Follow up to #17383. This is a minor optimization that only matters when a small allocation is resized/free'd soon after it is allocated.
The only real difference I was able to observe with this was via a synthetic benchmark that allocates a full bucket and then frees all but one of the slots, over and over in a loop:
Debug build:
Benchmark 1 (9 runs): gpa-degen-master.exe
measurement mean ± σ min … max outliers delta
wall_time 575ms ± 5.19ms 569ms … 583ms 0 ( 0%) 0%
peak_rss 43.8MB ± 1.37KB 43.8MB … 43.8MB 1 (11%) 0%
Benchmark 2 (10 runs): gpa-degen-search-cur.exe
measurement mean ± σ min … max outliers delta
wall_time 532ms ± 5.55ms 520ms … 539ms 0 ( 0%) ⚡- 7.5% ± 0.9%
peak_rss 43.8MB ± 65.2KB 43.8MB … 44.0MB 1 (10%) + 0.0% ± 0.1%
ReleaseFast build:
Benchmark 1 (129 runs): gpa-degen-master-release.exe
measurement mean ± σ min … max outliers delta
wall_time 38.9ms ± 1.12ms 36.7ms … 42.4ms 8 ( 6%) 0%
peak_rss 23.2MB ± 2.39KB 23.2MB … 23.2MB 0 ( 0%) 0%
Benchmark 2 (151 runs): gpa-degen-search-cur-release.exe
measurement mean ± σ min … max outliers delta
wall_time 33.2ms ± 999us 31.9ms … 36.3ms 20 (13%) ⚡- 14.7% ± 0.6%
peak_rss 23.2MB ± 2.26KB 23.2MB … 23.2MB 0 ( 0%) + 0.0% ± 0.0%
Notably, this contains bug fixes related to `@errorCast` which are
required by the changes to `std.io.Reader` in this branch, and the
compiler source code has a dependency on `std.io.Reader`.
The idea here is to avoid code bloat by having only one actual io.Reader
implementation, which is type erased, and then implement a GenericReader
that preserves type information on top of that as thin glue code.
The strategy here is for that glue code to `@errSetCast` the result of
the type-erased reader functions, however, while trying to do that I
ran into #17343.
Implement the stub for Elf.
I believe that separating the concerns, namely, having an interface
function that is responsible for signalling the linker to lower
the anon decl only, and a separate function to obtain the decl's
vaddr is preferable since it allows us to handle codegen errors
in a simpler way.
Introduce the new mechanism needed to render anonymous decls to C code
that the frontend is now using.
The current strategy is to collect the set of used anonymous decls into
one ArrayHashMap for the entire compilation, and then render them during
flush().
In the future this may need to be adjusted for incremental compilation
purposes, so that removing a Decl from decl_table means that newly
unused anonymous decls are no longer rendered. However, let's do one
thing at a time. The only goal of this branch is to stop using
Module.Decl objects for unnamed constants.
Start keeping track of dependencies on anon decls for dependency
ordering during flush()
Currently this causes use of undefined symbols because these
dependencies need to get rendered into the output.
Instead of explicitly creating a `Module.Decl` object for each anonymous
declaration, each `InternPool.Index` value is implicitly understood to
be an anonymous declaration when encountered by backend codegen.
The memory management strategy for these anonymous decls then becomes to
garbage collect them along with standard InternPool garbage.
In the interest of a smooth transition, this commit only implements this
new scheme for string literals and leaves all the previous mechanisms in
place.
The `makeMem*()` functions crashed under valgrind in Debug and
ReleaseSafe modes.
The reason being that `doMemCheckClientRequestExpr()` returns `0`
when not running under Valgrind, and `maxInt(usize)` when running
under Valgrind.
Thus, `@as(i1, @intCast(maxInt(usize)))` always fails and these
functions crashed before returning.
That being said, what these functions used to return was quite
unexpected: `0` on error and `-1` on success (=running under valgrind).
That doesn't match any Zig nor C conventions.
But that return value doesn't seem to be very useful. Either we are
running under Valgrind or we are not. There's no point in checking this
for every single call. Applications are likely to always discard it.
So, just return a `void` instead.
Also avoid function comments that start with `Similarly, ...` because
that doesn't refer to anything in the context of autodoc or in IDEs.
Before this commit, GeneralPurposeAllocator could run into incredibly degraded performance in scenarios where the bucket count for a particular size class grew to be large. For example, if exactly `slot_count` allocations of a single size class were performed and then all of them were freed except one, then the bucket for those allocations would have to be kept around indefinitely. If that pattern of allocation were done over and over, then the bucket list for that size class could grow incredibly large.
This allocation pattern has been seen in the wild: https://github.com/Vexu/arocc/issues/508#issuecomment-1738275688
In that case, the length of the bucket list for the `128` size class would grow to tens of thousands of buckets and cause Debug runtime to balloon to ~8 minutes whereas with the c_allocator the Debug runtime would be ~3 seconds.
To address this, there are three different changes happening here:
1. std.Treap is used instead of a doubly linked list for the lists of buckets. This takes the time complexity of searchBucket [used in resize and free] from O(n) to O(log n), but increases the time complexity of insert from O(1) to O(log n) [before, all new buckets would get added to the head of the list]. Note: Any data structure with O(log n) or better search/insert/delete would also work for this use-case.
2. If the 'current' bucket for a size class is full, the list of buckets is never traversed and instead a new bucket is allocated. Previously, traversing the bucket list could only find a non-full bucket in specific circumstances, and only because of a separate optimization that is no longer needed (before, after any resize/free, the affected bucket would be moved to the head of the bucket list to allow searchBucket to perform better on average). Now, the current_bucket for each size class only changes when either (1) the current bucket is emptied/freed, or (2) a new bucket is allocated (due to the current bucket being full or null). Because each bucket's alloc_cursor only moves forward (i.e. slots within a bucket are never re-used), we can therefore always know that any bucket besides the current_bucket will be full, so traversing the list in the hopes of finding an existing non-full bucket is entirely pointless.
3. Size + alignment information for small allocations has been moved into the Bucket data instead of keeping it in a separate HashMap. This offers an improvement over the HashMap since whenever we need to get/modify the length/alignment of an allocation it's extremely likely we will already have calculated any bucket-related information necessary to get the data.
The first change is the most relevant and accounts for most of the benefit here. Also note that the overall functionality of GeneralPurposeAllocator is unchanged.
In the degraded `arocc` case, these changes bring Debug performance from ~8 minutes to ~20 seconds.
Benchmark 1: test-master.bat
Time (mean ± σ): 481.263 s ± 5.440 s [User: 479.159 s, System: 1.937 s]
Range (min … max): 477.416 s … 485.109 s 2 runs
Benchmark 2: test-optim-treap.bat
Time (mean ± σ): 19.639 s ± 0.037 s [User: 18.183 s, System: 1.452 s]
Range (min … max): 19.613 s … 19.665 s 2 runs
Summary
'test-optim-treap.bat' ran
24.51 ± 0.28 times faster than 'test-master.bat'
Note: Much of the time taken on Windows in this particular case is related to gathering stack traces. With `.stack_trace_frames = 0` the runtime goes down to 6.7 seconds, which is a little more than 2.5x slower compared to when the c_allocator is used.
These changes may or mat not introduce a slight performance regression in the average case:
Here's the standard library tests on Windows in Debug mode:
Benchmark 1 (10 runs): std-tests-master.exe
measurement mean ± σ min … max outliers delta
wall_time 16.0s ± 30.8ms 15.9s … 16.1s 1 (10%) 0%
peak_rss 42.8MB ± 8.24KB 42.8MB … 42.8MB 0 ( 0%) 0%
Benchmark 2 (10 runs): std-tests-optim-treap.exe
measurement mean ± σ min … max outliers delta
wall_time 16.2s ± 37.6ms 16.1s … 16.3s 0 ( 0%) 💩+ 1.3% ± 0.2%
peak_rss 42.8MB ± 5.18KB 42.8MB … 42.8MB 0 ( 0%) + 0.1% ± 0.0%
And on Linux:
Benchmark 1: ./test-master
Time (mean ± σ): 16.091 s ± 0.088 s [User: 15.856 s, System: 0.453 s]
Range (min … max): 15.870 s … 16.166 s 10 runs
Benchmark 2: ./test-optim-treap
Time (mean ± σ): 16.028 s ± 0.325 s [User: 15.755 s, System: 0.492 s]
Range (min … max): 15.735 s … 16.709 s 10 runs
Summary
'./test-optim-treap' ran
1.00 ± 0.02 times faster than './test-master'
* add nested packed struct/union behavior tests
* use ptr_info.packed_offset rather than trying to duplicate the logic from Sema.structFieldPtrByIndex()
* use the container_ptr_info.packed_offset to account for non-aligned nested structs.
* dedup type.packedStructFieldBitOffset() and module.structPackedFieldBitOffset()
zig fetch [options] <url>
zig fetch [options] <path>
Fetches a package which is found at <url> or <path> into the global
cache directory, printing the package hash to stdout.
Closes#16972
Related to #14280
Additionally, this commit:
* Adds uncompressed .tar support to package fetching
* Introduces symlink support to package fetching
