Merge 2.073.2 front-end by kinke · Pull Request #2006 · ldc-developers/ldc

kinke · 2017-02-18T13:27:31Z

No description provided.

kinke · 2017-02-18T13:28:59Z

The compiler itself can be built on Win64, but compiling the libs fails for each (?) module due to 'recursive template expansion'.

kinke · 2017-02-18T20:39:41Z

Alright, compiler and libs build successfully now (incl. self-compile). The single remaining unittest compilation error is due to some sort of forward-referencing or other symbol resolution failure for std.regex. The first dmd-testsuite results are promising.
mars.d has changed quite a bit, primarily due to the new DMD -mcpu switch, with obviously incompatible values compared to LDC; a mapping to LLVM CPU and/or CPU features will probably be required for LDMD.

Ignore `-mcpu=baseline`, forward `-mcpu=native` to LDC and translate `-mcpu=avx` to `-mattr=+avx`. Show LLVM's (extensive) help for `-mcpu=?` (which is sadly actually printed 3 consecutive times).

Resolves issue #1962.

dnadlinger · 2017-02-19T00:53:05Z

driver/main.cpp


+  if (arch == llvm::Triple::x86 || arch == llvm::Triple::x86_64) {
+    if (traitsTargetHasFeature("sse2")) {
+      VersionCondition::addPredefinedGlobalIdent("D_SIMD");


D_SIMD refers to the DMD-style __simd opcode interface. (The docs are vague on this. Edit again: https://dlang.org/spec/version.html is clearer.)

Ah okay, and D_AVX isn't documented at all, but I guess it's just an extension for additional opcodes.

LDC doesn't support DMD's core.simd interface.

dnadlinger · 2017-02-19T02:06:16Z

https://dlang.org/changelog/2.073.0.html says "Add D_AVX predefined version when -mcpu=avx is used", so D_AVX might have been correct.

(Misunderstood UI.)

kinke · 2017-02-19T02:31:35Z

This concludes propagating the mars.d changes to LDC/LDMD. On Win64, there are 5 remaining failures wrt. dmd-testsuite-debug alone, 3 of which seem invalid.

kinke · 2017-02-19T16:02:20Z

Update:

unittests: std.regex.internal.shiftor fails to compile its unittests ('template not defined'), all others compile & work (at least on Win64).
lit-tests: work.
dmd-testsuite: works, except for:
- runnable/b16278.d: missing symbol on Win64/OSX (debug only), links on Linux (?)
- runnable/test15862.d: crash ('illegal instruction')

kinke · 2017-02-22T23:12:30Z

runnable/test15862.d shows that LDC still lacks propagating immutable constants to LL constants. The 4 failing asserts check that repeated calls to pure functions returning indirections to immutable stuff return pointers to the same LL constant.

dnadlinger · 2017-02-22T23:23:15Z

runnable/test15862.d … immutable constants to LL constants

DMD passes this not due to emitting stuff as constant, but because it does CSE on strongly pure functions. Actually, the assembly for e.g. the on() part is … rather funny:

   100000db4:   e8 b7 33 02 00          call   100024170          // allocate object
   100000db9:   48 89 45 f8             mov    QWORD PTR [rbp-0x8],rax
   100000dbd:   48 8b 4d f8             mov    rcx,QWORD PTR [rbp-0x8]
   100000dc1:   48 3b c8                cmp    rcx,rax

In other words, it copies the pointer to the allocated object to the stack, loads it into another register and then compares the two. Solid.

dnadlinger · 2017-02-22T23:26:12Z

(Just disabling the test until we implement optimizations on strongly pure functions – if ever – would be fine.)

kinke · 2017-02-22T23:37:42Z

So DMD caches the first result for all later detected identical invokations of strongly pure functions inside a function? Apparently only when optimizing. Shouldn't/couldn't this be done in the front-end already? At least optionally, in case the compile time/memory footprint impact could become noticeable.

Conflicts: runtime/phobos

dnadlinger · 2017-03-04T23:08:14Z

Shouldn't/couldn't this be done in the front-end already?

It could certainly be done as a pass over the AST, yes (whether you call that front-end or not).

Instead of letting our old copy silently go more and more out-of-sync. In this case, we missed the upstream fix wrt. skipping the declaration of ClassInfos for speculative class types. There are 2 noteworthy functional changes for function `getOrCreateTypeInfoDeclaration()`: 1) The old version always overwrote `torig->vtinfo` with `t->vtinfo` when declaring a new TypeInfo, whereas upstream's `genTypeInfo()` only sets it if it was null before. 2) The old version called `semanticTypeInfo()` during a semantic pass, upstream doesn't. The LDC-specific exception for class types in `builtinTypeInfo()` is still required. Fixes dmd-testsuite's runnable/b16278.d.

kinke · 2017-03-05T14:45:29Z

Apparently only remaining issue: std.regex.internal.shiftor unittest compile error.

kinke · 2017-03-05T15:27:44Z

Wow, that std.regex thing seems really ugly.

alias std.regex.internal.kickstart.Kickstart = std.regex.internal.kickstart.ShiftOr not being available in module std.regex.internal.shiftor may be okay (?), as it's only imported locally in a struct in imported std.regex.internal.ir.

When trying to work around this via explicit import:

diff --git a/std/regex/internal/shiftor.d b/std/regex/internal/shiftor.d
index 0a549f4..f3aca3a 100644
--- a/std/regex/internal/shiftor.d
+++ b/std/regex/internal/shiftor.d
@@ -6,7 +6,7 @@ module std.regex.internal.shiftor;

 package(std.regex):

-import std.regex.internal.ir;
+import std.regex.internal.ir, std.regex.internal.kickstart : Kickstart;
 import std.range.primitives, std.utf;

 //utility for shiftOr, returns a minimum number of bytes to test in a Char

it then fails with

C:\LDC\ldc\runtime\phobos\std\regex\internal\shiftor.d(29): Error: class std.regex.internal.shiftor.ShiftOr!char.ShiftOr base type must be class or interface, not Kickstart!char

which makes perfect sense, as the (very similar) base std.regex.internal.kickstart.ShiftOr!char is a struct, not a class!!

Pinging @JackStouffer due to dlang/phobos@5a2491a. Apparently there was a std.regex.internal.ir.Kickstart interface before the revert.

kinke · 2017-03-05T16:12:01Z

Upstream PR: dlang/phobos#5244

Awaiting some green now. :)

kinke · 2017-03-05T17:36:35Z

One more semantic/target_traits.d lit test failure (LDC crash) on Win32 (possibly x86 in general; the x86 AppVeyor job is the only one running the lit-tests in 32-bit, there's no multilib config for it yet) and accordingly no results for the Win32 unittests yet; green on all other CI systems (minus spurious OSX core.thread failure).

A ctor (I think I added it for convenience when merging 2.073) means that it's no POD type for C++ anymore. Non-POD structs are expected to be returned via sret etc. Fixes lit-test semantic/target_traits.d for x86 hosts.

Conflicts: tests/d2/dmd-testsuite

kinke · 2017-03-11T11:22:11Z

Upgraded to 2.073.2 (very few fixes only). The first of the 2 front-end fixes originated from this PR (06e88fe), the other one was already cherry-picked into master (b159de0).

dushibaiyu · 2017-03-18T08:48:56Z

will it marge 073.2 into ldc 1.2.0 release?

dnadlinger · 2017-03-18T09:10:54Z

@dushibaiyu: No, 2.073.2 will go into 1.3.x.

dnadlinger · 2017-03-18T09:11:26Z

@kinke: I can finally reproduce the spurious OS X core.thread issue on my local machine!

druntime-test-runner(43837,0x7fffb68223c0) malloc: *** error for object 0x1: pointer being freed was not allocated
*** set a breakpoint in malloc_error_break to debug
Abort trap: 6

Investigating…

dnadlinger · 2017-03-18T09:19:45Z

Ah, the issue seems to occur during garbage collection of stale Thread objects:

* thread #1: tid = 0x0000, 0x00007fffadb03dd6 libsystem_kernel.dylib`__pthread_kill + 10, stop reason = signal SIGSTOP
  * frame #0: 0x00007fffadb03dd6 libsystem_kernel.dylib`__pthread_kill + 10
    frame #1: 0x00007fffadbef787 libsystem_pthread.dylib`pthread_kill + 90
    frame #2: 0x00007fffada69420 libsystem_c.dylib`abort + 129
    frame #3: 0x00007fffadb5903f libsystem_malloc.dylib`free + 530
    frame #4: 0x000000010f87c537 druntime-test-runner`_D2rt4util9container6common8xreallocFNbNiPvmZPv + 71
    frame #5: 0x000000010f874869 druntime-test-runner`_D2rt19sections_elf_shared13finiTLSRangesFPS2rt4util9container5array13__T5ArrayTAvZ5ArrayZv + 73
    frame #6: 0x000000010f875d0c druntime-test-runner`_D2rt5tlsgc7destroyFPvZv + 12
    frame #7: 0x000000010f7dfccd druntime-test-runner`_D4core6thread6Thread6__dtorMFZv + 45
    frame #8: 0x000000010f86800e druntime-test-runner`rt_finalize2 + 94
    frame #9: 0x000000010f81a56a druntime-test-runner`_D2gc4impl12conservative2gc3Gcx5sweepMFNbZm + 314
    frame #10: 0x000000010f817165 druntime-test-runner`_D2gc4impl12conservative2gc3Gcx11fullcollectMFNbbZm + 533
    frame #11: 0x000000010f81734d druntime-test-runner`_D2gc4impl12conservative2gc14ConservativeGC150__T9runLockedS100_D2gc4impl12conservative2gc14ConservativeGC11fullCollectMFNbZ2goFNbPS2gc4impl12conservative2gc3GcxZmTPS2gc4impl12conservative2gc3GcxZ9runLockedMFNbKPS2gc4impl12conservative2gc3GcxZm + 45
    frame #12: 0x000000010f7e77da druntime-test-runner`_D4core6thread18__unittestL5742_34FZv + 122
    frame #13: 0x000000010f7ea07a druntime-test-runner`_D4core6thread10__unittestZ + 410
    frame #14: 0x000000010f7ba2c0 druntime-test-runner`_D11test_runner6doTestFPS6object10ModuleInfoKbZv + 64
    frame #15: 0x000000010f7ba1d0 druntime-test-runner`_D11test_runner11testModulesFZb + 176
    frame #16: 0x000000010f7d9d75 druntime-test-runner`runModuleUnitTests + 101
    frame #17: 0x000000010f863e46 druntime-test-runner`_d_run_main + 470
    frame #18: 0x00007fffad9d5255 libdyld.dylib`start + 1
    frame #19: 0x00007fffad9d5255 libdyld.dylib`start + 1

Seems like it might be related to dlang/druntime#1655, which applied to the Shared path.

dushibaiyu · 2017-03-18T09:42:17Z

@klickverbot Thanks.

dnadlinger · 2017-03-18T09:48:39Z

Seems like it might be related to dlang/druntime#1655

Yep, this is indeed the same issue. We should probably just move _tlsRanges to be on the C heap, with only the pointer being stored in TLS.

dnadlinger · 2017-03-18T10:32:56Z

@kinke: I pushed a patch which should fix the OS X core.thread crashes (if they were indeed the same as I got locally).

Happy to merge the branch to master after green?

kinke · 2017-03-18T10:47:49Z

Great that the OSX issue seems to be understood now (although it'll still happen for the shared libs as I understand).
We'll need it for LDC 1.2 too though, right?
Oh and merging is fine by me (and consequent cherry-picking only for branch release-1.2.[0x]). We'll be able to merge a bunch of other PRs deferred to 1.3 then too (#2016, #1960, #2030).

dnadlinger · 2017-03-18T11:43:18Z

although it'll still happen for the shared libs as I understand

This specific issue should have been fixed already for shared libraries.

kinke · 2017-03-18T12:04:24Z

I see, since 2.072. I don't remember whether it was only the static-libs job spuriously failing since 2.072, but we'll find out.

kinke · 2017-03-18T13:13:39Z

Looking good. I'll cherry-pick the fix onto master before merging 2.073.2 so that 1.2 doesn't (yet) need its own ldc-release-1.2.x druntime branch.

dnadlinger · 2017-03-18T13:31:04Z

Okay I'll let you do the merge, then.

Conflicts: runtime/druntime

Merge 2.073.1 front-end

0f620f4

kinke added 3 commits February 18, 2017 18:36

Add missing field in C++ global.params struct

06e88fe

druntime: Add yl2x() and yl2xp1() to core.math

4ae2149

Phobos: Fix std.numeric compile error

951ecdd

kinke force-pushed the merge-2.073 branch from b8fbdac to 951ecdd Compare February 18, 2017 19:51

Bump version number to 1.3 and front-end version to 2.073.1

2ab66d3

kinke added 5 commits February 18, 2017 22:42

LDMD: Warn about unsupported new -mscrtlib switch

b53f4cd

Add new switch -dip1000

8648b93

Support -mcpu switch in LDMD

f34b8ab

Ignore `-mcpu=baseline`, forward `-mcpu=native` to LDC and translate `-mcpu=avx` to `-mattr=+avx`. Show LLVM's (extensive) help for `-mcpu=?` (which is sadly actually printed 3 consecutive times).

Add DMD-compatible pre-defined versions D_SIMD and D_AVX

627b35d

Use upstream's new printPredefinedVersions()

3940f6f

Resolves issue #1962.

dnadlinger requested changes Feb 19, 2017

View reviewed changes

kinke added 2 commits February 19, 2017 02:37

Don't pre-define versions D_SIMD and D_AVX

1d4c6fa

LDC doesn't support DMD's core.simd interface.

Add new switch -mv=<package.module>=<filespec>

0b03cda

dnadlinger previously approved these changes Feb 19, 2017

View reviewed changes

kinke added 2 commits February 19, 2017 14:28

druntime: Fix inlining of core.bitop.bsf() and bsr()

45479bb

dmd-testsuite: Disable/adapt some new tests for LDC

f3d9f7f

kinke added 2 commits March 4, 2017 23:06

dmd-testsuite: Partially disable a test due to missing CSE

fd79016

Merge branch 'master' into merge-2.073

162c264

Conflicts: runtime/phobos

Phobos: Fix std.regex.internal.shiftor

2ce0d10

Fix gen/ldctraits D/C++ interop ABI regression

8db147a

A ctor (I think I added it for convenience when merging 2.073) means that it's no POD type for C++ anymore. Non-POD structs are expected to be returned via sret etc. Fixes lit-test semantic/target_traits.d for x86 hosts.

kinke force-pushed the merge-2.073 branch from fe27f4c to 8db147a Compare March 5, 2017 18:44

kinke added 2 commits March 9, 2017 01:11

Merge branch 'master' into merge-2.073

97188a1

Conflicts: tests/d2/dmd-testsuite

Merge 2.073.2 front-end

88f69e9

kinke changed the title ~~Merge 2.073.1 front-end~~ Merge 2.073.2 front-end Mar 11, 2017

dnadlinger mentioned this pull request Mar 18, 2017

sections_elf_shared: Do not access TLS of dead thread in finiTLSRanges() dlang/druntime#1655

Merged

Merge branch 'master' into merge-2.073

2aebbe4

Conflicts: runtime/druntime

kinke force-pushed the merge-2.073 branch from 70560c4 to 2aebbe4 Compare March 18, 2017 13:34

Merge branch 'master' into merge-2.073

2ffba59

Conflicts: runtime/druntime

kinke merged commit 5c5831a into master Mar 18, 2017

kinke deleted the merge-2.073 branch March 18, 2017 22:02

Uh oh!

Conversation

kinke commented Feb 18, 2017

Uh oh!

kinke commented Feb 18, 2017

Uh oh!

kinke commented Feb 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dnadlinger Feb 19, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kinke Feb 19, 2017

Choose a reason for hiding this comment

Uh oh!

dnadlinger commented Feb 19, 2017

Uh oh!

kinke commented Feb 19, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Feb 19, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Feb 22, 2017

Uh oh!

dnadlinger commented Feb 22, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dnadlinger commented Feb 22, 2017

Uh oh!

kinke commented Feb 22, 2017

Uh oh!

dnadlinger commented Mar 4, 2017

Uh oh!

kinke commented Mar 5, 2017

Uh oh!

kinke commented Mar 5, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Mar 5, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Mar 5, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Mar 11, 2017

Uh oh!

dushibaiyu commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

dushibaiyu commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kinke commented Mar 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

kinke commented Mar 18, 2017

Uh oh!

kinke commented Mar 18, 2017

Uh oh!

dnadlinger commented Mar 18, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

kinke commented Feb 18, 2017 •

edited

Loading

dnadlinger Feb 19, 2017 •

edited

Loading

kinke commented Feb 19, 2017 •

edited

Loading

kinke commented Feb 19, 2017 •

edited

Loading

dnadlinger commented Feb 22, 2017 •

edited

Loading

kinke commented Mar 5, 2017 •

edited

Loading

kinke commented Mar 5, 2017 •

edited

Loading

kinke commented Mar 5, 2017 •

edited

Loading

dnadlinger commented Mar 18, 2017 •

edited

Loading

kinke commented Mar 18, 2017 •

edited

Loading