The initial configure & generate of a build on Windows should have
the necessary VCVARS set in the environment. After that, we should be
incrementally building with those same vars to avoid conflicts between
Windows SDKs.
Cache these vars so that the same Windows SDKs are consistently used
in future incremental builds.
Due to the horrible attrocities against software of the attempt to perform
cross-compilation in the swift build system, we need to emulate the linking
behaviour for Windows with the link against the import library. The emulation
requires the custom creation of import library targets. In order to actually
get the linking semantics correct, the dependendency targets must be created
prior to use (unlike standard CMake). The reordering ensures that we get
correct linkage when building for Windows.
Perform a simple optimization to avoid a number of string comparisions for the
host system.
Normally, the C++ shared library would have link against the C++ ABI
shared library, but the Android NDK does not distribute the later, so
one need to link manually against the static C++ ABI from the NDK.
This will ensure that additional target executables can not be added to the rest
of the swift project without anyone noticing since the non-stdlib parts of
Swift's cmake will not have visibility of the declaration unless they change the
cmake lookup paths.
`-fblocks` is a core driver option now, so it can be used with both the GCC
style driver as well as the cl style driver. Simplify the logic for the
handling of this option.
The swift image registrar uses the extension `.obj` as is traditional on
Windows. Ensure that we get the extension correct when cross-compiling with the
Swift specific cross-compilation system.
The Android targetted libraries already link (manually) with the Android
NDK C++ libraries. When using Clang and lld, an extra link to libc++ is
requested, which will fail because Android names it libc++_shared.so
instead. To avoid looking for the wrong name, and since the library is
referenced manually either way, request no linking with the standard C++
libraries (but add the math library in, which is implicitly linked when
C++ is linked, and it is needed for Glibc at least).
Ensure that the use the target specific names for the fat libraries for
non-MachO targets which do not support fat libraries. This fixes the
windows build.
Attempt to repair the build for the unified swift build. This allows us
to build a single unified toolchain with swift support. In this layout
assume that cmark and clang are peers of LLVM rather than located in
`tools`. Doing so allows a uniform layout of the tree and a simpler
build approach.
gold is more strict than lld about the order of the arguments, that's
why CMake offers two different properties for the linker: LINK_FLAGS and
LINK_LIBRARIES. The former _add_variant_link_flags was adding the
libraries to LINK_FLAGS, when the correct thing is to add them to
LINK_LIBRARIES.
The change adds a new output variable for _add_variant_link_flags which
will containt the linked libraries, and CMake will be able to generate
the correct command line invocation for when gold is used.
This should fix the Android CI build.
Windows does not link against the library but the import library. When
building the target specific bits, we unfortunately do not use the cmake
build infrastructure properly. This results in us trying to link
against libraries which do not exist. Redirect the link to the right
files. This allows us to build swift-reflection-test.
This reverts commit 121f5b64be.
Sorry to revert this again. This commit makes some pretty big changes. After
messing with the merge-conflict created by this internally, I did not feel
comfortable landing this now. I talked with Saleem and he agreed with me that
this was the right thing to do.
When building on case insensitive filesystems, there is no need to
create the library symlink forest as the paths will be resolved properly
due to the insensitivity. This avoids a bit of work and spew on
Windows.
This change could impact Swift programs that previously appeared
well-behaved, but weren't fully tested in debug mode. Now, when running
in release mode, they may trap with the message "error: overlapping
accesses...".
Recent optimizations have brought performance where I think it needs
to be for adoption. More optimizations are planned, and some
benchmarks should be further improved, but at this point we're ready
to begin receiving bug reports. That will help prioritize the
remaining work for Swift 5.
Of the 656 public microbenchmarks in the Swift repository, there are
still several regressions larger than 10%:
TEST OLD NEW DELTA RATIO
ClassArrayGetter2 139 1307 +840.3% **0.11x**
HashTest 631 1233 +95.4% **0.51x**
NopDeinit 21269 32389 +52.3% **0.66x**
Hanoi 1478 2166 +46.5% **0.68x**
Calculator 127 158 +24.4% **0.80x**
Dictionary3OfObjects 391 455 +16.4% **0.86x**
CSVParsingAltIndices2 526 604 +14.8% **0.87x**
Prims 549 626 +14.0% **0.88x**
CSVParsingAlt2 1252 1411 +12.7% **0.89x**
Dictionary4OfObjects 206 232 +12.6% **0.89x**
ArrayInClass 46 51 +10.9% **0.90x**
The common pattern in these benchmarks is to define an array of data
as a class property and to repeatedly access that array through the
class reference. Each of those class property accesses now incurs a
runtime call. Naturally, introducing a runtime call in a loop that
otherwise does almost no work incurs substantial overhead. This is
similar to the issue caused by automatic reference counting. In some
cases, more sophistacated optimization will be able to determine the
same object is repeatedly accessed. Furthermore, the overhead of the
runtime call itself can be improved. But regardless of how well we
optimize, there will always a class of microbenchmarks in which the
runtime check has a noticeable impact.
As a general guideline, avoid performing class property access within
the most performance critical loops, particularly on different objects
in each loop iteration. If that isn't possible, it may help if the
visibility of those class properties is private or internal.
