[SR-9986] Test - IRGen/pic.swift failing on AArch64

[futurejones]

Previous ID	SR-9986
Radar	None
Original Reporter	@futurejones
Type	Bug

Environment

Linux Ubuntu 16.04 LTS on AArch64

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, Linux, Test, aarch64
Assignee	None
Priority	Medium

md5: 3f57543657dc35089583afc2202407fb

Issue Description:

Test - Swift(linux-aarch64) :: IRGen/pic.swift is failing on 4.2.2, 5.0 and master.

******************** TEST 'Swift(linux-aarch64) :: IRGen/pic.swift' FAILED ********************
Script:
--
: 'RUN: at line 4';   /home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/build/buildbot_linux/swift-linux-aarch64/bin/swift -frontend -target aarch64-unknown-linux-gnu  -module-cache-path '/home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/build/buildbot_linux/swift-linux-aarch64/swift-test-results/aarch64-unknown-linux-gnu/clang-module-cache' -swift-version 4  -assume-parsing-unqualified-ownership-sil /home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/swift/test/IRGen/pic.swift -module-name main -S -o - | /home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/swift/utils/PathSanitizingFileCheck --sanitize 'BUILD_DIR=/home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/build/buildbot_linux/swift-linux-aarch64' --sanitize 'SOURCE_DIR=/home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/swift' --use-filecheck '/home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/build/buildbot_linux/llvm-linux-aarch64/bin/FileCheck' -check-prefix=aarch64 /home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/swift/test/IRGen/pic.swift
--
Exit Code: 1

Command Output (stderr):
--
/home/worksonarm_test/jenkins_slave/workspace/swift-5.0-aarch64/swift/test/IRGen/pic.swift:56:13: error: aarch64: expected string not found in input
// aarch64: adrp [[REG1:x[0-9]+]], ($s4main6globalSivp@PAGE)
            ^
<stdin>:24:2: note: scanning from here
.cfi_startproc
^
<stdin>:31:2: note: possible intended match here
adrp x8, ($s4main6globalSivp)
^

--

********************

[futurejones]

A fix for this was submitted and merged but the test is still failing.

#19866

// aarch64-LABEL: $s4main10use_globalSiyF:
// aarch64:        adrp [[REG1:x[0-9]+]], ($s4main6globalSivp@PAGE)
// aarch64:        add [[REG1]], [[REG1]], :lo12:($s4main6globalSivp)
// aarch64:        bl swift_beginAccess
// aarch64:        ldr [[REG2:x[0-9]+]], {{\[}}[[REG1]]{{\]}}
// aarch64:        str [[REG2]], [sp]
// aarch64:        bl swift_endAccess
// aarch64:        ldr x0, [sp]

If you remove "@page" in line 2 the test passes

-// aarch64:        adrp [[REG1:x[0-9]+]], ($s4main6globalSivp@PAGE)
+// aarch64:        adrp [[REG1:x[0-9]+]], ($s4main6globalSivp)

But I don't know enough about this to know if this fix is correct.

[belkadan]

cc @rjmccall

[rjmccall]

I think that's just a target-specific assembler difference: Darwin's assembler calls out the fact that the operand isn't actually the full offset to the label, just the page offset, but armasm doesn't. So a fix to accept either should be fine.