying GC doesn't even have to store the size of the cell, for example). The LRU doesn't need GC.sizeOf. It could always conservatively reallocate whenever in doubt. I hope you eventually realize that you are complicating things much more than necessary. I actually did a couple of days ago :o). Andrei
Re: LRU cache for ~=
dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? Andrei 1. I assume the LRU cache would be thread-local, so no lock would be necessary? Yes. 2. I don't understand how this solves the safety problem: It's rather subtle. I'd figured it out in the morning and forgot it by the time I explained to Walter. // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Upon the call to ~=, bar is not in the LRU cache so ~= conservatively reallocates it. As a rule of thumb (which takes care of the subtler issues): if a growing slice is not found in the LRU cache, it will always be conservatively reallocated. This is exactly because you don't know whether that slice was reduced from a larger slice. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. 3. I'm pretty familiar with these parts of druntime and would probably be willing to do the implementation after I understand a few details of it a little better. That would be awesome!!! Andrei
Re: LRU cache for ~=
dsimcha wrote: 2. I don't understand how this solves the safety problem: // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. Let me stress a point harder that I think I expressed poorly: The LRU cache stores the capacity of a given slice given _BOTH_ the slice's left and right bounds. If you later come with a slice that has only one correct bound, the LRU doesn't care about it. That's the safety tidbit. Andrei
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: 2. I don't understand how this solves the safety problem: // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. Let me stress a point harder that I think I expressed poorly: The LRU cache stores the capacity of a given slice given _BOTH_ the slice's left and right bounds. If you later come with a slice that has only one correct bound, the LRU doesn't care about it. That's the safety tidbit. Andrei I think I get it now, although it's very conservative. One more question: Is this going to take the place of ArrayBuilder or be inaddition? The LRU is a good hack to preserve syntactic elegance and ease of use, but it's somewhat of a kludge nonetheless and I'd ideally still like to see a real ArrayBuilder with full array-like semantics if T[new] is definitely out.
Re: LRU cache for ~=
dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: 2. I don't understand how this solves the safety problem: // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. Let me stress a point harder that I think I expressed poorly: The LRU cache stores the capacity of a given slice given _BOTH_ the slice's left and right bounds. If you later come with a slice that has only one correct bound, the LRU doesn't care about it. That's the safety tidbit. Andrei I think I get it now, although it's very conservative. That's why we need experimentation. My gut tells me that linear search will scram through 8 elements real fast, and if there's something in the cache it will almost always be in a top position. Also, I think the LRU will solve all cases that matter. One more question: Is this going to take the place of ArrayBuilder or be inaddition? The LRU is a good hack to preserve syntactic elegance and ease of use, but it's somewhat of a kludge nonetheless and I'd ideally still like to see a real ArrayBuilder with full array-like semantics if T[new] is definitely out. Ideally we'd be able to render ArrayBuilder/Appender unnecessary. I think there is a need for a UniqueArray, but that's only loosely related. Andrei
Re: LRU cache for ~=
Andrei Alexandrescu wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. So whenever code calls arr ~= b, the LRU is searched first. If the system finds arr (both bounds) in the LRU, that means the cached capacity is correct and can solve the matter without an actual trip to the GC at all! Otherwise, do the deed and cache the new slice and the new capacity. This also solves the lack of safety: if you request a growth on an array you just grew, it's impossible to have a valid slice beyond that array. This LRU would allow us to keep the slice API as it currently is, and also at excellent efficiency. What do you think? Andrei Its an interesting idea, and if I have time tonight I'll take a crack at it. For those others who may, the function you care about is _d_arrayappendcT in compiler/dmd/lifetime. -- Chris Nicholson-Sauls
Re: LRU cache for ~=
On 2009-10-19 21:53:53 +0200, Andrei Alexandrescu seewebsiteforem...@erdani.org said: dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: 2. I don't understand how this solves the safety problem: // foo lives on the heap b/c we've idup'd it. string foo = This is only a test..idup; string bar = foo[0..4]; bar ~= is _not ; writeln(foo); // prints This is _not a test. Having access to the capacity in an LRU cache doesn't help if I understand it correctly. Let me stress a point harder that I think I expressed poorly: The LRU cache stores the capacity of a given slice given _BOTH_ the slice's left and right bounds. If you later come with a slice that has only one correct bound, the LRU doesn't care about it. That's the safety tidbit. Andrei I think I get it now, although it's very conservative. That's why we need experimentation. My gut tells me that linear search will scram through 8 elements real fast, and if there's something in the cache it will almost always be in a top position. Also, I think the LRU will solve all cases that matter. yes you are probably right One more question: Is this going to take the place of ArrayBuilder or be inaddition? The LRU is a good hack to preserve syntactic elegance and ease of use, but it's somewhat of a kludge nonetheless and I'd ideally still like to see a real ArrayBuilder with full array-like semantics if T[new] is definitely out. Ideally we'd be able to render ArrayBuilder/Appender unnecessary. I think there is a need for a UniqueArray, but that's only loosely related. a library array knowing its capacity is still useful I think. Fawzi
Re: LRU cache for ~=
Andrei Alexandrescu Wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. Shouldn't it be MRU (Most Recently Used)? I assume ~= will be disallowed for shared arrays? Will all arrays overallicate when created in order to allow efficient appending? I feel like bolting on ~= to slices gives mediocre performance and a lot of gotchas. Some users may expect slices to be reference copied arrays and want to see appended data in other slices. Some may get away with mutating the array in the MRU or a slice and seeing the changes, only to get bitten later. A separate type really makes more sense to me...
Re: LRU cache for ~=
Jason House wrote: Andrei Alexandrescu Wrote: I just wrote this to Sean and Walter and subsequently discussed it with Walter. Walter thinks this should work. Does anyone have the time and inclination to test this out? It would involve hacking into druntime's implementation of ~= (I'm not sure what the function name is). I'd really appreciate this; I'm overloaded as it is. == In wake of the recent demise of T[new], I was thinking of finding ways of making ~= efficient for T[]. Currently ~= is slow because accessing GC.sizeOf(void*) acquires a global lock and generally must figure out a lot of things about the pointer to make a decision. Also, ~= is dangerous because it allows slices to stomp over other slices. I was thinking of solving these issues by keeping an LRU (Least Recently Used) cache inside the implementation of ~=. The LRU would only have a few entries (4-8) and would store the parameters of the last ~= calls, and their cached capacities. Shouldn't it be MRU (Most Recently Used)? Yes. LRU would be the eviction strategy. I assume ~= will be disallowed for shared arrays? I think so. Will all arrays overallicate when created in order to allow efficient appending? Upon the first append. I feel like bolting on ~= to slices gives mediocre performance and a lot of gotchas. You may be right about the gotchas, but you better produce some numbers wrt performance. Some users may expect slices to be reference copied arrays and want to see appended data in other slices. Some may get away with mutating the array in the MRU or a slice and seeing the changes, only to get bitten later. A separate type really makes more sense to me... Making ~= work well for slices does not preclude creating a distinct library type. Andrei
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article Making ~= work well for slices does not preclude creating a distinct library type. Andrei Yes, I see LRU/MRU/whatever as a kludge to allow convenience without being unsafe egregiously inefficient. The fact that it is a kludge is not a criticism because there is clearly no easy answer, and thus a kludge is genuinely necessary. However, I think there needs to be a separate array builder type for heavy duty appending operations. In TDPL I would just say that concatenating slices can be inefficient and that people should use array builder for heavy duty appending, length changing, etc., w/o getting into the details.
Re: LRU cache for ~=
On 2009-10-20 00:13:05 +0200, dsimcha dsim...@yahoo.com said: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article Making ~= work well for slices does not preclude creating a distinct library type. Andrei Yes, I see LRU/MRU/whatever as a kludge to allow convenience without being unsafe egregiously inefficient. The fact that it is a kludge is not a criticism because there is clearly no easy answer, and thus a kludge is genuinely necessary. However, I think there needs to be a separate array builder type for heavy duty appending operations. In TDPL I would just say that concatenating slices can be inefficient and that people should use array builder for heavy duty appending, length changing, etc., w/o getting into the details. I fully agree
Re: LRU cache for ~=
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article
Making ~= work well for slices does not preclude creating a distinct
library type.
Andrei
Yes, I see LRU/MRU/whatever as a kludge to allow convenience without being
unsafe
egregiously inefficient. The fact that it is a kludge is not a criticism
because
there is clearly no easy answer, and thus a kludge is genuinely necessary.
However, I think there needs to be a separate array builder type for heavy
duty
appending operations. In TDPL I would just say that concatenating slices can be
inefficient and that people should use array builder for heavy duty appending,
length changing, etc., w/o getting into the details.
But how about the opposite view. What if the *previous* implementation
of GC and ~= were a kludge that led to egregious inefficiency and
revolting unsafety, kludge that that now is getting fixed.
What I'm saying is that with the cache in place we'll have slices that
are safe and efficient. Right now they are not safe and not efficient. I
can hardly find reasons to characterize the new state of affairs as kludgey.
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Andrei
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article
Making ~= work well for slices does not preclude creating a distinct
library type.
Andrei
Yes, I see LRU/MRU/whatever as a kludge to allow convenience without being
unsafe
egregiously inefficient. The fact that it is a kludge is not a criticism
because
there is clearly no easy answer, and thus a kludge is genuinely necessary.
However, I think there needs to be a separate array builder type for heavy
duty
appending operations. In TDPL I would just say that concatenating slices
can be
inefficient and that people should use array builder for heavy duty
appending,
length changing, etc., w/o getting into the details.
But how about the opposite view. What if the *previous* implementation
of GC and ~= were a kludge that led to egregious inefficiency and
revolting unsafety, kludge that that now is getting fixed.
What I'm saying is that with the cache in place we'll have slices that
are safe and efficient. Right now they are not safe and not efficient. I
can hardly find reasons to characterize the new state of affairs as kludgey.
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Andrei
Started playing w/ the implementation a little and I see a problem. What about
the garbage collector? There are two possibilities:
1. The pointer gets stored as a pointer. The array never gets freed until it
gets evicted from the cache. If it's a huge array, this is very bad.
2. The pointer gets cast to a size_t so it doesn't look like a reference.
Then,
we can have something like:
string foo;
foreach(i; 0..5) {
foo ~= 'a';
}
foo = null;
GC.collect;
// Now we have (foo.ptr, 5) in our LRU cache, but foo has been GC'd.
string bar = 123456789; // bar gets the memory reclaimed from foo.
string baz = bar[0..5]; // Same length, ptr as foo.
baz ~= '1'; // Finds stale information from foo in cache.
writeln(bar); // Prints [1 2 3 4 5 1 7 8 9].
The only possible solutions I see would be to have the GC know everything about
the LRU cache and evict stale entries (probably slows down GC a lot, a huge PITA
to implement, couples things that shouldn't be tightly coupled), or clear the
cache every time GC is run (probably would make appending so slow as to defeat
the
purpose of having the cache).
Re: LRU cache for ~=
Jason House Wrote: Will all arrays overallicate when created in order to allow efficient appending? They effectively do anyway, since the GC block sizes are in powers of 2 and the runtime relies on the GC to determine block capacity. In effect, it's simply the GC that's over-allocating instead of the runtime itself.
Re: LRU cache for ~=
Andrei Alexandrescu Wrote: What I'm saying is that with the cache in place we'll have slices that are safe and efficient. Right now they are not safe and not efficient. I can hardly find reasons to characterize the new state of affairs as kludgey. I'm not sure I agree that they'd be safe. Given only a pointer to the head of a block there's no way to know whether it represents the array or a slice of the array from [0..n].
Re: LRU cache for ~=
Andrei Alexandrescu Wrote: I think GC.collect may simply evict the entire cache. The collection cycle costs so much, the marginal cost of losing cached information is lost in the noise. Right. The GC already has an LRU cache of size 1 (ie. LRU entry) for GC.sizeOf() and GC.query(), and these values are cleared on a collection by necessity, since pages can be repurposed if they're been completely emptied.
Re: LRU cache for ~=
Sean Kelly wrote: Andrei Alexandrescu Wrote: What I'm saying is that with the cache in place we'll have slices that are safe and efficient. Right now they are not safe and not efficient. I can hardly find reasons to characterize the new state of affairs as kludgey. I'm not sure I agree that they'd be safe. Given only a pointer to the head of a block there's no way to know whether it represents the array or a slice of the array from [0..n]. That's the beauty of the cache: with the cache in place, you may actually know that the memory beyond the end of a cached slice is seen by nobody else. Andrei
Re: LRU cache for ~=
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
--
Rainer Deyke - rain...@eldwood.com
Re: LRU cache for ~=
Andrei Alexandrescu, el 19 de octubre a las 17:24 me escribiste:
dsimcha wrote:
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article
Making ~= work well for slices does not preclude creating a distinct
library type.
Andrei
Yes, I see LRU/MRU/whatever as a kludge to allow convenience without being
unsafe
egregiously inefficient. The fact that it is a kludge is not a criticism
because
there is clearly no easy answer, and thus a kludge is genuinely necessary.
However, I think there needs to be a separate array builder type for heavy
duty
appending operations. In TDPL I would just say that concatenating slices
can be
inefficient and that people should use array builder for heavy duty
appending,
length changing, etc., w/o getting into the details.
But how about the opposite view. What if the *previous*
implementation of GC and ~= were a kludge that led to egregious
inefficiency and revolting unsafety, kludge that that now is getting
fixed.
What I'm saying is that with the cache in place we'll have slices
that are safe and efficient. Right now they are not safe and not
efficient. I can hardly find reasons to characterize the new state
of affairs as kludgey.
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
And for me this is enough to remove appending from slicing and having
a proper array type.
There is no reason to keep that buggy behaviour.
--
Leandro Lucarella (AKA luca) http://llucax.com.ar/
--
GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145 104C 949E BFB6 5F5A 8D05)
--
DONAN UN PENE EN NICARAGUA
-- Crónica TV
Re: LRU cache for ~=
On Tue, 20 Oct 2009 03:00:57 +0400, Rainer Deyke rain...@eldwood.com
wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
Safe as in SafeD (i.e. no memory corruption) :)
Re: LRU cache for ~=
dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: Started playing w/ the implementation a little and I see a problem. What about the garbage collector? There are two possibilities: [snip] The only possible solutions I see would be to have the GC know everything about the LRU cache and evict stale entries (probably slows down GC a lot, a huge PITA to implement, couples things that shouldn't be tightly coupled), or clear the cache every time GC is run (probably would make appending so slow as to defeat the purpose of having the cache). I think GC.collect may simply evict the entire cache. The collection cycle costs so much, the marginal cost of losing cached information is lost in the noise. Andrei But then you have to copy the whole array again, likely triggering another GC if the array is large. Then things really get ugly as, for all practical purposes, you've completely done away with the cache. This happens whether or not a cache is in use. Andrei
Re: LRU cache for ~=
Rainer Deyke wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
What's yours?
Andrei
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: Started playing w/ the implementation a little and I see a problem. What about the garbage collector? There are two possibilities: [snip] The only possible solutions I see would be to have the GC know everything about the LRU cache and evict stale entries (probably slows down GC a lot, a huge PITA to implement, couples things that shouldn't be tightly coupled), or clear the cache every time GC is run (probably would make appending so slow as to defeat the purpose of having the cache). I think GC.collect may simply evict the entire cache. The collection cycle costs so much, the marginal cost of losing cached information is lost in the noise. Andrei But then you have to copy the whole array again, likely triggering another GC if the array is large. Then things really get ugly as, for all practical purposes, you've completely done away with the cache. This happens whether or not a cache is in use. Andrei But the array isn't guaranteed to get reallocated immediately after *every* GC run. If you're appending to a huge array, the GC will likely run several times while you're appending, leading to several unnecessary reallocations. Each of those unnecessary reallocations will increase the memory footprint of your program, possibly triggering another GC run and wiping out your cache again in short order, until, for sufficiently large arrays, a ~= b; is almost equivalent to a = a ~ b;
Re: LRU cache for ~=
dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: Started playing w/ the implementation a little and I see a problem. What about the garbage collector? There are two possibilities: [snip] The only possible solutions I see would be to have the GC know everything about the LRU cache and evict stale entries (probably slows down GC a lot, a huge PITA to implement, couples things that shouldn't be tightly coupled), or clear the cache every time GC is run (probably would make appending so slow as to defeat the purpose of having the cache). I think GC.collect may simply evict the entire cache. The collection cycle costs so much, the marginal cost of losing cached information is lost in the noise. Andrei But then you have to copy the whole array again, likely triggering another GC if the array is large. Then things really get ugly as, for all practical purposes, you've completely done away with the cache. This happens whether or not a cache is in use. Andrei But the array isn't guaranteed to get reallocated immediately after *every* GC run. If you're appending to a huge array, the GC will likely run several times while you're appending, leading to several unnecessary reallocations. I don't think I understand this. 1. Request for an append comes that runs out of memory 2. GC runs and clears memory 3. Array is reallocated and the capacity cached. No? Each of those unnecessary reallocations will increase the memory footprint of your program, possibly triggering another GC run and wiping out your cache again in short order, until, for sufficiently large arrays, a ~= b; is almost equivalent to a = a ~ b; I don't understand how the cache makes that all worse. Andrei
Re: LRU cache for ~=
Bill Baxter wrote:
On Mon, Oct 19, 2009 at 5:07 PM, Andrei Alexandrescu
seewebsiteforem...@erdani.org wrote:
Rainer Deyke wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
What's yours?
Probably something like safe from easy-to-write hard-to-debug
programming errors.
I agree that it would stink if all these changes were made and *still*
one of the top gotchas with arrays/slices remained. It also makes me
think slices and appending just don't belong together. Appending to a
view ... just say no.
How to reconcile this viewpoint with that of people who find ~= all too
darn convenient?
Andrei
Re: LRU cache for ~=
Denis Koroskin wrote: Safe as in SafeD (i.e. no memory corruption) :) Right. The problems with other definitions of safe is they are too ill-defined.
Re: LRU cache for ~=
Denis Koroskin wrote:
On Tue, 20 Oct 2009 03:00:57 +0400, Rainer Deyke rain...@eldwood.com
wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
Safe as in SafeD (i.e. no memory corruption) :)
If the caller wasn't expecting the array to be modified, then that's a
textbook case of memory corruption. If the caller /was/ expecting the
array to be modified, then it's the opposite, which isn't much better.
--
Rainer Deyke - rain...@eldwood.com
Re: LRU cache for ~=
Andrei Alexandrescu wrote: Rainer Deyke wrote: Your definition of safe is clearly not aligned with mine. What's yours? Safety is not an absolute, but a question of degree. The harder it is to write incorrect code, the safer the language. One key element of this is deterministic behavior. If you rely on the whim of the runtime to determine if two slices refer to the same data, then it becomes much harder to reason about or test the code. -- Rainer Deyke - rain...@eldwood.com
Re: LRU cache for ~=
On Mon, 19 Oct 2009, Walter Bright wrote: Denis Koroskin wrote: Safe as in SafeD (i.e. no memory corruption) :) Right. The problems with other definitions of safe is they are too ill-defined. There's SafeD, which has a fairly formal definition. The other side of it is the general principle of D which is that the right way should be the easy and obvious way. Slices and arrays have issue with the principle.
Re: LRU cache for ~=
Andrei Alexandrescu, el 19 de octubre a las 20:12 me escribiste:
Bill Baxter wrote:
On Mon, Oct 19, 2009 at 5:07 PM, Andrei Alexandrescu
seewebsiteforem...@erdani.org wrote:
Rainer Deyke wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
What's yours?
Probably something like safe from easy-to-write hard-to-debug
programming errors.
I agree that it would stink if all these changes were made and *still*
one of the top gotchas with arrays/slices remained. It also makes me
think slices and appending just don't belong together. Appending to a
view ... just say no.
How to reconcile this viewpoint with that of people who find ~= all
too darn convenient?
Use a proper array type, not a view (slice).
--
Leandro Lucarella (AKA luca) http://llucax.com.ar/
--
GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145 104C 949E BFB6 5F5A 8D05)
--
No recuerdo las flores, no conozco el viento
Aquí donde envejecen juntos presos y carceleros
Los días no pasan, el tiempo no pasa
Y vivimos contando el tiempo
Y moriremos contando el tiempo
Re: LRU cache for ~=
Andrei Alexandrescu wrote:
I was thinking of solving these issues by keeping an LRU (Least Recently
Used) cache inside the implementation of ~=. The LRU would only have a
few entries (4-8) and would store the parameters of the last ~= calls,
and their cached capacities.
Sounds like a bad hack, but at least it would solve the issue about
overwritten slices.
But for some uses of array appending, you had to use a library based
Appender (or whatever you have in Phobos 2):
class Something {
T[] data;
void add(T x) {
data ~= x; //chances that data are in the cache are minimal
}
}
There's no cache locality or anything, but obviously you still would
like to have efficient appender behavior.
Also look at this:
string[] t;
t ~= somefunction();
t ~= someotherfunction();
Chances are high that those functions will remove t from the cache,
and it would all be inefficient again. Nothing solved!
Now you could try to make the cache function local to solve this issue.
Whenever a function contains the ~= operator, the compiler allocates a
cache struct, and the ~= operator passes a hidden pointer to it to the
runtime.
But that wouldn't work if you want pass slices to other functions and to
append to them (but it would still be safe, I guess?).
Looks like these performance hacks just don't work out. It'd be so much
simpler to make a~=b; equivalent to a=a~b;. Then there's no
mystery about the performance of the ~= operator. You can explain
everyone: yes, this allocates; if you want performance, use
ArrayAppender. Hey, you could alias this ArrayAppender type to
something like T[new] to make it feel more natural. But then, we're at
the beginning again.
Re: LRU cache for ~=
== Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: == Quote from Andrei Alexandrescu (seewebsiteforem...@erdani.org)'s article dsimcha wrote: Started playing w/ the implementation a little and I see a problem. What about the garbage collector? There are two possibilities: [snip] The only possible solutions I see would be to have the GC know everything about the LRU cache and evict stale entries (probably slows down GC a lot, a huge PITA to implement, couples things that shouldn't be tightly coupled), or clear the cache every time GC is run (probably would make appending so slow as to defeat the purpose of having the cache). I think GC.collect may simply evict the entire cache. The collection cycle costs so much, the marginal cost of losing cached information is lost in the noise. Andrei But then you have to copy the whole array again, likely triggering another GC if the array is large. Then things really get ugly as, for all practical purposes, you've completely done away with the cache. This happens whether or not a cache is in use. Andrei But the array isn't guaranteed to get reallocated immediately after *every* GC run. If you're appending to a huge array, the GC will likely run several times while you're appending, leading to several unnecessary reallocations. I don't think I understand this. 1. Request for an append comes that runs out of memory 2. GC runs and clears memory 3. Array is reallocated and the capacity cached. No? This is entirely correct. Each of those unnecessary reallocations will increase the memory footprint of your program, possibly triggering another GC run and wiping out your cache again in short order, until, for sufficiently large arrays, a ~= b; is almost equivalent to a = a ~ b; I don't understand how the cache makes that all worse. Andrei The cache doesn't make anything *worse* than with no cache. The only point I'm trying to make is that, for large arrays, if the GC clears the cache every time it runs, things would start to get *almost as bad as* having no cache because the copy operations become expensive and the GC may run frequently.
Re: LRU cache for ~=
Rainer Deyke wrote:
Denis Koroskin wrote:
On Tue, 20 Oct 2009 03:00:57 +0400, Rainer Deyke rain...@eldwood.com
wrote:
Andrei Alexandrescu wrote:
One surprising (but safe) behavior that remains with slices is this:
void fun(int[] a) {
a[0] = 0;
a ~= 42;
a[0] = 42;
}
The caller may or may not see 42 in the first slot after the call.
Your definition of safe is clearly not aligned with mine.
Safe as in SafeD (i.e. no memory corruption) :)
If the caller wasn't expecting the array to be modified, then that's a
textbook case of memory corruption.
[citation needed]
Andrei
Re: LRU cache for ~=
Rainer Deyke wrote: Andrei Alexandrescu wrote: Rainer Deyke wrote: Your definition of safe is clearly not aligned with mine. What's yours? Safety is not an absolute, but a question of degree. The harder it is to write incorrect code, the safer the language. Well you got a point. I think I will from here on talk about soundness instead of safety, because the former is absolute. One key element of this is deterministic behavior. If you rely on the whim of the runtime to determine if two slices refer to the same data, then it becomes much harder to reason about or test the code. I agree. Andrei
Re: LRU cache for ~=
grauzone wrote:
Andrei Alexandrescu wrote:
I was thinking of solving these issues by keeping an LRU (Least
Recently Used) cache inside the implementation of ~=. The LRU would
only have a few entries (4-8) and would store the parameters of the
last ~= calls, and their cached capacities.
Sounds like a bad hack, but at least it would solve the issue about
overwritten slices.
But for some uses of array appending, you had to use a library based
Appender (or whatever you have in Phobos 2):
class Something {
T[] data;
void add(T x) {
data ~= x; //chances that data are in the cache are minimal
}
}
There's no cache locality or anything, but obviously you still would
like to have efficient appender behavior.
Why is there no cache locality?
Also look at this:
string[] t;
t ~= somefunction();
t ~= someotherfunction();
Chances are high that those functions will remove t from the cache,
and it would all be inefficient again. Nothing solved!
Why are there chances those functions will remove t from the cache?
For it to become a performance problem, there must be a loop with nine
or more round-robin appends. When that does happen, yes, appends will
become slower. (We may be able to make that better by using random
eviction.)
Now you could try to make the cache function local to solve this issue.
Whenever a function contains the ~= operator, the compiler allocates a
cache struct, and the ~= operator passes a hidden pointer to it to the
runtime.
But that wouldn't work if you want pass slices to other functions and to
append to them (but it would still be safe, I guess?).
Looks like these performance hacks just don't work out.
Caches have a long and solid history of working. You'd have a very hard
time convincing me that a cache that directly addresses a performance
problem is a hack.
It'd be so much
simpler to make a~=b; equivalent to a=a~b;. Then there's no
mystery about the performance of the ~= operator. You can explain
everyone: yes, this allocates; if you want performance, use
ArrayAppender. Hey, you could alias this ArrayAppender type to
something like T[new] to make it feel more natural. But then, we're at
the beginning again.
I think making ~= faster is worth pursuing.
Andrei
Re: LRU cache for ~=
Andrei Alexandrescu wrote: Rainer Deyke wrote: If the caller wasn't expecting the array to be modified, then that's a textbook case of memory corruption. [citation needed] I guess we need to define memory corruption first. Memory corruption is when a piece of code erroneously overwrites memory. That applies here. Do you have a better definition? -- Rainer Deyke - rain...@eldwood.com
Re: LRU cache for ~=
Rainer Deyke wrote: Andrei Alexandrescu wrote: Rainer Deyke wrote: If the caller wasn't expecting the array to be modified, then that's a textbook case of memory corruption. [citation needed] I guess we need to define memory corruption first. Memory corruption is when a piece of code erroneously overwrites memory. Where's that quote from? That definition is too vague to be of any use. Even an int that's written with a value when the spec meant to write another value is erroneously written and would be, by your definition, memory corruption. It is not. That applies here. Do you have a better definition? I do, but since you mentioned a textbook case of memory corruption, I was curious which textbook that wold come from. That's why I asked for a _citation_. A vague useless ad-hoc definition is easy to put together. Andrei
Re: LRU cache for ~=
Andrei Alexandrescu wrote: Rainer Deyke wrote: I guess we need to define memory corruption first. Memory corruption is when a piece of code erroneously overwrites memory. Where's that quote from? It's my own attempt to define memory corruption. It's equivalent to the definition in Wikipedia: Memory corruption happens when the contents of a memory location are unintentionally modified due to programming errors. [...] Do you have a better definition? I do, Then post it. but since you mentioned a textbook case of memory corruption, I was curious which textbook that wold come from. My use of the word textbook was idiomatic, not literal. From http://www.answers.com/topic/textbook: text·book (tĕkst'bʊk') [...] adj. Being a characteristic example of its kind; classic: a textbook case of schizophrenia. -- Rainer Deyke - rain...@eldwood.com
