aboutsummaryrefslogtreecommitdiffstats
path: root/stm-arm-neon-ref.h
Commit message (Collapse)AuthorAgeFilesLines
* Make arm-neon-tests buildable with optimization.Victor Khimenko2016-10-191-1/+1
| | | | | | | | | | | | | | If arguments are not marked as volatile then Clang sometimes manages to elide intrinsics entirely which makes test useless as CPU test. Worse: if saturating intrinsic is eliminated then nothing touches FPSCR.Q anymore thus tests fail. Test: tests ARE tests! Bug: 27353392 Change-Id: Iaf555a068118250d12d7cc259c4124dfcf90601e
* Make sure FP16 tests are enabled only if the target actually supports them.Christophe Lyon2016-08-241-20/+20
| | | | | This implies checking __ARM_FP or __ARM_NEON_FP16_INTRINSICS to account for different compilers.
* __set_neon_cumulative_sat() modifies the contents on the QC flag, andChristophe Lyon2015-01-201-16/+21
| | | | | | | | | | | | | some intrinsics do so too: this patch adds the explicit dependency on the asm statement, to avoid code reordering or removal. When writing QC, the asm statement now has a fake input dependency, which is the output of the intrinsic being tested. Modifying the __set_neon_cumulative_sat macro is necessary, to be able to accept all the possible input types. Update the generic code in ref_v_binary_sat_op.c and ref_v_unary_sat_op.c accordingly, as well as all the tests involving QC.
* Rename helper macros:Christophe Lyon2014-12-161-2/+2
| | | | | TEST_VLOAD into VLOAD TEST_VDUP into VDUP
* Fix detection of FP16 support: test __ARM_FP16_FORMAT_IEEE instead of ↵Christophe Lyon2014-09-021-23/+21
| | | | __ARM_NEON_FP16_INTRINSICS.
* Make sure poly* types are handled as unsigned when they get printed.Christophe Lyon2014-07-111-8/+20
|
* Remove floating-point outputs to avoid miscomparison of results when using ↵Christophe Lyon2014-07-111-1/+1
| | | | different libC, some of which can output "-nan", the others always "nan".
* Dump fp16 values in expected_input4gcc.txt too.Christophe Lyon2014-07-111-10/+13
| | | | Add expected_input4gcc-nofp16.txt.
* Support Neon Cumulative Saturation flag for aarch64 too.Christophe Lyon2014-07-101-14/+30
|
* Renamed Neon overflow to Neon cumulative saturation, since it's the real name.Christophe Lyon2014-06-031-20/+21
|
* Add new output file, to make conversion to GCC dejagnu testsuite easier.Christophe Lyon2014-05-161-3/+62
|
* Added support for half-precision (fp16) variants.Christophe Lyon2013-04-111-0/+74
|
* Fix typo in poly16_t printf format.Christophe Lyon2013-04-091-1/+1
|
* Add support for polynomial variants (*_p8, *_p16).Christophe Lyon2013-03-291-3/+53
|
* Reorganize input data buffer to try to detect some out of bounds accesses.Christophe Lyon2013-01-171-32/+144
| | | | Add dedicated input for vdup, vld1_dup, vldX, vldX_dup, vldX_lane, vstX_lane.
* Fix input data for vld3 and vld4 tests.Christophe Lyon2013-01-141-6/+9
| | | | Reported by Victoria Zhislina.
* Improve support for endianness of the target processor. (eg when compiling ↵Christophe Lyon2012-08-311-1/+15
| | | | with llvm).
* Add accessor functions for Neon_Overflow, such that its value can beChristophe Lyon2012-05-111-5/+19
| | | | read/changed when compiling with GCC/ARM too.
* Fix definition of NAN, INF under Win32.Christophe Lyon2011-10-171-4/+7
| | | | Fix build with GCC/ARM.
* Define NAN and HUGE_VALF to be compatible with MSVC.Christophe Lyon2011-10-031-0/+3
|
* Merge from SVN #539Christophe Lyon2011-07-191-1/+1
| | | | [CL] force precision of FP output to help comparison when compiled with MSVC
* Merge from SVN #532Christophe Lyon2011-07-191-0/+4
| | | | [CL] include msinttypes.h/msstdint.h instead of inttypes.h/stdint.h when compiling with MSVC
* Merge from SVN #514Christophe Lyon2011-07-191-3/+9
| | | | [CL] fixes to support compilation with a C++ compiler
* First public release of the ARM/Neon tests.Christophe Lyon2011-01-241-0/+438