Age | Commit message (Collapse) | Author |
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@374295 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@374156 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Summary:
This patch adds support for testing clang/LLVM against the GCC Torture
suite.
This patch adds the CMake configuration and licence information for these tests. A follow-up patch will add the testcases themselves (which are too large to review, and included without modifications). They will be committed together.
Reviewers: hfinkel, kristof.beyls, asb
Reviewed By: kristof.beyls
Subscribers: khcheang, mehdi_amini, jvesely, krytarowski, fedor.sergeev, zzheng, steven_wu, dexonsmith, arphaman, jfb, mstorsjo, lewis-revill, simoncook, s.egerton, riccibruno, asb, mgorny, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D66887
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@374155 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This patch adds necessary flags for the test stdthreadbug.cpp, so it can
be executed without error, when tests are compiled with -static flag.
Differential Revision: https://reviews.llvm.org/D52878
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@352385 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
For details, see r351675 which intalled the new license for the test
suite, and r351676 which handled headers in the ABI test suite.
This also involved cleaning up some weirdly mis-formatted headers in the
glib compat stubs.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@351677 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Summary: Here is the next bunch of avx512 tests. In each test we do some load, store or move operations and also check result.
Reviewers: MatzeB, craig.topper, zvi, RKSimon
Reviewed By: RKSimon
Subscribers: mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D51599
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@350816 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Summary: Here is another bunch of avx512 tests. In each test we do miscellaneous operations(shuffles, compress, shifts and etc) and also check result.
Reviewers: craig.topper, RKSimon, MatzeB, zvi
Reviewed By: RKSimon
Subscribers: mgorny, eraman, llvm-commits
Differential Revision: https://reviews.llvm.org/D51621
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@349334 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
No functional change intended for other OSes.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@349273 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Summary:
This patch separates current avx512 tests to avx512f, avx512bw and other instruction set directories.
This allows to specify tests sets on specific avx512 hardware, for example, knl.
Reviewers: RKSimon, MatzeB, craig.topper
Reviewed By: RKSimon
Subscribers: mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D55603
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@349039 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
The forward declared protocols are no longer valid in @protocol expressions.
rdar://43516174
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@340196 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
There are various variables leftover from the Makefile build conversion
that aren't actually used in the cmake build. Remove them:
- OBJS, XCORE_TARGET_NEEDS_MEMORY, REQUIRES_EH_SUPPORT, DIFFPROG,
INCLUDES, JIT_OPTS, DISABLE_JIT were never respected by the cmake system.
- PROG is only used by llvm_multisource() not from llvm_singlesource()
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@338788 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Misc-C++-EH test fails when compiling with -static option. Adding
-pthread flag resolves the linker error.
Differential Revision: https://reviews.llvm.org/D49538
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@338515 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This benchmark wants to use vector::operator[] to form a pointer-past-the-end.
This is not valid to do via operator[], because doing so requires forming an
(intermediate) reference-past-the-end (dangling references are always UB, simply
by existing). Forming address-past-the-end is valid under pointer arithmetic
(basic.compound p3.2) and iterators (iterator.requirements.general p7), however.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@331314 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Summary:
Here is the bunch of tests. In each test we do some operations with intrinsics and also check result.
This set depends on cpu features, so it needs support to identify avx512 features in cmake and makefile approaches of test-suite. Here are patches for makefile and cmake: [[https://reviews.llvm.org/D38182|D38182]], [[https://reviews.llvm.org/D38484|D38484]]
Reviewers: craig.topper, zvi, MatzeB
Reviewed By: MatzeB
Subscribers: llvm-commits, mgorny
Differential Revision: https://reviews.llvm.org/D41249
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@321144 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This modifies the build scripts so that the test-suite should still pass when
Clang switches its default dialect to C++14. This pretty much just involves
explicitly specifying an older version because (particularly with benchmarks)
we don't want to get involved in hacking the source.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@320249 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Differential Revision: https://reviews.llvm.org/D38962
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@316918 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
checks should also disable the benchmark's code
This fixes an issue with our
http://lab.llvm.org:8080/green/job/Testsuite-Tester/ bot that couldn't compile
the benchmark because of __builtin_bitreverse uses.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@311079 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Differential Revision: https://reviews.llvm.org/D35188
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@311010 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
When running the test-suite on AArch64 at -O0, we sometimes get timeouts
in stepanov-v1p2. It usually takes more than 450s to run it, and the
threshold is only 500s. Increase it to 750s.
We're modifying the Makefiles because this is what the AArch64 buildbots
currently use.
Similar to r305671, which increased the timeouts for PAQ8p and SciMark-2.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@308165 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
There are more external suites added nowadays which all require
traditional_output = False, so switch the default to False and rather
set it explicitely to True for SingleSource/MultiSource and some
External/XXX tests.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@302959 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
When we removed -faltivec from Clang in favor of just supporting -maltivec, we
missed updating the test-suite CMake file.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@299826 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@296896 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This is a regression test benchmark for indirect branch critical edge
splitting, based on the python eval loop.
Not splitting these edges can cause a roughly 3x slowdown on x86-64.
Differential Revision: https://reviews.llvm.org/D30313
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@296044 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
- Include the minus sign in the PREFIX name instead of having it
magically be appended by SingleMultiSource.cmake.
- Make sure prefix spelling corresponds to benchmark directory name.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286276 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Those two variables were set but never used (coming from the automatic
Makefile->CMakeLists.txt conversion).
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286200 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
The former works as well for us and is more in line with usual cmake
style.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286199 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Nothing in the current cmake build does something with this variable
(and so far nobody has complained either). Remove the variable, we can
always bring it back when we have a system in place.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286198 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Instead of having additional PROGRAMS_TO_SKIP magic that filters
benchmarks after the fact, we rather explicitely construct the list of
benchmarks to run with normal cmake commands.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286197 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
They have been used in the Makefile build, but the cmake build never
sets them.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286196 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
The initial Makefile->cmake conversion missed *.m/*.mm files.
Add an option to llvm_singlesource() to specify the set of files to
test.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286195 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Get rid of remnants from the automated Makefile->cmake conversion.
Nobody noticed because somehow this whole directory isn't included
anyway.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286180 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
CPPFLAGS are for the C-PreProcessor; codegen and warning flags need to
go to CFLAGS or CXXFLAGS.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286178 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
- People familiar with cmake already know add_subdirectory() so they do
not need lookup whether llvm_add_subdirectories() has additional
semantics.
- Replacing one-line lists with list over multiple lines sorted in
alphabetical order is merge friendly.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286143 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Perform add_subdirectory() calls before setting up
singlesource/multisource benchmarks to avoid settings of the parent
directory influencing independent benchmarks in the subdirectory. As far
as I can tell there is no problem today, but we should get used to the
safe style.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@286142 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
These attributes do not have the effect that I was hoping to achieve:
that is to remove the effects of -ffast-math on the strictFP functions.
See discussion at https://llvm.org/bugs/show_bug.cgi?id=30745
To fix the polybench for -ffast-math we will have to either add another function
attribute to disable fast-math, or split compilation, or disable the tests when
the compiler flags contain fast-math.
This patch fixes two tests in the build-bots for Mips: most likely a mips
back-end problem with optnone:
Nikola Fejsov <Nikola.Fejsov@rt-rk.com>
Oct 17 (9 days ago)
After commit for r284354, number of failing tests dropped to 2. They are:
SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.execution_time
SingleSource/Benchmarks/Polybench/linear-algebra/solvers/gramschmidt/gramschmidt.execution_time
You can see which configurations have problems on following link: http://www.rt-rk.com/yamb/waterfall
[...]
Yes, with __attribute__ commented out it is working fine.
We have been able to narrow the area where error occurs. When code reaches line
p[i] = 1.0 / sqrt(x) in first function, sqrt.d assembler instruction is used to
calculate square root. However, when it reaches same line in second function
(strictFP), instead of using sqrt.d instruction, program tries to use sqrt
function from libm. While doing it, it somewhere overwrites t9 register and
there it breaks. We are not sure why this happens, we will try to investigate
further.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@285230 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
I have missed to add the correct tolerance in a few places, printing
the strictFP array, and avoid useless computation in init_array.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@284480 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@284354 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This patch makes each test in polybench to run twice: once with the flags
specified by the user, and once with strict FP computations: optnone and
fp-contract=off. The results are then compared in memory with FP_ABSTOLERANCE.
If there are differences, the program stops returning 1. The output of the
strict FP computation is printed in a file and compared against a hashed
reference output.
With this patch the polybench will pass when the test-suite is compiled with
-ffp-contract=on.
Differential Revision: https://reviews.llvm.org/D25346
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@284334 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
.. and provide a way for users to find out generated executable name.
This makes it possible to further customize build process for particular
target in the user's CMake files.
Differential Revision: http://reviews.llvm.org/D20842
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@280738 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
We need to add -faltivec, to enable Altivec language extensions, to compile the
test suite unit tests in SingleSource/UnitTests/Vector/Altivec. That should be
all that is necessary to the CMake configuration, as -faltivec also implies
-maltivec.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@280514 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This prevents name clashes for test targets CMake generates on arm.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@267439 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Differential Revision: http://reviews.llvm.org/D19423
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@267243 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264907 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264900 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
This reverts most of the cmake hackery from r264499, instead
communicate to lit which tests are singlesource tests so it can narrow
down the compiletime search to files starting with the benchmark name.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264617 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264498 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
The cmake build has never set ARCH to x86_64 or x86_64h, so no need to
check for it (things work fine without).
This way we can get rid of the llvm_filter() function.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264496 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@264101 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@263615 91177308-0d34-0410-b5e6-96231b3b80d8
|
|
Instead of trying to list every benchmark in the subdirectories in
PROGRAMS_TO_SKIP (and failing to really list them all) simply do not
include the whole Regression directory in BENCHMARKING_ONLY mode.
git-svn-id: https://llvm.org/svn/llvm-project/test-suite/trunk@263164 91177308-0d34-0410-b5e6-96231b3b80d8
|