[NFCI][bolt][test] Enable AT&T syntax explicitly

[tambry]

Files using the Intel syntax typically explicitly specify it. Do the same for the few relevant files for AT&T.

This enables building LLVM with -mllvm -x86-asm-syntax=intel in one's Clang config files (i.e. a global preference for Intel syntax).

Prior art: #164453 #166818

[llvmbot]

@llvm/pr-subscribers-bolt

Author: Raul Tambre (tambry)

Changes

Files using the Intel syntax typically explicitly specify it. Do the same for the few relevant files for AT&T.

This enables building LLVM with -mllvm -x86-asm-syntax=intel in one's Clang config files (i.e. a global preference for Intel syntax).

Prior art: #164453 #166818

---

Full diff: https://github.com/llvm/llvm-project/pull/167225.diff

4 Files Affected:

• (modified) bolt/test/runtime/X86/Inputs/basic-instrumentation.s (+1)
• (modified) bolt/test/runtime/X86/Inputs/internalcall.S (+1)
• (modified) bolt/test/runtime/X86/Inputs/pie-exceptions-failed-split.s (+1)
• (modified) bolt/test/runtime/X86/exceptions-lpstart-zero.s (+1)

diff --git a/bolt/test/runtime/X86/Inputs/basic-instrumentation.s b/bolt/test/runtime/X86/Inputs/basic-instrumentation.s
index 388d2cdbd085b..c28fa2210d256 100644
--- a/bolt/test/runtime/X86/Inputs/basic-instrumentation.s
+++ b/bolt/test/runtime/X86/Inputs/basic-instrumentation.s
@@ -1,3 +1,4 @@
+  .att_syntax
   .globl main
   .type main, %function
 main:
diff --git a/bolt/test/runtime/X86/Inputs/internalcall.S b/bolt/test/runtime/X86/Inputs/internalcall.S
index 775a67122d2d7..18f7914ef65ec 100644
--- a/bolt/test/runtime/X86/Inputs/internalcall.S
+++ b/bolt/test/runtime/X86/Inputs/internalcall.S
@@ -1,6 +1,7 @@
 // This is reduced test case for BOLT containing an internal call based on
 // GetCoreDump (from google core dumper).
 
+  .att_syntax
   .text
   .globl getCallback
   .type getCallback, %function
diff --git a/bolt/test/runtime/X86/Inputs/pie-exceptions-failed-split.s b/bolt/test/runtime/X86/Inputs/pie-exceptions-failed-split.s
index 5195d298b1bbe..fd12c26750b2c 100644
--- a/bolt/test/runtime/X86/Inputs/pie-exceptions-failed-split.s
+++ b/bolt/test/runtime/X86/Inputs/pie-exceptions-failed-split.s
@@ -49,6 +49,7 @@
 #			return !throw_test(argc, argv);
 #		}
 
+	.att_syntax
 	.text
 	.file	"exceptions-failed-split.cpp"
 	.globl	_Z3bari                         # -- Begin function _Z3bari
diff --git a/bolt/test/runtime/X86/exceptions-lpstart-zero.s b/bolt/test/runtime/X86/exceptions-lpstart-zero.s
index b487ff0fa2f59..1e377ea26cfd0 100644
--- a/bolt/test/runtime/X86/exceptions-lpstart-zero.s
+++ b/bolt/test/runtime/X86/exceptions-lpstart-zero.s
@@ -21,6 +21,7 @@
 # The exception table was modified with udata4 LPStartEncoding and sdata4
 # CallSiteEncoding.
 
+	.att_syntax
 	.text
 	.globl	main                            # -- Begin function main
 	.p2align	4, 0x90

[maksfb]

Thanks for adjusting BOLT tests. I'm surprised only a few needed a change since we use AT&T syntax everywhere.

I wonder if it will be easier to set proper flags in lit.local.cfg?

[tambry]

@maksfb A vast majority of the tests don't use the Clang frontend to compile assembly so they're not affected. Across the whole LLVM codebase only a few tens of changes have been necessary (~30'ish).

I've switched to passing -mllvm -x86-asm-syntax=att explicitly in the lit local config and removed the explicit annotations per your preference.
Worth noting that #167234 would add these to all LLVM MC generated outputs going forward for better interoperability, and not just only for Intel. Explicit annotations would track somewhat better with that and is what the fixes elsewhere went with. To me that also seems cleaner than passing a x86-specific flag on all architectures, ignoring some accidental breakages while the explicit syntax output change hopefully gets merged and propagates.

[github-actions]

🐧 Linux x64 Test Results

• 579 tests passed
• 36 tests skipped

[maksfb]

Thanks. Setting flags via config will prevent future failures for newly added tests.

[peterwaller-arm]

@tambry @maksfb @paschalis-mpeis - this appears to break tests unless the x86 backend is enabled as a part of LLVM_TARGETS_TO_BUILD. Can we entertain a revert or is there another way forward?

The argument is defined in an x86 target module:

llvm-project/llvm/lib/Target/X86/MCTargetDesc/X86MCAsmInfo.cpp

Lines 26 to 30 in 1c3b10f

	static cl::opt<AsmWriterFlavorTy> X86AsmSyntax(
	"x86-asm-syntax", cl::init(ATT), cl::Hidden,
	cl::desc("Select the assembly style for input"),
	cl::values(clEnumValN(ATT, "att", "Emit AT&T-style assembly"),
	clEnumValN(Intel, "intel", "Emit Intel-style assembly")));

(e.g. tests now fail if -DLLVM_TARGETS_TO_BUILD=AArch64 is supplied to cmake):

******************* TEST 'BOLT :: timers.c' FAILED ********************
...
# .---command stderr------------
# | clang (LLVM option parsing): Unknown command line argument '-x86-asm-syntax=att'.  Try: 'clang (LLVM option parsing) --help'
# | clang (LLVM option parsing): Did you mean '--asan-stack=att'?
# `-----------------------------
# error: command failed with exit status: 1

[maksfb]

Let me put up a fix.

[pawosm-arm]

FYI, enabling X86 targeting on AArch64 just leads to another test case failing:

# .---command stderr------------
# | build/bolt/test/X86/register-fragments-bolt-symbols.s:46:15: error: CHECK-SYMS: expected string not found in input
# | # CHECK-SYMS: l F .text [[#]] chain
# |               ^
# | <stdin>:8:62: note: scanning from here
# | 0000000000400080 l F .text.cold 0000000000000026 chain.cold.0
# |                                                              ^
# | <stdin>:9:18: note: possible intended match here
# | 0000000000400000 l F .text 000000000000003c chain
# |                  ^
# |
# | Input file: <stdin>
# | Check file: build/bolt/test/X86/register-fragments-bolt-symbols.s
# |
# | -dump-input=help explains the following input dump.
# |
# | Input was:
# | <<<<<<
# |             1:
# |             2: build/stage/product_build/tools/bolt/test/X86/Output/register-fragments-bolt-symbols.s.tmp.bolt: file format elf64-x86-64
# |             3:
# |             4: SYMBOL TABLE:
# |             5: 0000000000000000 l df *ABS* 0000000000000000 chain.s
# |             6: 0000000000001258 l F .bolt.org.text 0000000000000001 chain
# |             7: 0000000000001258 l d .bolt.org.text 0000000000000000 .bolt.org.text
# |             8: 0000000000400080 l F .text.cold 0000000000000026 chain.cold.0
# | check:46'0                                                                  X error: no match found
# |             9: 0000000000400000 l F .text 000000000000003c chain
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# | check:46'1                      ?                                 possible intended match
# |            10: 0000000000000000 l d .comment 0000000000000000 .comment
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# |            11: 0000000000000200 l d .interp 0000000000000000 .interp
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# |            12: 0000000000002310 l .dynamic 0000000000000000 .hidden _DYNAMIC
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# |            13: 0000000000000000 l df *ABS* 0000000000000000 bolt-pseudo.o
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# |            14: 00000000004000c0 l F .text.cold 0000000000000022 dummy.cold.0
# | check:46'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# |             .
# |             .
# |             .
# | >>>>>>
# `-----------------------------
# error: command failed with exit status: 1

[maksfb]

FYI, enabling X86 targeting on AArch64 just leads to another test case failing:

Thanks - I'll take a look. Can you try #169541 meanwhile?