summaryrefslogtreecommitdiff
path: root/compiler/optimizing/codegen_test_utils.h
diff options
context:
space:
mode:
authorAart Bik <ajcbik@google.com>2017-07-20 16:47:45 -0700
committerAart Bik <ajcbik@google.com>2017-07-20 17:03:35 -0700
commitf60525793a1fd784ce7de82f18e7ad9de242c431 (patch)
tree5b63ffbc991a2b80ef8e779b975cf737c6d5ab3c /compiler/optimizing/codegen_test_utils.h
parent8fe41cf9c995d321da607fa45ad0e752444696a1 (diff)
Assume all x86/x86_64 hosts support at least sse4.x/popcount.
Note: a small change with potentially big consequences please review carefully Rationale: Since all true Android x86/x86_64 devices and all x86/x86_64 hosts support at least sse4.x/popcount, ideally we would simply drop support for anything "lower". However, since the emulator running on x86/x86_64 only supports as much as the underlying host (a bad design to start with!), we cannot do the full drop until emulator support includes sse4.x/popcount. This CL is the next best thing to get much more SIMD testing "out there" It assumes that all x86/x86_64 hosts support at least sse4.x/popcount. This firstly implies that the ART runtime binary itself may exploit these features. Secondly, this implies thatthe ART runtime passes these feature flags to dex2oat and JIT by calling the method InstructionSetFeatures::FromCppDefines(). Since invoking dex2oat directly does not pick up these flags, cross-compiling from a x86/x86_64 host to a x86/x86_64 target should not be affected. Bug: 63585550 Test: test-art-host and test-art-target (FUGU) Change-Id: I93736cd7b781e4432f3cea036368e230af3d03d0
Diffstat (limited to 'compiler/optimizing/codegen_test_utils.h')
0 files changed, 0 insertions, 0 deletions