[X86,AsmPrinter] Set assembler dialect for module inline asm

[MaskRay]

clang -c -masm=intel compiling a source file with file scope basic asm
incorrectly uses the AT&T dialect.

% cat a.c
asm("mov rax, rax");
% clang a.c -c -masm=intel
<inline asm>:1:1: error: unknown use of instruction mnemonic without a size suffix
mov rax, rax
^

Fix this by setting the assembler dialect from the MCAsmInfo object.

Note: clang -c -flto -masm=intel a.c still fails because of
https://reviews.llvm.org/D82862 for #34830: it tried to support AT&T
syntax for clang-cl, but the forced AT&T syntax is not compatible with
intended Intel syntax.

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Fangrui Song (MaskRay)

Changes

clang -c -masm=intel compiling a source file with file scope basic asm
incorrectly uses the AT&T dialect.

% cat a.c
asm("mov rax, rax");
% clang a.c -c -masm=intel
<inline asm>:1:1: error: unknown use of instruction mnemonic without a size suffix
mov rax, rax
^

Fix this by setting the assembler dialect from the MCAsmInfo object.

Note: clang -c -flto -masm=intel a.c still fails due to a known LTO
problem of ModuleSymbolTable::CollectAsmSymbols not using sufficient
target information.

---

Full diff: https://github.com/llvm/llvm-project/pull/85367.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp (+4-2)
• (added) llvm/test/CodeGen/X86/asm-dialect-module.ll (+10)

diff --git a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
index 0efe7a0e733672..a15538755d73b3 100644
--- a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -538,8 +538,10 @@ bool AsmPrinter::doInitialization(Module &M) {
   if (!M.getModuleInlineAsm().empty()) {
     OutStreamer->AddComment("Start of file scope inline assembly");
     OutStreamer->addBlankLine();
-    emitInlineAsm(M.getModuleInlineAsm() + "\n", *TM.getMCSubtargetInfo(),
-                  TM.Options.MCOptions);
+    emitInlineAsm(
+        M.getModuleInlineAsm() + "\n", *TM.getMCSubtargetInfo(),
+        TM.Options.MCOptions, nullptr,
+        InlineAsm::AsmDialect(TM.getMCAsmInfo()->getAssemblerDialect()));
     OutStreamer->AddComment("End of file scope inline assembly");
     OutStreamer->addBlankLine();
   }
diff --git a/llvm/test/CodeGen/X86/asm-dialect-module.ll b/llvm/test/CodeGen/X86/asm-dialect-module.ll
new file mode 100644
index 00000000000000..2c00a44424c2c3
--- /dev/null
+++ b/llvm/test/CodeGen/X86/asm-dialect-module.ll
@@ -0,0 +1,10 @@
+;; Test that we respect the assembler dialect when parsing module-level inline asm.
+; RUN: not llc < %s -mtriple=x86_64 2>&1 | FileCheck %s --check-prefix=ERR
+; RUN: llc < %s -mtriple=x86_64 -x86-asm-syntax=intel | FileCheck %s
+
+; ERR: <inline asm>:1:1: error: unknown use of instruction mnemonic without a size suffix
+
+; CHECK: .intel_syntax noprefix
+; CHECK: mov eax, eax
+
+module asm "mov eax, eax"

[KanRobert]

LGTM