Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes #901

hpux735 · 2016-01-07T00:35:15Z

this pull request adds support for ARMv6 (including the original Raspberry pi), and fixes several bugs encountered on ARMv7. I've been using this code for a few weeks, and is has been rather stable. My hope is that AMRv6 and ARMv7 on linux can be included in the Swift 2.2 release.

jrose-apple · 2016-01-07T00:54:45Z

I'm not really the right person to review (just trying to get past the easy stuff), but what tests currently are and aren't passing?

hpux735 · 2016-01-07T23:09:12Z

ARMv7 test results:

Testing Time: 754.18s

Failing Tests (28):
Swift :: 1_stdlib/ArrayCore.swift
Swift :: 1_stdlib/Interval.swift
Swift :: 1_stdlib/Optional.swift
Swift :: 1_stdlib/Range.swift
Swift :: 1_stdlib/Reflection.swift
Swift :: 1_stdlib/VarArgs.swift
Swift :: ClangModules/autolinking.swift
Swift :: DebugInfo/ASTSection_ObjC.swift
Swift :: DebugInfo/variables.swift
Swift :: Driver/Dependencies/one-way-external.swift
Swift :: Driver/subcommands.swift
Swift :: IRGen/c_layout.sil
Swift :: Interpreter/archetype_casts.swift
Swift :: Interpreter/formal_access.swift
Swift :: Interpreter/generic_class.swift
Swift :: Interpreter/generic_struct.swift
Swift :: Interpreter/protocol_extensions.swift
Swift :: Interpreter/protocol_lookup.swift
Swift :: Interpreter/slices.swift
Swift :: Interpreter/testability.swift
Swift :: Misc/expression_too_complex.swift
Swift :: Parse/BOM.swift
Swift :: Prototypes/CollectionsMoveIndices.swift
Swift :: Prototypes/FloatingPoint.swift
Swift :: Prototypes/MutableIndexableDict.swift
Swift :: Prototypes/TextFormatting.swift
Swift :: Serialization/autolinking.swift
Swift-Unit :: runtime/SwiftRuntimeTests/MetadataTest.getExistentialMetadata

Expected Passes : 1611
Expected Failures : 78
Unsupported Tests : 639
Unexpected Failures: 28

hpux735 · 2016-01-08T00:49:53Z

ARMv6 results:

Testing Time: 7199.76s

Failing Tests (30):
Swift :: 1_stdlib/ArrayCore.swift
Swift :: 1_stdlib/FloatingPointIR.swift
Swift :: 1_stdlib/Interval.swift
Swift :: 1_stdlib/Optional.swift
Swift :: 1_stdlib/Range.swift
Swift :: 1_stdlib/Reflection.swift
Swift :: 1_stdlib/VarArgs.swift
Swift :: ClangModules/autolinking.swift
Swift :: DebugInfo/ASTSection_ObjC.swift
Swift :: DebugInfo/variables.swift
Swift :: Driver/Dependencies/one-way-external.swift
Swift :: Driver/subcommands.swift
Swift :: IRGen/c_layout.sil
Swift :: IRGen/multithread_module.swift
Swift :: Interpreter/archetype_casts.swift
Swift :: Interpreter/formal_access.swift
Swift :: Interpreter/generic_class.swift
Swift :: Interpreter/generic_struct.swift
Swift :: Interpreter/protocol_extensions.swift
Swift :: Interpreter/protocol_lookup.swift
Swift :: Interpreter/slices.swift
Swift :: Interpreter/testability.swift
Swift :: Misc/expression_too_complex.swift
Swift :: Parse/BOM.swift
Swift :: Prototypes/CollectionsMoveIndices.swift
Swift :: Prototypes/FloatingPoint.swift
Swift :: Prototypes/MutableIndexableDict.swift
Swift :: Prototypes/TextFormatting.swift
Swift :: Serialization/autolinking.swift
Swift-Unit :: runtime/SwiftRuntimeTests/MetadataTest.getExistentialMetadata

Expected Passes : 1609
Expected Failures : 78
Unsupported Tests : 639
Unexpected Failures: 30

gribozavr · 2016-01-08T13:00:27Z

cmake/modules/AddSwift.cmake

@@ -159,7 +159,7 @@ function(_add_variant_link_flags
      result)

  if("${sdk}" STREQUAL "LINUX")
-    list(APPEND result "-lpthread" "-ldl")
+    list(APPEND result "-lpthread" "-ldl" "-Wl,-Bsymbolic")


-Bsymbolic seems to be a big hammer (and it might be just a workaround for what is an LLVM or a linker bug). Could you scope it as narrowly as possible, just to armv6?

Yes, it is. After 2.2 is forked, I'm going to work with @tienex on a different (probably better) solution based upon the gold linker. I'll restrict this to arm-linux.

gribozavr · 2016-01-08T17:51:28Z

@hpux735 The general approach is to run %target-swiftc_driver -### and add appropriate options that you would expect to trigger this behavior. The driver in -### will print the command line, but not execute it. I think in this case the changes you made will affect the linker options, so the tests should go into test/Driver/linker.swift -- or you might just need to add a few extra CHECK lines.

hpux735 · 2016-01-09T00:29:51Z

test/Driver/linker.swift

+// LINUX-armv6-DAG: [[OBJECTFILE]]
+// LINUX-armv6-DAG: -lswiftCore
+// LINUX-armv6-DAG: -L [[STDLIB_PATH:[^ ]+/lib/swift]]
+// LINUX-armv6-DAG: --target=junk-unknown-linux-gnueabihf


While attempting to add testing for whether --target is added to the clang++ invocation, this seems like the correct approach. However, this does not cause a failure as expected.

When lit is running a test on a file with more than one test (such as linker.swift), it looks like it bails when it finds the first failure, right? I deleted the tests relating to macos and the relative linker and verified that the tests for providing --target to clang functions as intended.

hpux735 · 2016-01-23T21:26:24Z

I'm almost done with a patch that will do away with the swift.ld linker script, but in the spirit of small changes, can we get this moved forward so I can open up a new PR? In your opinion (@gribozavr )? what are the remaining issues that need to be addressed?

Thanks!!

gribozavr · 2016-01-25T01:49:23Z

Sure! Could you rebase the patch? (there are conflicts)

It also seems that @jckarter added -Wl,-Bsymbolic to the build anyway.

modocache · 2016-01-25T03:00:09Z

lib/Driver/Driver.cpp

+        triple.getSubArch() == llvm::Triple::SubArchType::ARMSubArch_v6) {
+      triple.setArchName("armv6");
+    }
+  }


nit-pick: Perhaps combine some of these conditionals, and use else if? For example:

if (triple.getOS() == llvm::Triple::Linux && triple.getArch() == llvm::Triple::arm) { if (triple.getSubArch() == llvm::Triple::SubArchType::ARMSubArch_v7) { triple.setArchName("armv7"); } else if (triple.getSubArch() == llvm::Triple::SubArchType::ARMSubArch_v6) { triple.setArchName("armv6"); } }

Personally I find this more readable, but that's just my two cents. 😄

Totally agree.

hpux735 · 2016-01-27T00:11:37Z

I has conflicts with apple/master and resulting regressions. I'm working on resolving them at the moment.

hpux735 · 2016-01-29T13:31:06Z

I believe that all the regression issues have been resolved. I think this is ready to go, finally.

modocache · 2016-01-29T14:14:04Z

@hpux735 Fantastic! Are all tests passing? If not, could you post a list of which aren't? I'd like to help if possible.

Also, I have a feeling someone on the core team will ask you to squash these commits--or at least rebase onto master, to get rid of intermediate commits like these:

But I'm not sure, so don't take my word for it. 😛

hpux735 · 2016-01-29T18:00:54Z

test/Driver/linker.swift

+// LINUX-armv6-DAG: [[OBJECTFILE]]
+// LINUX-armv6-DAG: -lswiftCore
+// LINUX-armv6-DAG: -L [[STDLIB_PATH:[^ ]+/lib/swift]]
+// LINUX-armv6-DAG: --target=armv6-unknown-linux-gnueabihf


Here's the test you asked for, @gribozavr

gribozavr · 2016-01-29T18:47:32Z

@swift-ci Please test

gribozavr · 2016-01-29T21:11:23Z

@hpux735 Sorry, this failed on OS X:

TEST 'Swift :: Driver/linker.swift' FAILED
swift/test/Driver/linker.swift:135:17: error: expected string not found in input
// LINUX-armv6: swift
                ^
<stdin>:4:1: note: scanning from here

Do you have access to an OS X machine to investigate?

hpux735 · 2016-01-29T21:43:49Z

Sorry about that! I just fixed an issue and ran it on my OSX machine. I think it will pass this time, it did for me.

gribozavr · 2016-01-29T21:47:10Z

@swift-ci Please test

hpux735 · 2016-01-29T23:31:47Z

Is there a list of machines/platforms used for CI and the arguments used? I definitely don't want to waste people's time with these problems, if it can be avoided. I just tried it again on my mac and the tests completed without error.

gribozavr · 2016-01-30T01:46:04Z

@hpux735 It failed because of infrastructure reasons. Linux passed, we don't know about OS X. We'll get it fixed and I will re-run.

gribozavr · 2016-01-30T01:52:21Z

@swift-ci Please test

gribozavr · 2016-01-30T04:07:15Z

It failed with:

Failing Tests (1):
    Swift :: IDE/crashers_fixed/019-swift-vardecl-emitlettovarnoteifsimple.swift

which I fixed.

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes

gribozavr · 2016-01-30T04:08:22Z

@hpux735 Thank you for working on the Linux arm port!

hpux735 · 2016-01-30T04:17:03Z

Awesome!!! Thanks so much Dmitri!

MacMeDan · 2016-02-01T16:31:06Z

@hpux735 you are my hero.

hpux735 · 2016-02-01T18:25:57Z

LOL, Glad I could help, @peterdan

The target was added for Unix toolchains in swiftlang#901, but a later pull swiftlang#1891 added it again. Since clang only uses the last target flag that's passed in, all customization done for the first one was unused these last 4+ years, so remove it and change tests that look for custom strings passed by the first one.

gribozavr reviewed Jan 8, 2016
View reviewed changes

hpux735 reviewed Jan 9, 2016
View reviewed changes

modocache reviewed Jan 25, 2016
View reviewed changes

hpux735 reviewed Jan 29, 2016
View reviewed changes

Implemented ARMv6 and fixed up ARMv7

ab7c87e

gribozavr added a commit that referenced this pull request Jan 30, 2016

Merge pull request #901 from hpux735/master

b51bc1c

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes

gribozavr merged commit b51bc1c into swiftlang:master Jan 30, 2016

finagolfin mentioned this pull request Dec 5, 2020

[Driver] Remove redundant clang '-target' flag that wasn't being used #34978

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes #901

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes #901

hpux735 commented Jan 7, 2016

jrose-apple commented Jan 7, 2016

hpux735 commented Jan 7, 2016

hpux735 commented Jan 8, 2016

gribozavr Jan 8, 2016

hpux735 Jan 8, 2016

gribozavr commented Jan 8, 2016

hpux735 Jan 9, 2016

hpux735 Jan 9, 2016

hpux735 commented Jan 23, 2016

gribozavr commented Jan 25, 2016

modocache Jan 25, 2016

hpux735 Jan 25, 2016

hpux735 commented Jan 27, 2016

hpux735 commented Jan 29, 2016

modocache commented Jan 29, 2016

hpux735 Jan 29, 2016

gribozavr commented Jan 29, 2016

gribozavr commented Jan 29, 2016

hpux735 commented Jan 29, 2016

gribozavr commented Jan 29, 2016

hpux735 commented Jan 29, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

hpux735 commented Jan 30, 2016

MacMeDan commented Feb 1, 2016

hpux735 commented Feb 1, 2016

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes #901

Support ARMv6 (original Raspberry Pi), ARMv7 bugfixes #901

Conversation

hpux735 commented Jan 7, 2016

jrose-apple commented Jan 7, 2016

hpux735 commented Jan 7, 2016

hpux735 commented Jan 8, 2016

gribozavr Jan 8, 2016

Choose a reason for hiding this comment

hpux735 Jan 8, 2016

Choose a reason for hiding this comment

gribozavr commented Jan 8, 2016

hpux735 Jan 9, 2016

Choose a reason for hiding this comment

hpux735 Jan 9, 2016

Choose a reason for hiding this comment

hpux735 commented Jan 23, 2016

gribozavr commented Jan 25, 2016

modocache Jan 25, 2016

Choose a reason for hiding this comment

hpux735 Jan 25, 2016

Choose a reason for hiding this comment

hpux735 commented Jan 27, 2016

hpux735 commented Jan 29, 2016

modocache commented Jan 29, 2016

hpux735 Jan 29, 2016

Choose a reason for hiding this comment

gribozavr commented Jan 29, 2016

gribozavr commented Jan 29, 2016

hpux735 commented Jan 29, 2016

gribozavr commented Jan 29, 2016

hpux735 commented Jan 29, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

gribozavr commented Jan 30, 2016

hpux735 commented Jan 30, 2016

MacMeDan commented Feb 1, 2016

hpux735 commented Feb 1, 2016