[X86] Always use 64-bit relocations in no-PIC large code model

[aeubanks]

This matches other types of relocations, e.g. to constant pool. And makes things more consistent with PIC large code model.

Some users of the large code model may not place small data in the lower 2GB of the address space (e.g. ClangBuiltLinux/linux#2016), so just unconditionally use 64-bit relocations in the large code model.

So now functions in a section not marked large will use 64-bit relocations to reference everything when using the large code model.

This also fixes some lldb tests broken by #88172 (https://lab.llvm.org/buildbot/#/builders/68/builds/72458).

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Arthur Eubanks (aeubanks)

Changes

This matches other types of relocations, e.g. to constant pool.

Some users of the large code model may not place small data in the lower 2GB of the address space (e.g. ClangBuiltLinux/linux#2016), so just unconditionally use 64-bit relocations in the large code model.

---

Full diff: https://github.com/llvm/llvm-project/pull/89101.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelDAGToDAG.cpp (+3-4)
• (modified) llvm/test/CodeGen/X86/code-model-elf.ll (+2-2)

diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 4e4241efd63d6b..7dcde2a508949d 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2927,11 +2927,10 @@ bool X86DAGToDAGISel::selectAddr(SDNode *Parent, SDValue N, SDValue &Base,
 }
 
 bool X86DAGToDAGISel::selectMOV64Imm32(SDValue N, SDValue &Imm) {
-  // Cannot use 32 bit constants to reference objects in kernel code model.
-  // Cannot use 32 bit constants to reference objects in large PIC mode since
-  // GOTOFF is 64 bits.
+  // Cannot use 32 bit constants to reference objects in kernel/large code
+  // model.
   if (TM.getCodeModel() == CodeModel::Kernel ||
-      (TM.getCodeModel() == CodeModel::Large && TM.isPositionIndependent()))
+      TM.getCodeModel() == CodeModel::Large)
     return false;
 
   // In static codegen with small code model, we can get the address of a label
diff --git a/llvm/test/CodeGen/X86/code-model-elf.ll b/llvm/test/CodeGen/X86/code-model-elf.ll
index 0da62e3e7a6519..f60f75bc26911e 100644
--- a/llvm/test/CodeGen/X86/code-model-elf.ll
+++ b/llvm/test/CodeGen/X86/code-model-elf.ll
@@ -350,7 +350,7 @@ define dso_local ptr @lea_forced_small_data() #0 {
 ;
 ; LARGE-STATIC-LABEL: lea_forced_small_data:
 ; LARGE-STATIC:       # %bb.0:
-; LARGE-STATIC-NEXT:    movl $forced_small_data, %eax
+; LARGE-STATIC-NEXT:    movabsq $forced_small_data, %rax
 ; LARGE-STATIC-NEXT:    retq
 ;
 ; SMALL-PIC-LABEL: lea_forced_small_data:
@@ -403,7 +403,7 @@ define dso_local i32 @load_forced_small_data() #0 {
 ;
 ; LARGE-STATIC-LABEL: load_forced_small_data:
 ; LARGE-STATIC:       # %bb.0:
-; LARGE-STATIC-NEXT:    movl $forced_small_data+8, %eax
+; LARGE-STATIC-NEXT:    movabsq $forced_small_data+8, %rax
 ; LARGE-STATIC-NEXT:    movl (%rax), %eax
 ; LARGE-STATIC-NEXT:    retq
 ;

[MaskRay]

64 bit abs relocation without large section flag? Worth clarifying

[aeubanks]

updated description