Auto merge of #83044 - kubo39:set-llvm-code-model, r=nikic

Add support for storing code model to LLVM module IR

This patch avoids undefined behavior by linking different object files.
Also this would it could be propagated properly to LTO.

See https://reviews.llvm.org/D52322 and https://reviews.llvm.org/D52323.

This patch is based on #74002

Author: bors

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -140,7 +140,7 @@ fn to_llvm_relocation_model(relocation_model: RelocModel) -> llvm::RelocModel {
     }
 }
 
-fn to_llvm_code_model(code_model: Option<CodeModel>) -> llvm::CodeModel {
+pub(crate) fn to_llvm_code_model(code_model: Option<CodeModel>) -> llvm::CodeModel {
     match code_model {
         Some(CodeModel::Tiny) => llvm::CodeModel::Tiny,
         Some(CodeModel::Small) => llvm::CodeModel::Small,

compiler/rustc_codegen_llvm/src/context.rs

@@ -1,4 +1,5 @@
 use crate::attributes;
+use crate::back::write::to_llvm_code_model;
 use crate::callee::get_fn;
 use crate::coverageinfo;
 use crate::debuginfo;
@@ -181,6 +182,13 @@ pub unsafe fn create_module(
         }
     }
 
+    // Linking object files with different code models is undefined behavior
+    // because the compiler would have to generate additional code (to span
+    // longer jumps) if a larger code model is used with a smaller one.
+    //
+    // See https://reviews.llvm.org/D52322 and https://reviews.llvm.org/D52323.
+    llvm::LLVMRustSetModuleCodeModel(llmod, to_llvm_code_model(sess.code_model()));
+
     // If skipping the PLT is enabled, we need to add some module metadata
     // to ensure intrinsic calls don't use it.
     if !sess.needs_plt() {

compiler/rustc_codegen_llvm/src/llvm/ffi.rs

@@ -2326,6 +2326,7 @@ extern "C" {
     pub fn LLVMRustUnsetComdat(V: &Value);
     pub fn LLVMRustSetModulePICLevel(M: &Module);
     pub fn LLVMRustSetModulePIELevel(M: &Module);
+    pub fn LLVMRustSetModuleCodeModel(M: &Module, Model: CodeModel);
     pub fn LLVMRustModuleBufferCreate(M: &Module) -> &'static mut ModuleBuffer;
     pub fn LLVMRustModuleBufferPtr(p: &ModuleBuffer) -> *const u8;
     pub fn LLVMRustModuleBufferLen(p: &ModuleBuffer) -> usize;

compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp

@@ -1264,6 +1264,14 @@ extern "C" void LLVMRustSetModulePIELevel(LLVMModuleRef M) {
   unwrap(M)->setPIELevel(PIELevel::Level::Large);
 }
 
+extern "C" void LLVMRustSetModuleCodeModel(LLVMModuleRef M,
+                                           LLVMRustCodeModel Model) {
+  auto CM = fromRust(Model);
+  if (!CM.hasValue())
+    return;
+  unwrap(M)->setCodeModel(*CM);
+}
+
 // Here you'll find an implementation of ThinLTO as used by the Rust compiler
 // right now. This ThinLTO support is only enabled on "recent ish" versions of
 // LLVM, and otherwise it's just blanket rejected from other compilers.

src/test/codegen/codemodels.rs

@@ -0,0 +1,20 @@
+// only-x86_64
+
+// revisions: NOMODEL MODEL-SMALL MODEL-KERNEL MODEL-MEDIUM MODEL-LARGE
+//[NOMODEL] compile-flags:
+//[MODEL-SMALL] compile-flags: -C code-model=small
+//[MODEL-KERNEL] compile-flags: -C code-model=kernel
+//[MODEL-MEDIUM] compile-flags: -C code-model=medium
+//[MODEL-LARGE] compile-flags: -C code-model=large
+
+#![crate_type = "lib"]
+
+// MODEL-SMALL: !llvm.module.flags = !{{{.*}}}
+// MODEL-SMALL: !{{[0-9]+}} = !{i32 1, !"Code Model", i32 1}
+// MODEL-KERNEL: !llvm.module.flags = !{{{.*}}}
+// MODEL-KERNEL: !{{[0-9]+}} = !{i32 1, !"Code Model", i32 2}
+// MODEL-MEDIUM: !llvm.module.flags = !{{{.*}}}
+// MODEL-MEDIUM: !{{[0-9]+}} = !{i32 1, !"Code Model", i32 3}
+// MODEL-LARGE: !llvm.module.flags = !{{{.*}}}
+// MODEL-LARGE: !{{[0-9]+}} = !{i32 1, !"Code Model", i32 4}
+// NOMODEL-NOT: Code Model