| Age | Commit message (Collapse) | Author |
|---|
|
Bug: 37670543
Bug: 36006390
Bug: 37679566
Bug: 37636434
In debug context, a script is forced to be recompiled every time it
is initialized.
To avoid the same .so file being written to by another thread,
while it is being loaded and used by one thread, do not save the .so
file. Delete it right after loading it.
Test: RefocusTest and RSTest (including ScriptGroup2 tests) with
debug.rs.debug set to 1 and CTS on Angler
Change-Id: If63e3d21e3d9abd007a66e0ec79c9e6f1c9f13a0
|
|
parts of libRS_internal.
NOTE: we're resolving dependencies to provide a model for vendors.
For us, all this code is above the HAL, thus the dependencies
are acceptable; whereas for vendors, their equivalent of this code is
below the HAL, and so the dependencies are not acceptable.
This CL resolves the libutils dependency by:
- Implement the timings functions in android::renderscript namespace
using NDK APIs, instead of using libutils counterparts.
- Replace android::Vector and android::String8 by std::vector and
std::string.
- PROPERTY_VALUE_MAX is replaced as PROP_VALUE_MAX.
This CL didn't resolve the libutils dependency of rsFont.cpp and
rsDebugger.cpp in libRS_internal:
The dependent functionality in rsDebugHelper.h is off by default, and
only intended for use during development; and rsFont.cpp is part of
graphics API which is not implemented below the HAL and is not used as
a model by vendors.
Additionally, this CL fixed the bug that mNodes was sorted in a
decreasing order. Nodes in ScriptGroup should be executed in
ascending order. The bad sort was only for support lib; so there was a
previously-unknown bug in support lib implementation of ScriptGroup.
Background:
libutils contains a collection of things like Vector, String8,
CallStack. It served the purpose similar to a STL library, when there
was no stable STL implementation available in Android. And most
importantly, it is not part of NDK.
Support lib used to use our own implementations of android::Vector and
android::String8, because it can only depend on NDK, similarly for the
timing related functions.
As part of the Treble requirements, native RS, including vendor version
libRS_internal, libRSDriver, libRSCpuRef could only depend on NDK
too. So we need to break the dependency on libutils. And since we now
have reasonable support of STLs, we should use that instead.
Bug: 34396220
Test: mm, and all CTS tests pass on Bullhead;
RsTest and RSTest_CompatLib (both native and compat path) also pass.
Change-Id: Ib9a37d16235c1dcd0f5bae3b95c374e394483c91
|
|
- Implement property_get using __system_property_get, remove
dependencies on cutils/properties.h.
- remove rsCompatibilityLib.cpp since it only contains propery_get, and
it is now in rsCppUtils.cpp.
- Disable ATRACE. The systrace is current done in Java level. If we
want to do it in the driver, we need to have our own implementation
similar to native/utils/trace.cc
- Remove CC_LIKELY macro.
Bug: 34396220
Test: build
Change-Id: I6a6cbcfb2ca15b9d1f09b9ce0579d39fdbe98727
|
|
- The RS_SERVER macros were intended to have a version of RenderScript
running on the desktop. It has not fully worked for a very long time.
Bug: 35473587
Test: build and CTS tests pass
Change-Id: I0da3fc9d4601951442384d05f9dbecaeddeff8cc
|
|
Just fix a few include paths.
Bug: 34265954
Test: Compiled everything, ran a regular and a support lib app.
Change-Id: I5d608e58734e50c0d2ac895290ea2a66ebe46345
|
|
This change fixes a few typos and clang tidy warnings related to the
"llvm-namespace-comment" checks.
Bug: 26936282
Test: WITH_TIDY=1 WITH_TIDY_CHECKS="llvm-namespace-comment" mm
Change-Id: Ic65182e5b4999fbd48d6a8ad7172e4bfeeb541f4
|
|
In that mode libraries could be found in /system/lib/arm, not in
/system/lib or /system/lib64
Test: cts-tradefed run cts -s 77DCDB7A -m CtsRsCppTestCases
BUG=27783760
Change-Id: I31227b281637b93c75b507b03cb4e71a31d670a0
|
|
Bug: 29772332
Fixed code making reference to internal data of a temporary object.
Change-Id: I940cce17cd5ada4bb33fa63fde8a63a9038778a8
|
|
Bug: 27298560
Change-Id: I8c3d568e98aaf0b7d86881c985d13ed5b8e95338
(cherry picked from commit ae2ec3febedfc29376b9104413fb4042028f1265)
|
|
Bug: 28463187
Launch options array[2-4] are not initialized by the support library
in the current SDK.
They may have arbitrary values that invalide the checks that this CL
is removing, and cause a kernel launch to bail out unnecessarily.
This CL removes those checks, since the array.*Start and array.*End
fields in the launch options are not used anywhere in the current
runtime.
Change-Id: Ib8a2e551619f770b402f569acbde635ee047675a
|
|
overhaul.
A reduction kernel is eligible for multithreaded execution if it has a
combiner function and it is launched over a 1D iteration space.
Note: Properties debug.rs.reduce-accum and debug.rs.reduce-split-accum
are added for debugging multithreaded reduction.
The following changes are made to reduction tests in RsTest:
- Overhaul the test framework -- now data-driven, and can execute the same
test multiple times with different seeds and input sizes, features
separate sets of quick correctness tests, full correctness tests,
and performance tests. (Performance tests are not run by default.)
- Report timing information for test execution.
- Report more information for fz* kernel testing.
- Remove dp kernel testing -- this involved floating-point arithmetic
which is not guaranteed to produce identical results between java
and rs or for different rs multithreaded executions.
- Add sumgcd kernel testing. This is intended to be representative of
a compute-heavy kernel.
- findMinAndMax kernel testing must compare cell value not cell index
-- two or more cells might have the same min or max value, and java
and various rs multithreaded executions are not guaranteed to find
the same cell.
- Fix bug in findMinAndMax kernel's combiner function. (It behaved
incorrectly when operating on an accumulator datum that has been
initialized but never passed to the accumulator function.)
- RsTest now requests largeHeap.
Bug: 27299475
Change-Id: I58f99c21389dbae5c8e3ad85d98700dc165664bb
(cherry picked from commit 10adb0c2029f112b5738228617d5645f6ecea0c5)
|
|
Bug: 27874893
Bug: 27925095
rsClearObject() may be called on rsContextDestroy() in the JNI thread,
where TLS is not initialized.
This also reverted Icca50fade1199b6f530f963e9e694366b7d8cf67 that
made rsContextDestroy() go through the FIFO. rsContextDestroy() must
happen off the FIFO, since it joins the FIFO consuming thread.
Both the rsClearObject() change and the rsContextDestroy() revert need
to go together to make RenderScript and Camera CTS tests pass. Hence
this single CL for both changes.
Change-Id: I5f1d91a4cc1ce9bf2b3386fbef6f1171638b5266
|
|
Bug: 27439842
An RS intrinsic relies on RSExpandKernelDriverInfo.usr for the "this"
pointer to the ScriptIntrinsic object.
Need to to correctly set that field in a script execution for intrinsics.
Also fixed a bcc command-line option.
Added a test of script group with intrinsics to ImageProcessing GroupTest(native).
Change-Id: Ia8b45bca3396ac987ba167fd1709ee89123a2557
|
|
Bug: 27367378
Change-Id: Ide29c702d7a70dfe087ddc600d3fd17279223196
|
|
|
|
Requires coordinated change in frameworks/base.
Requires coordinated change in frameworks/compile/libbcc in order
for RsTest to run.
At present, general reduction kernels are run single-threaded.
Also: Remove dead struct field MTLaunchStructForEach::sig.
Bug: 23535724
Change-Id: Ice17ccf20a902f8a106eaa62ec071d46e3c0ad8c
|
|
Add a new libclcore_g.bc which is functionally equivalent to
libclcore.bc but is compiled with debug information and -O0.
To achieve this some allocation functions had to be written in C,
because they were previously available in LLVM IR without debug
information.
The RenderScript cts has been run to verify these changes, both as it
is and forcing it to link with the new libclcore_g library. Both runs
show no regressions.
Change-Id: Iffc3c54fbf69b9a43703d3c2776101ef881deffd
Signed-off-by: Verena Beckham <verena@codeplay.com>
|
|
reference implementation.
Bug: 22631253
For now, this just runs a serial reduction on one thread.
Change-Id: I34c96d24bb6f44274de72bb53160abcf79d143b0
|
|
Add a flag RS_CONTEXT_OPT_LEVEL_0 which sets a field mOptLevel in the Context
to 0 when calling bcc.
Using this flag will result in an object file that is easier to debug.
Change-Id: I9673506710cc9f34c415b694fe5d6bc7e967f1db
Signed-off-by: verena beckham <verena@codeplay.com>
|
|
b/21949580
Writing different contents to an existing .so file, which is currently
open, may corrupt its globals and code. Choose a different file name
instead.
Change-Id: I3e649b09a55b43339283aa1c46f2844c66434c17
(cherry picked from commit 9844cffc8af28c4829012bc0c48f85d04b671f88)
|
|
Bug: 20894664
The system partition is read-only, and can/should be compiled correctly
during the offline creation of the system image. Since we cannot replace
these precompiled blobs (short of app update/OTA), there is no reason or
correct way to validate/replace the checksum.
Change-Id: Ia66bfdbe178bf215e146c3699f5bc7804222e978
|
|
|
|
Change-Id: Ibe8725074724b75e35c25a404daaba07ffbca2ab
|
|
Change-Id: I17c701dfda5b3031c1fc50bfc186e4b7f8b7ac1e
Signed-off-by: Yong Chen <yong.a.chen@intel.com>
|
|
* changes:
Update RS driver to support extraction of global variable properties.
Use "override" instead of "virtual" when replacing methods.
|
|
Bug: 20306487
This patch adds some new enums to classify properties (such as "static",
"constant", and "pointer") for global variables. The reference driver
is also extended to provide methods to examine these properties (when
the bitcode is compiled with bcc).
Change-Id: I331756f8a8990caf5ebdf85599060434a7cfdcb7
|
|
Bug: http://b/20501816
Change-Id: I63a7c690bc5bde3dc1354f6e04ad8d32b3b43015
|
|
|
|
The existing linker path hard-coded "-lRSDriver" into the command line, but
this won't work for partner drivers that have a different implementation
loaded. In order to still properly handle use of the CPU driver, this needs
to change depending on whether we actually loaded an OVERRIDE_RS_DRIVER or
not.
Change-Id: I0c4a4f12f5db819b234952bc8f364ac6300f147b
|
|
Bug: 20306487
This change enables vendor drivers to configure support for including
additional information about global variables in the emitted CPU code.
This information includes the number of total global variables, the
names of these variables, the addresses of these variables and the
sizes of these variables. The driver can also select whether the
information includes constant (immutable) globals or not.
The reference driver defaults to embedding information about each of
the existing, non-constant global variables.
Change-Id: I1e55fc3f08e518f04eeee3e4f9dc7b6ea3b80d7c
|
|
bug 19866850
Explicitly set error when a kernel launch fails.
Change-Id: If42419f3cf7557c524afc5bb27a0753460ca7e4e
|
|
Change-Id: I48dafb2bc1dc335a52b289db2981397251f673c8
|
|
|
|
I introduced a separate routine to cacluate checksum for ScriptGroup
in my previous CL, in addition to the one we use for regular scripts.
This CL removes the new one and uses the old one.
While I am on it, I made some other minor changes, e.g., changing
mBuildChecksum in RsdCpuScriptIml from char* to uint32_t, and a few
other minor cleanups in ScriptGroup2 implementation.
Change-Id: I168fdbb4e7bd14f1549a687e7b0d0ca6dd4da866
|
|
Change-Id: I6d6b46c55f3e88a810ebe51def3ebaccb1fd3fa2
|
|
Change-Id: I5f469266fd11b5031d1357f356ac40e00f5c0348
|
|
BUG 19934827
Wrap TEMP_FAILURE_RETRY around system calls that can return EINTR
(waitpid, close).
Refactor fork/exec flows in various places into a utility function
and log errors so we can better understand failures in the test server.
Fix a small use-after-free issue in ScriptGroups.
Change-Id: I60b192f83c395a13c27cd6bd2289c44132b84791
|
|
Also made two other cleanups:
- Changed KernelID/InvokeID into IDBase in class Closure
Rather than having two fields in class Closure, one of type ScriptKernelID and
the other InovkeID, use a single field of the common base class IDBase. Added a
boolean field to indicate whether it is kernel or invoke, since -fno-rtti is on.
- Removed user pointer from CPU closure
Change-Id: I5553f86b2e58325f85649078d48685a38f12d62f
|
|
bug 19216507
If ro.debuggable system property is set,
- Compute the Adler32 checksum of bitcode, compiler arguments, bcc,
libclcore files.
- On shared object load, check that there is an embedded checksum value
in .rs.info and it matches the value computed in the previous step.
The .so is not loaded if the checksums don't match.
- If invoking bcc, pass the checksum as an argument.
If ro.debuggable is not set, a constant checksum is added to .rs.info.
But it is never checked.
Change-Id: I0a9d16b34148b748daf744007a8515511714a5a5
|
|
b/19283946
This move also included the related class SharedObjectUtils.
Change-Id: Iefd587b659c4ad99340c330955b41cd92df45563
|
|
b/19148482
Replaced std::string with android::String8, and std::pair with
android::renderscript::Pair, in header files.
Change-Id: I20ac1a3b7a019e5eee12fac46ad6adf64d850459
|
|
libbcc had a utility function to concatenate string arguments. This
patch moves that function to frameworks/rs. The function will no longer
be needed in libbcc once we move all the checksum/cache invalidation
logic to frameworks/rs.
Change-Id: I2b1c15c2763d90d7b099e41257bb72f200680943
|
|
This also includes support for InvokeID
Change-Id: I5b59df166ea30b309b8dd9623825ac0e72d03856
|
|
64bit Renderscript compat lib."
|
|
for 64bit Renderscript compat lib.
- Native lib path stored in Java and queried from JNI.
- Then set to Context during context creation.
Change-Id: I3c997cf849efb7b9a0b0ac35d5b62289d70e7434
|
|
The runtime will start a seperate process to call the new bcc to fuse kernels.
Change-Id: Ia73ea917a126a5055ec97f13d90a5feaafd6a2f5
|
|
b/19148482
Change-Id: I2138664d8ef51d15da23be69ac7d224f30224f42
|
|
bug 19095896
Read embedded flag in .rs.info and use it to mark Scripts as threadable.
In compat lib path, scripts are always threadable.
Change-Id: I85a3753ffae788dbdec5fd130c20581790c4ee07
|
|
Change-Id: I66cd89b5b44eafa92f391708a06464cd7cdde3ed
|
|
|
