Eric Erhardt [Wed, 27 Jan 2016 23:41:14 +0000 (17:41 -0600)]
TimeZoneInfo.DisplayName values are not localized on Linux
Fixed by calling ICU's ucal_getTimeZoneDisplayName to read the display names for the current locale.
Fix https://github.com/dotnet/corefx/issues/2748
Jan Kotas [Tue, 2 Feb 2016 08:35:22 +0000 (00:35 -0800)]
Merge pull request #2978 from jkotas/prof-breaks
Fix fallout from enabling profiling
Jan Kotas [Tue, 2 Feb 2016 05:40:35 +0000 (21:40 -0800)]
Merge pull request #2976 from krytarowski/netbsd-support-27
Fix build on NetBSD: Introduce check for pthread_getcpuclockid(3)
Jan Kotas [Tue, 2 Feb 2016 05:38:20 +0000 (21:38 -0800)]
Merge pull request #2977 from krytarowski/netbsd-support-28
Prefer PAL_PTRACE() over hardcoded FreeBSD-style ptrace(2) calls
Jan Kotas [Tue, 2 Feb 2016 05:37:48 +0000 (21:37 -0800)]
Merge pull request #2972 from krytarowski/netbsd-support-24
Fix build on NetBSD: Don't increment iterator twice in for(;;)
Jan Kotas [Tue, 2 Feb 2016 05:34:40 +0000 (21:34 -0800)]
Merge pull request #2971 from krytarowski/netbsd-support-23
Add NetBSD target in src/pal/src/CMakeLists.txt
Kamil Rytarowski [Tue, 2 Feb 2016 01:58:56 +0000 (02:58 +0100)]
Fix build on NetBSD: Introduce check for pthread_getcpuclockid(3)
This function is part of the optional part of POSIX.
It's available in:
- FreeBSD,
- AIX,
- QNX,
- Linux.
It's missing in NetBSD.
According to Jan Kotas (Microsoft) this function is used for undocumented
resource monitoring feature (FEATURE_APPDOMAIN_RESOURCE_MONITORING) and
unlikely anybody is using this today.
Solution suggested in #2947 "NetBSD doesn't implement pthread_getcpuclockid(3)"
Jan Kotas [Tue, 2 Feb 2016 02:13:41 +0000 (18:13 -0800)]
Fix fallout from enabling profiling
- Compiler optimization assumption invalid: EE expects method to exist: System.StubHelpers.StubHelpers:ProfilerBeginTransitionCallback
- Add stubs for ARM
Kamil Rytarowski [Tue, 2 Feb 2016 02:13:02 +0000 (03:13 +0100)]
Prefer PAL_PTRACE() over hardcoded FreeBSD-style ptrace(2) calls
Jan Kotas [Tue, 2 Feb 2016 01:20:28 +0000 (17:20 -0800)]
Merge pull request #2953 from janvorli/fix-unix-readytorun
Fix ReadyToRun on Unix
Kamil Rytarowski [Tue, 2 Feb 2016 01:18:39 +0000 (02:18 +0100)]
Fix build on NetBSD: Don't increment iterator twice in for(;;)
If I'm reading it correctly, we will never catch the ".cormeta" value.
Kamil Rytarowski [Tue, 2 Feb 2016 01:10:43 +0000 (02:10 +0100)]
Add NetBSD target in src/pal/src/CMakeLists.txt
Kamil Rytarowski [Tue, 2 Feb 2016 01:05:02 +0000 (02:05 +0100)]
Sort platforms alphabetically in src/pal/src/CMakeLists.txt
Jan Kotas [Tue, 2 Feb 2016 00:56:33 +0000 (16:56 -0800)]
Merge pull request #2962 from dotnet-bot/from-tfs
Merge changes from TFS
Jan Kotas [Tue, 2 Feb 2016 00:43:48 +0000 (16:43 -0800)]
Merge pull request #2958 from jkotas/arm-fixes
Fix build breaks in Linux ARM build
Sivarv [Tue, 2 Feb 2016 00:19:10 +0000 (16:19 -0800)]
Merge pull request #2930 from sivarv/retfix
Fix to issue #2700
Brian Sullivan [Mon, 1 Feb 2016 22:48:53 +0000 (14:48 -0800)]
Merge pull request #2960 from briansull/issue_2925
Remove assert NYI: "Varargs method"
Jan Kotas [Mon, 1 Feb 2016 19:02:31 +0000 (11:02 -0800)]
Fix build breaks in Linux ARM build
- Delete redundant WAITORTIMERCALLBACK declarations. These redundant declarations were needed for very old versions of Windows SDK that are no longer supported.
- Remove unnecessary __cdecl modifiers
- Add default case to switch statement
Jan Kotas [Mon, 1 Feb 2016 21:41:03 +0000 (13:41 -0800)]
Fix Unix build breaks
[tfs-changeset: 1571015]
Brian Sullivan [Mon, 1 Feb 2016 21:39:13 +0000 (13:39 -0800)]
Remove assert NYI: "Varargs method"
Of the 27 test cases that hit this assert 22 of them are passing after this change
The remaining test failures are:
arglist.exe_2994
arglist.exe_2971
arglist.exe_2983
arglist.exe_3006
b16423.exe_4877
Jan Vorlicek [Mon, 1 Feb 2016 16:15:37 +0000 (17:15 +0100)]
Fix ReadyToRun on Unix
While making the ready to run tests run on Unix, I have discovered that the feature
doesn't really work on Unix. First, I have found that the test was generating the
native images into the NI subfolder, but the corerun on Unix doesn't use this
notation and expects these images to live side by side with the IL executables.
So, the test to verify that an image cannot be loaded multiple times was failing,
since the main exe loaded the IL test.dll instead of the native one and so explicit
load of the native test.dll wasn't considered a secondary load of the same image.
After fixing the test, I have found that attempt to load a ready to run image is
failing with an assert.
There were these issues:
- the loaded native image was not marked as relocated
- the EPILOG_WITH_TRANSITION_BLOCK_RETURN macro was missing popping argument registers
- One of the asserts and one getter didn't take into account the fact that the ready to
run native image is not considered to have native header.
- The assembler helpers DelayLoad_Helper and DelayLoad_MethodCall were passing a hex
number as a parameter to the PROLOG_WITH_TRANSITION_BLOCK, but it was using the "h"
suffix notation. There is a known issue in the clang assembler which causes such
suffix to be swallowed so the macro was getting just "10" instead of "10h". The fix
was to use 0x prefix instead.
This change fixes the problem. The ready to run tests pass with it correctly.
dotnet-bot [Mon, 1 Feb 2016 18:48:17 +0000 (10:48 -0800)]
Fix OACR errors for ProjectK
[tfs-changeset: 1570939]
Jan Kotas [Mon, 1 Feb 2016 16:39:26 +0000 (08:39 -0800)]
Merge pull request #2951 from Dmitry-Me/dontAccessDeletedObject
Don't access deleted object, return proper error code
Dmitry-Me [Mon, 1 Feb 2016 10:40:05 +0000 (13:40 +0300)]
Don't access deleted object, return proper error code
Jan Kotas [Mon, 1 Feb 2016 09:59:42 +0000 (01:59 -0800)]
Merge pull request #2949 from Dmitry-Me/castToMostDerivedInterface
Fix cast to interface other than the most derived one
Jan Kotas [Mon, 1 Feb 2016 09:58:33 +0000 (01:58 -0800)]
Merge pull request #2950 from jkotas/WIN32
Stop defining WIN32 in Unix builds
Jan Kotas [Mon, 1 Feb 2016 07:42:27 +0000 (23:42 -0800)]
Stop defining WIN32 in Unix builds
WIN32 was defined for 32-bit Unix builds in number of places which is incorrect. It caused build break in lldb header since they use WIN32 define for Windows-specific code.
Dmitry-Me [Mon, 1 Feb 2016 07:06:35 +0000 (10:06 +0300)]
Fix cast to interface other than the most derived one
Jan Kotas [Mon, 1 Feb 2016 05:42:44 +0000 (21:42 -0800)]
Merge pull request #2948 from krytarowski/netbsd-support-22
Fix build on NetBSD, casting pointer to function to pointer to data
Jan Kotas [Mon, 1 Feb 2016 05:15:27 +0000 (21:15 -0800)]
Merge pull request #2933 from stephentoub/enum_parse
Improve performance of Enum.Parse/TryParse
Kamil Rytarowski [Mon, 1 Feb 2016 02:57:17 +0000 (03:57 +0100)]
Fix build on NetBSD, casting pointer to function to pointer to data
/tmp/pkgsrc-tmp/wip/coreclr-git/work/coreclr/src/vm/method.cpp:5196:16: error: cannot initialize return object of type 'LPVOID' (aka 'void *') with an rvalue of type 'FARPROC' (aka 'long (*)()')
return GetProcAddress(hMod, (LPCSTR)(size_t)((UINT16)ordinal));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
stephentoub [Sat, 30 Jan 2016 19:43:54 +0000 (11:43 -0800)]
Improve performance of Enum.Parse/TryParse
The allocation profile of Enum.Parse today is fairly poor. While the design of the method returning an Enum requires at least one allocation for the boxed result (the generic TryParse could be overhauled to avoid this, but still has it), additional allocations shouldn't be necessary in the common cases. However, for some reason the current code is boxing a 0 on every call. It's also using string.Trim() to remove whitespace (once for the overall string and then once for each substring), and using String.Split to parse multiple values, which ends up allocating a string[] and an int[] even if there's only one value.
This commit removes all allocations from Enum.Parse other than the boxing of the Enum result and some allocations on the code path where Enum.Parse is handed a string containing a number, in which case additional allocations are involved in using the Convert.ChangeType call.
With an enum like:
```C#
[Flags]
enum Colors
{
Red = 0x1,
Orange = 0x2,
Yellow = 0x4,
Green = 0x8,
Blue = 0x10
}
```
using ```Enum.Parse(typeof(Color), "Red")``` repeatedly now results in 5x fewer gen0 GCs and is ~15% faster. Using ```Enum.Parse(typeof(Color), "Red, Orange, Yellow, Green, Blue")``` repeatedly now results in 23x fewer gen0 GCs is is similarly ~15% faster.
Jan Kotas [Mon, 1 Feb 2016 01:52:46 +0000 (17:52 -0800)]
Merge pull request #2940 from krytarowski/netbsd-support-18
NetBSD requires <stdarg.h> for variable argument lists in output.inl
Jan Kotas [Mon, 1 Feb 2016 01:52:25 +0000 (17:52 -0800)]
Merge pull request #2943 from krytarowski/netbsd-support-20
NetBSD doesn't ship with SI_KERNEL
Jan Kotas [Mon, 1 Feb 2016 01:49:16 +0000 (17:49 -0800)]
Merge pull request #2946 from krytarowski/netbsd-support-21
NetBSD: Disable check for minimal stack size for undefined PTHREAD_ST…
Kamil Rytarowski [Mon, 1 Feb 2016 00:10:41 +0000 (01:10 +0100)]
NetBSD: Disable check for minimal stack size for undefined PTHREAD_STACK_MIN
An alternative would be 'sysconf(_SC_THREAD_STACK_MIN)', but it might be no
good reason to prohibit single page stack size (PAGE_SIZE).
Jan Kotas [Sun, 31 Jan 2016 22:23:12 +0000 (14:23 -0800)]
Merge pull request #2520 from sperling/profiler
Enabling profiling on Unix/OS X
Kamil Rytarowski [Sun, 31 Jan 2016 10:17:48 +0000 (11:17 +0100)]
NetBSD requires <stdarg.h> for variable argument lists in output.inl
Kamil Rytarowski [Sun, 31 Jan 2016 19:23:08 +0000 (20:23 +0100)]
NetBSD doesn't ship with SI_KERNEL
SI_KERNEL is available in Linux and FreeBSD. It's unspecified what does it mean.
Linux:
#define SI_KERNEL 0x80 /* sent by the kernel from somewhere */
-- include/uapi/asm-generic/siginfo.h
FreeBSD:
SI_KERNEL signal generated by miscellaneous
parts of the kernel
-- siginfo(3)
Andreas Strid [Mon, 21 Dec 2015 14:33:18 +0000 (15:33 +0100)]
Enabling profiling on Unix/OS X.
Jan Kotas [Sun, 31 Jan 2016 15:31:07 +0000 (07:31 -0800)]
Merge pull request #2939 from krytarowski/netbsd-support-17
NetBSD doesn't ship with <machine/npx.h>.
Jan Kotas [Sun, 31 Jan 2016 15:30:24 +0000 (07:30 -0800)]
Merge pull request #2938 from krytarowski/netbsd-support-16
NetBSD defines _lwp_self(2) in <lwp.h>
Jan Kotas [Sun, 31 Jan 2016 15:22:30 +0000 (07:22 -0800)]
Merge pull request #2935 from jkotas/arm-fixes
Fix build breaks in Linux ARM port
Kamil Rytarowski [Sun, 31 Jan 2016 09:59:53 +0000 (10:59 +0100)]
NetBSD doesn't ship with <machine/npx.h>.
This is first step towards stopping to assume that HAVE_BSD_REGS_T implies
FreeBSD-style code. This breaks on NetBSD in several places.
Perhaps HAVE_BSD_REGS_T should be renamed to HAVE_FREEBSD_REGS_T.
Check headers with this fashion:
- FreeBSD requires <sys/types.h> for <machine/npx.h>
- FreeBSD requires <sys/cdefs.h> for <machine/reg.h>
- NetBSD doesn't need extra includes for <machine/reg.h>
- NetBSD requires <sys/param.h> before inclusion of <sys/cdefs.h>
Jan Kotas [Sun, 31 Jan 2016 01:33:29 +0000 (17:33 -0800)]
Introduce central location for turning FEATURE_EVENT_TRACE on/off
Turn FEATURE_EVENT_TRACE on for x64 Linux only. aarch64 and arm do not
have working lttng in their ubuntu/debian repos. We can expand to other
platforms in the future as they become supported.
Kamil Rytarowski [Sun, 31 Jan 2016 09:49:09 +0000 (10:49 +0100)]
NetBSD defines _lwp_self(2) in <lwp.h>
The _lwp_self(2) function prototype is missing in <sys/lwp.h>.
It's not sufficient to include <sys/lwp.h> or <lwp.h> -- as they are not
including one the other. We need to check for both headers and include them
both.
NAME
_lwp_self - get light-weight process identification
LIBRARY
Standard C Library (libc, -lc)
SYNOPSIS
#include <lwp.h>
lwpid_t
_lwp_self(void);
DESCRIPTION
_lwp_self() returns the LWP ID of the calling LWP.
Jan Kotas [Sun, 31 Jan 2016 04:54:03 +0000 (20:54 -0800)]
Cleanup top-level defines
- Delete left-over cruft from internal build
- Move Windows-specific defines under WIN32
Jan Kotas [Sun, 31 Jan 2016 05:33:12 +0000 (21:33 -0800)]
Merge pull request #2936 from dotnet/revert-2931-glob-revert
Revert "Revert "Add un-prefixed signatures as temporary workaround""
Jan Kotas [Sun, 31 Jan 2016 05:31:34 +0000 (21:31 -0800)]
Revert "Revert "Add un-prefixed signatures as temporary workaround""
Jan Kotas [Sun, 31 Jan 2016 01:40:53 +0000 (17:40 -0800)]
Add missing files for ARM legacy JIT
Jan Kotas [Sun, 31 Jan 2016 01:29:57 +0000 (17:29 -0800)]
Merge pull request #2934 from dotnet-bot/from-tfs
Merge changes from TFS
Jan Kotas [Sat, 30 Jan 2016 22:59:46 +0000 (14:59 -0800)]
Fix build break in internal build
[tfs-changeset: 1570568]
Jan Kotas [Sat, 30 Jan 2016 19:38:18 +0000 (11:38 -0800)]
Merge pull request #2931 from jkotas/glob-revert
Revert "Add un-prefixed signatures as temporary workaround"
Jan Kotas [Sat, 30 Jan 2016 16:12:16 +0000 (08:12 -0800)]
Merge pull request #2922 from rahku/updateBuildTools
update version of buildtools
Jan Kotas [Sat, 30 Jan 2016 16:10:25 +0000 (08:10 -0800)]
Revert "Add un-prefixed signatures as temporary workaround"
This reverts commit
6b1d2938ec4a5a2c64fd849797ec7800ed3ab575.
sivarv [Sat, 30 Jan 2016 02:14:31 +0000 (18:14 -0800)]
Fixes #2700
Sergiy Kuryata [Sat, 30 Jan 2016 00:57:21 +0000 (16:57 -0800)]
Merge pull request #2902 from wtgodbe/disableTests
Disable failing pri 1 tests that are unsupported outside of Windows
Bruce Forstall [Sat, 30 Jan 2016 00:53:36 +0000 (16:53 -0800)]
Merge pull request #2928 from BruceForstall/FixArm
Fix ARM JIT
Brian Sullivan [Sat, 30 Jan 2016 00:47:55 +0000 (16:47 -0800)]
Merge pull request #2923 from briansull/issue_2849
Fix for Issue 2849 - Assertion failed 'parent->OperGet() != GT_LDOBJ'
Bruce Forstall [Fri, 29 Jan 2016 23:57:48 +0000 (15:57 -0800)]
Fix ARM JIT
During a previous code cleanup, an assert condition was incorrectly inverted.
William Godbe [Fri, 29 Jan 2016 23:46:25 +0000 (15:46 -0800)]
Merge pull request #2924 from wtgodbe/pri1
Switch ubuntu PR tests from Release to Checked, and run Pri 1 tests on each commit
William Godbe [Fri, 29 Jan 2016 22:26:13 +0000 (14:26 -0800)]
Switch ubuntu PR tests from Release to Checked, and run Pri 1 tests on each commit
Jan Kotas [Fri, 29 Jan 2016 23:05:57 +0000 (15:05 -0800)]
Merge pull request #2917 from jkotas/globalization
Unique names for GlobalizationNative exports
Rahul Kumar [Fri, 29 Jan 2016 22:00:48 +0000 (14:00 -0800)]
fix for path to buildtools.coreclr package
Brian Sullivan [Fri, 29 Jan 2016 21:42:16 +0000 (13:42 -0800)]
Fix for Issue 2849 - Assertion failed 'parent->OperGet() != GT_LDOBJ'
Change the ifdef to use FEATURE_MULTIREG_STRUCT_ARGS instead of
FEATURE_UNIX_AMD64_STRUCT_PASSING because the ARM64 target
also uses GT_LDOBJ when passing 16-byte sized structs
Eugene Rozenfeld [Fri, 29 Jan 2016 21:34:12 +0000 (13:34 -0800)]
Merge pull request #2916 from erozenfeld/Checkedx86BuildFix
Fix for x86 checked build error.
Rahul Kumar [Fri, 29 Jan 2016 21:29:03 +0000 (13:29 -0800)]
update version of buildtools
Jan Kotas [Fri, 29 Jan 2016 20:15:57 +0000 (12:15 -0800)]
Add un-prefixed signatures as temporary workaround
Jan Kotas [Fri, 29 Jan 2016 18:25:41 +0000 (10:25 -0800)]
Unique names for GlobalizationNative exports
For consistency and to enable eventual sharing of the same code with CoreRT, I have changed the naming convention for System.Globalization.Native exports to match dotnet/corefx#4818.
Matt Ellis [Fri, 29 Jan 2016 21:31:26 +0000 (13:31 -0800)]
Merge pull request #2904 from ellismg/add-ubuntu-15.10-to-ci
Add Ubuntu 15.10 runs
Kyungwoo Lee [Fri, 29 Jan 2016 21:27:28 +0000 (13:27 -0800)]
Merge pull request #2919 from kyulee1/ver
RyuJit version update
William Godbe [Fri, 29 Jan 2016 21:01:27 +0000 (13:01 -0800)]
Merge pull request #2918 from wtgodbe/serverGC
This should enable Server GC for ubuntu PR CI builds
William Godbe [Fri, 29 Jan 2016 20:59:19 +0000 (12:59 -0800)]
This should enable Server GC for ubuntu PR CI builds
William Godbe [Thu, 28 Jan 2016 22:52:38 +0000 (14:52 -0800)]
Disable failing pri 1 tests that are unsupported outside of Windows
Eugene Rozenfeld [Fri, 29 Jan 2016 19:56:25 +0000 (11:56 -0800)]
Fix for x86 checked build error.
Remove FORCEINLINE from GetUnwindDataBlob so that the compiler
always emits it in codeman.obj. The code in nidump.cpp
(NativeImageDumper::DumpReadyToRunMethod) calls this method so
we were getting link errors if GetUnwindDataBlob wasn't emitted in codeman.obj.
GetUnwindDataBlob is small enough for a compiler to inline it even without FORCEINLINE.
Jan Kotas [Fri, 29 Jan 2016 18:20:17 +0000 (10:20 -0800)]
Merge pull request #2914 from Dmitry-Me/unifyRefcountAccess
Unify reference count access
Kyungwoo Lee [Fri, 29 Jan 2016 17:30:07 +0000 (09:30 -0800)]
RyuJit version update
John Chen [Fri, 29 Jan 2016 16:11:41 +0000 (08:11 -0800)]
Merge pull request #2888 from JohnChen0/coremsbuild
Build mscorlib with CoreCLR MSBuild on Linux and Mac
Dmitry-Me [Fri, 29 Jan 2016 14:21:31 +0000 (17:21 +0300)]
Unify reference count access
Bruce Forstall [Fri, 29 Jan 2016 05:05:45 +0000 (21:05 -0800)]
Merge pull request #2900 from BruceForstall/ClrStlUpdate
Fixes and additions to CLR STL functionality
Jan Kotas [Fri, 29 Jan 2016 04:08:57 +0000 (20:08 -0800)]
Merge pull request #2901 from JohnChen0/r2r-av
Modify crossgen to always set DLL bit in PE header
Jan Kotas [Fri, 29 Jan 2016 04:07:26 +0000 (20:07 -0800)]
Merge pull request #2913 from krytarowski/netbsd-support-15
NetBSD: Fix portability of isspace(3) usage
Jan Kotas [Fri, 29 Jan 2016 04:06:46 +0000 (20:06 -0800)]
Merge pull request #2911 from krytarowski/netbsd-support-13
NetBSD: Stop assuming that __hppa__ is HP-UX
Jan Kotas [Fri, 29 Jan 2016 04:04:19 +0000 (20:04 -0800)]
Merge pull request #2906 from krytarowski/netbsd-support-12
Stop setting for NetBSD -D_DEFAULT_SOURCE -D_POSIX_C_SOURCE=200809L
Jan Kotas [Fri, 29 Jan 2016 04:02:57 +0000 (20:02 -0800)]
Merge pull request #2912 from krytarowski/netbsd-support-14
NetBSD defines VM_MAXUSER_ADDRESS in <machine/vmparam.h>
Andy Ayers [Fri, 29 Jan 2016 01:59:42 +0000 (17:59 -0800)]
Merge pull request #2703 from AndyAyersMS/InlineRefactor
Initial round of jit inliner refactoring
Kamil Rytarowski [Fri, 29 Jan 2016 01:42:53 +0000 (02:42 +0100)]
NetBSD: Fix portability of isspace(3) usage
NAME
isspace - white-space character test
LIBRARY
Standard C Library (libc, -lc)
SYNOPSIS
#include <ctype.h>
int
isspace(int c);
STANDARDS
The isspace() function conforms to ANSI X3.159-1989 (``ANSI C89'').
CAVEATS
The argument to isspace() must be EOF or representable as an unsigned
char; otherwise, the behavior is undefined.
-- NetBSD man-page
This behavior is POSIX and Standard C:
The c argument is an int, the value of which the application shall ensure is a
character representable as an unsigned char or equal to the value of the macro
EOF. If the argument has any other value, the behavior is undefined.
-- The Open Group Base Specifications Issue 6
IEEE Std 1003.1, 2004 Edition
C11 standard:
The header declares several functions useful for classifying and mapping
characters In all cases the argument is an int, the value of which shall be
representable as an unsigned char or shall equal the value of the macro EOF. If
the argument has any other value, the behavior is undefined.
-- 7.4 Character handling <ctype.h> paragraph 1
Kamil Rytarowski [Fri, 29 Jan 2016 01:30:07 +0000 (02:30 +0100)]
NetBSD defines VM_MAXUSER_ADDRESS in <machine/vmparam.h>
/*
* USRSTACK is the top (end) of the user stack. Immediately above the
* user stack resides the user structure, which is UPAGES long and contains
* the kernel stack.
*
* Immediately after the user structure is the page table map, and then
* kernel address space.
*/
-- /usr/include/machine/vmparam.h
Kamil Rytarowski [Fri, 29 Jan 2016 01:23:21 +0000 (02:23 +0100)]
NetBSD: Stop assuming that __hppa__ is HP-UX
HPPA is one of the NetBSD ports.
Andy Ayers [Thu, 14 Jan 2016 23:33:01 +0000 (15:33 -0800)]
Initial round of jit inliner refactoring
In the jit, a particular inline decision is the aggregate result of checks
in many different parts of the code. The primary goal of this refactoring
is to establity a single object instance that will accumulate the necessary
information, and to ensure there are a small number of places in the code
where the final outcome is known.
That object is the `JitInlineResult`. I've updated it so it is no longer
copyable or assignable. Instead each root instance is passed down stack
by reference to helper methods that update the instance as the inline decision
making process evolves. Methods on the JitInlineResult can be used to advance
the state of what's known. The allowable state transitions are limited and the
code verifies that (when appropriate) a decision was actually made.
There are 4 such root instances:
* the main check done in the inlining pass.
* the candidate identification pass done in the importer.
* a pre-jit viability scan also done in the importer.
* another pre-jit viability scan done in the importer.
(These latter two identify un-inlinable methods at pre-jit time, so
subsequent jit invocations don't expend any cycles considering them).
I've updated various APIs to try and avoid introducing aliased copies or
projections of the inline result, for instance passing it via argument to
a helper that already has access to it via a member variable, or returning
a bool result from one of the various 'canInline' methods. More work is possible
here -- for example the compiler instance probably doesn't need an inline
result member of its own, since whenever it is inlining it has an inline
instance that has an inline result. I've added a few asserts in places to ensure
that if there is more than one plausible path to the result that they all refer
to the same object.
I've also started streamlining a bit of the goto-heavy case analysis in favor
of early returns, which I find more readable. The reporting obligation
is now held by the JitInlineResult destructor. Future work may go further in this
direction, and make the JitInlineResult responsible for never attribute flagging,
logging and dump messages, and perhaps even responsibility for undoing some of
the speculative state modification done when preparing for inlining (eg cleaning up
the locals in `fgMorphCallInlineHelper`).
At this stage the various candidate and failure reasons are still strings.
I added some new strings for in-progress checks so the dumps now show the
"success reason" for an inline. This is an area that will undergo
further refinement. Success reasons and failures and failure reasons
are still not captured in the inline tree; that too is forthcoming.
This change has minor codegen diffs. In cases where the callee has pinned locals
and returns a bool, we'll now identify the callee as a "never" candidate and so
some calls to it will never get the anticipatory wrapping that inline candidates
get. While this wrapping is undone later when inlining fails, it leaves a type
conversion behind and that conversion impacts the generated code slightly: I see
`test eax, eax` to query the return value rather than `test al, al`.
While I was striving for a no-diff change I think taking this one change is
reasonable: it's preferable to identify never candidates rather than repeatedly try
inlining them and fail each time.
Kamil Rytarowski [Thu, 28 Jan 2016 23:59:48 +0000 (00:59 +0100)]
Stop setting for NetBSD -D_DEFAULT_SOURCE -D_POSIX_C_SOURCE=200809L
This makes ucontext.h related discovery functions work.
Matt Ellis [Thu, 28 Jan 2016 23:35:58 +0000 (15:35 -0800)]
Add Ubuntu 15.10 runs
Part of dotnet/corefx#5349
Jan Kotas [Thu, 28 Jan 2016 22:51:35 +0000 (14:51 -0800)]
Merge pull request #2893 from swaroop-sridhar/redirect
Fix an assert failure in GCStress testing
Andy Ayers [Thu, 28 Jan 2016 22:49:42 +0000 (14:49 -0800)]
Merge pull request #2899 from dotnet-bot/from-tfs
Merge changes from TFS
Kyungwoo Lee [Thu, 28 Jan 2016 22:24:39 +0000 (14:24 -0800)]
Merge pull request #2868 from kyulee1/cfi
Support for CFI unwind info
Bruce Forstall [Thu, 28 Jan 2016 22:00:05 +0000 (14:00 -0800)]
Fixes and additions to CLR STL functionality
vector:
Added some const qualifiers.
Added cbegin(), cend(), and data() members.
algorithm:
Added transform()
string:
Implemented a simple subset of std::string. This was implemented for use by a component that will be checked in later. It is currently unused.
Removed ownership tags.
Added "magic" comments to help VIM editor properly handle these files without filename extensions.
John Chen (CLR) [Thu, 28 Jan 2016 21:26:46 +0000 (13:26 -0800)]
Modify crossgen to always set DLL bit in PE header
Andy Ayers [Thu, 28 Jan 2016 21:22:20 +0000 (13:22 -0800)]
The jit may try to inline in cases where there are serious mismatches between the type the caller is passing and the type the callee expects. This can cause the jit to issue malformed queries through the jit interface.
The fix is to take an existing assertion check which detects mismatches the jit is not prepared to handle and make the check active in retail builds. If a bad type mismatch is detected, inlining at that call site is aborted.
[tfs-changeset: 1569969]
Kyungwoo Lee [Sat, 23 Jan 2016 14:13:58 +0000 (06:13 -0800)]
Support for CFI unwind info
For Unix targeting CoreRT, this will provide platform specific unwind
info which is a dwarf format.
Unlike window’s UNWIND_INFO, we won’t encode the format within RyuJit
since it is not only complex/error-prone but also needs to be adjusted
depending on platforms.
Instead, CFI (call frame information) pseudo instructions are encoded,
which will be passed to CoreRT/ObjectWriter which will translate
them to directly emit CFI directives to LLVM MC stream to establish
frames and layout eh_frame section.
A jit flag is used to dynamically dispatch either Windows’s unwind
blob or this CFI instruction table. No JIT/EE interface change is needed
since the API already expects an opaque blob.
Initially when I looked at what Clang does, the prologue and the
sequence of CFI emissions are a bit different than Windows.
Since we will emit the same sequence of code with the same runtime that
we define, I assume this is unnecessary – I’ve verified the CFI sequence
here can work correctly with libunwind.
Basically we need only 3 operations – push reg is a combination of 1 and
2 below.
1. Allocation – increase stack frame. Normally subtract esp in x64.
2. Store – Copy a (callee save) register to stack (memory)
3. SaveFP – Set frame pointer register
Since Windows operation is based on the relative value (all offsets,
etc are computed from the current point), I also use the similar form of
CFI instructions.
So, mostly one window’s code corresponds to two CFIs – we might optimize
this by aggregating allocation, but I’d like to keep the current
syntax/semantic same as Windows.
This results in a very simple transformation on par with Windows.
Swaroop Sridhar [Thu, 28 Jan 2016 02:49:14 +0000 (18:49 -0800)]
Fix an assert failure in GCStress testing
When GCStress uses redirection, it pushes a
RedirectedThreadFrame (Windows) or
ResumableFrame(Unix) on the stack.
The stack walker, when checking for consistency of frame chain
makes a special case for this redirection when running under
GCStress -- but compares only against `RedirectedThreadFrame'.
This caused the StackWalker to think that certain invoke-s were
not redirected, resulting in the failures in some GCStress tests
on Linux.
This change fixes the problem.
Fixes #2848