Prior to this CL, there were several different code paths for reading/writing message data. Generated code, MiniTable accessors, and reflection all performed direct manipulation of the bits and bytes in a message, but they all had distinct implementations that did not share much of any code. This divergence meant that they could easily have different behavior, bugs could creep into one but not another, and we would need three different sets of tests to get full test coverage. This also made it very difficult to change the internal representation in any way, since it would require updating many places in the code.
With this CL, the three different APIs for accessing message data now all share a common set of functions. The common functions all take a `upb_MiniTableField` as the canonical description of a field's type and layout. The lowest-level functions are very branchy, as they must test for every possible variation in the field type (field vs oneof, hasbit vs no-hasbit, different field sizes, whether a nonzero default value exists, extension vs. regular field), however these functions are declared inline and designed to be very optimizable when values are known at compile time.
In generated accessors, for example, we can declare constant `upb_MiniTableField` instances so that all values can constant-propagate, and we can get fully specialized code even though we are calling a generic function. On the other hand, when we use the generic functions from reflection, we get runtime branches since values are not known at compile time. But even the function is written to still be as efficient as possible even when used from reflection. For example, we use memcpy() calls with constant length so that the compiler can optimize these into inline loads/stores without having to make an out-of-line call to memcpy().
In this way, this CL should be a benefit to both correctness and performance. It will also make it easier to change the message representation, for example to optimize the encoder by giving hasbits to all fields.
Note that we have not completely consolidated all access in this CL:
1. Some functions outside of get/set such as clear and hazzers are not yet unified.
2. The encoder and decoder still touch the message without going through the common functions. The encoder and decoder require a bit more specialized code to get good performance when reading/writing fields en masse.
PiperOrigin-RevId: 490016095
upb has traditionally returned 16-byte-aligned pointers from arena allocation. This was out of an abundance of caution, since users could theoretically be using upb arenas to allocate memory that is then used for SSE/AVX values (eg. [`__m128`](https://docs.microsoft.com/en-us/cpp/cpp/m128?view=msvc-170), which require 16-byte alignment.
In practice, the protobuf C++ arena has used 8-byte alignment for 8 years with no significant problems I know of arising from SSE etc.
Reducing the alignment requirement to 8 will save memory. It will also help with compatibility on 32-bit architectures where `malloc()` only returns 8-byte aligned memory. The immediate motivation is to fix the win32 build for Python protobuf.
PiperOrigin-RevId: 448331777
Unfortunately a few of the Clang warnings did not have easy fixes:
../../../../ext/google/protobuf_c/ruby-upb.c: In function ‘fastdecode_err’:
../../../../ext/google/protobuf_c/ruby-upb.c:353:13: warning: function might be candidate for attribute ‘noreturn’ [-Wsuggest-attribute=noreturn]
353 | const char *fastdecode_err(upb_decstate *d) {
| ^~~~~~~~~~~~~~
../../../../ext/google/protobuf_c/ruby-upb.c: In function ‘_upb_decode’:
../../../../ext/google/protobuf_c/ruby-upb.c:867:30: warning: argument ‘buf’ might be clobbered by ‘longjmp’ or ‘vfork’ [-Wclobbered]
867 | bool _upb_decode(const char *buf, size_t size, void *msg,
I even tried to suppress the first error, but it still shows up.
upb previously attempted to support C89 and pre-2015 versions
of Visual Studio. This was to support older compilers with
limited C99 support (particularly MSVC). But as of last August,
even gRPC has dropped support for MSVC prior to 2015
c87276d058
Therefore it seems safe for upb to no longer attempt C89 support
(we were already not truly C89 compliant, with our use of "bool").
We now explicitly require C99 or greater and MSVC 2015 or greater.
This cleaned up port_def.inc a fair bit. I took the chance to
also remove some obsolete macros.
* resolvename is declared to return a bool value, but instead can return NULL. MSVC 2019 does not like that an throws a compile error. Fixed by returning false instead of NULL.
* When compiling with MSVC 2019, the UPB_ASSUME macro expands out to:
do {} if (false && (ok))
That isn't valid C code. Fixed by adding an elif for MSVC that uses __assume(0), which is similar to gcc's __builtin_unreachable according to http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0627r0.pdf.
* WIP.
* WIP.
* Tests are passing.
* Recover some perf: LIKELY doesn't propagate through functions. :(
* Added some more benchmarks.
* Simplify & optimize upb_arena_realloc().
* Only add owned blocks to the freelist.
* More optimization/simplification.
* Re-fixed the bug.
* Revert unintentional changes to parser.rl.
* Revert Lua changes for now.
* Revert the arena fuse changes for now.
* Added last_size to the arena representation.
* Fixed compile errors.
* Fixed compile error and changed benchmarks to do one allocation.
New code is smaller (in both source size and compiled size) and faster.
# Speed
The decoder speeds up on all machines I tested, though the amount of speedup varies. I was only able to test Intel CPUs.
### Linux Desktop
```
CPU: Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
OS: Linux
name old time/op new time/op delta
CreateArena 4.72ns ± 0% 4.93ns ± 0% +4.47% (p=0.000 n=11+11)
ParseDescriptor 12.4µs ± 1% 9.1µs ± 1% -26.65% (p=0.000 n=11+11)
```
### Mac Laptop
```
CPU: Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
OS: macOS
name old time/op new time/op delta
CreateArena 5.33ns ± 3% 5.58ns ± 2% +4.69% (p=0.000 n=12+12)
ParseDescriptor 15.0µs ± 2% 11.9µs ± 2% -20.20% (p=0.000 n=12+12)
```
### Linux Workstation
```
CPU: Intel(R) Xeon(R) Gold 6154 CPU @ 3.00GHz
OS: Linux
name old time/op new time/op delta
CreateArena 5.29ns ± 0% 5.52ns ± 0% +4.37% (p=0.000 n=10+12)
ParseDescriptor 18.6µs ± 0% 16.4µs ± 0% -11.54% (p=0.000 n=12+12)
```
# Size
A few source files grow marginally because of some arena functionality moved inline. But `upb/decode.c` shrinks by 30% on Linux:
```
VM SIZE
--------------
+2.1% +283 upb/json_decode.c
+24% +205 upb/msg.c
+8.4% +115 upb/upb.c
+0.9% +28 upb/reflection.c
[ = ] 0 upb/def.c
[ = ] 0 upb/encode.c
[ = ] 0 upb/json_encode.c
[ = ] 0 upb/table.c
-30.3% -1.51Ki upb/decode.c
-0.7% -738 TOTAL
```
upb/json/parser.rl: In function 'end_member.isra.150':
bazel-out/k8-opt/bin/upb.c:5536:13: error: 'sel' may be used uninitialized in this function [-Werror=maybe-uninitialized]
upb_func *ret = (upb_func *)h->table[s].func;
Joshua Haberman
Protobuf Team Bot
Eric Salo
Joshua Haberman
Esun Kim
Igor Kostenko
Charlie Savage
Anna R
Nicolas "Pixel" Noble