Auto merge of rust-lang#98162 - nextsilicon:support_lto_embed_bitcode…

…, r=davidtwco

Allow to disable thinLTO buffer to support lto-embed-bitcode lld feature

Hello
This change is to fix issue (rust-lang#84395) in which passing "-lto-embed-bitcode=optimized" to lld when linking rust code via linker-plugin-lto doesn't produce the expected result.

Instead of emitting a single unified module into a llvmbc section of the linked elf, it emits multiple submodules.
This is caused because rustc emits the BC modules after running llvm `createWriteThinLTOBitcodePass` pass.
Which in turn triggers a thinLTO linkage and causes the said issue.

This patch allows via compiler flag (-Cemit-thin-lto=<bool>) to select between running `createWriteThinLTOBitcodePass` and `createBitcodeWriterPass`.
Note this pattern of selecting between those 2 passes is common inside of LLVM code.
The default is to match the old behavior.

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -199,7 +199,7 @@ pub(crate) fn run_thin(
 
 pub(crate) fn prepare_thin(module: ModuleCodegen<ModuleLlvm>) -> (String, ThinBuffer) {
     let name = module.name.clone();
-    let buffer = ThinBuffer::new(module.module_llvm.llmod());
+    let buffer = ThinBuffer::new(module.module_llvm.llmod(), true);
     (name, buffer)
 }
 
@@ -695,9 +695,9 @@ unsafe impl Send for ThinBuffer {}
 unsafe impl Sync for ThinBuffer {}
 
 impl ThinBuffer {
-    pub fn new(m: &llvm::Module) -> ThinBuffer {
+    pub fn new(m: &llvm::Module, is_thin: bool) -> ThinBuffer {
         unsafe {
-            let buffer = llvm::LLVMRustThinLTOBufferCreate(m);
+            let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin);
             ThinBuffer(buffer)
         }
     }

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -790,7 +790,7 @@ pub(crate) unsafe fn codegen(
             let _timer = cgcx
                 .prof
                 .generic_activity_with_arg("LLVM_module_codegen_make_bitcode", &*module.name);
-            let thin = ThinBuffer::new(llmod);
+            let thin = ThinBuffer::new(llmod, config.emit_thin_lto);
             let data = thin.data();
 
             if let Some(bitcode_filename) = bc_out.file_name() {

compiler/rustc_codegen_llvm/src/llvm/ffi.rs

@@ -2470,7 +2470,7 @@ extern "C" {
     pub fn LLVMRustModuleBufferFree(p: &'static mut ModuleBuffer);
     pub fn LLVMRustModuleCost(M: &Module) -> u64;
 
-    pub fn LLVMRustThinLTOBufferCreate(M: &Module) -> &'static mut ThinLTOBuffer;
+    pub fn LLVMRustThinLTOBufferCreate(M: &Module, is_thin: bool) -> &'static mut ThinLTOBuffer;
     pub fn LLVMRustThinLTOBufferFree(M: &'static mut ThinLTOBuffer);
     pub fn LLVMRustThinLTOBufferPtr(M: &ThinLTOBuffer) -> *const c_char;
     pub fn LLVMRustThinLTOBufferLen(M: &ThinLTOBuffer) -> size_t;

compiler/rustc_codegen_ssa/src/back/write.rs

@@ -99,6 +99,7 @@ pub struct ModuleConfig {
     pub emit_ir: bool,
     pub emit_asm: bool,
     pub emit_obj: EmitObj,
+    pub emit_thin_lto: bool,
     pub bc_cmdline: String,
 
     // Miscellaneous flags.  These are mostly copied from command-line
@@ -218,6 +219,7 @@ impl ModuleConfig {
                 false
             ),
             emit_obj,
+            emit_thin_lto: sess.opts.unstable_opts.emit_thin_lto,
             bc_cmdline: sess.target.bitcode_llvm_cmdline.to_string(),
 
             verify_llvm_ir: sess.verify_llvm_ir(),

compiler/rustc_interface/src/tests.rs

@@ -735,6 +735,7 @@ fn test_unstable_options_tracking_hash() {
     tracked!(drop_tracking, true);
     tracked!(dual_proc_macros, true);
     tracked!(dwarf_version, Some(5));
+    tracked!(emit_thin_lto, false);
     tracked!(fewer_names, Some(true));
     tracked!(force_unstable_if_unmarked, true);
     tracked!(fuel, Some(("abc".to_string(), 99)));

compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp

@@ -34,6 +34,7 @@
 #include "llvm/Transforms/Utils/AddDiscriminators.h"
 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
 #include "llvm/LTO/LTO.h"
+#include "llvm/Bitcode/BitcodeWriterPass.h"
 #include "llvm-c/Transforms/PassManagerBuilder.h"
 
 #include "llvm/Transforms/Instrumentation.h"
@@ -1638,13 +1639,17 @@ struct LLVMRustThinLTOBuffer {
 };
 
 extern "C" LLVMRustThinLTOBuffer*
-LLVMRustThinLTOBufferCreate(LLVMModuleRef M) {
+LLVMRustThinLTOBufferCreate(LLVMModuleRef M, bool is_thin) {
   auto Ret = std::make_unique<LLVMRustThinLTOBuffer>();
   {
     raw_string_ostream OS(Ret->data);
     {
       legacy::PassManager PM;
-      PM.add(createWriteThinLTOBitcodePass(OS));
+      if (is_thin) {
+        PM.add(createWriteThinLTOBitcodePass(OS));
+      } else {
+        PM.add(createBitcodeWriterPass(OS));
+      }
       PM.run(*unwrap(M));
     }
   }

compiler/rustc_session/src/options.rs

@@ -1279,6 +1279,8 @@ options! {
         "version of DWARF debug information to emit (default: 2 or 4, depending on platform)"),
     emit_stack_sizes: bool = (false, parse_bool, [UNTRACKED],
         "emit a section containing stack size metadata (default: no)"),
+    emit_thin_lto: bool = (true, parse_bool, [TRACKED],
+        "emit the bc module with thin LTO info (default: yes)"),
     fewer_names: Option<bool> = (None, parse_opt_bool, [TRACKED],
         "reduce memory use by retaining fewer names within compilation artifacts (LLVM-IR) \
         (default: no)"),

src/test/run-make-fulldeps/issue-84395-lto-embed-bitcode/Makefile

@@ -0,0 +1,11 @@
+# needs-matching-clang
+
+# This test makes sure the embed bitcode in elf created with
+# lto-embed-bitcode=optimized is valid llvm BC module.
+
+-include ../../run-make-fulldeps/tools.mk
+
+all:
+	$(RUSTC) test.rs --target $(TARGET) -Clink-arg=-fuse-ld=lld -Clinker-plugin-lto -Clinker=$(CLANG) -Clink-arg=-Wl,--plugin-opt=-lto-embed-bitcode=optimized -Zemit-thin-lto=no
+	$(LLVM_BIN_DIR)/objcopy --dump-section .llvmbc=$(TMPDIR)/test.bc $(TMPDIR)/test
+	$(LLVM_BIN_DIR)/llvm-dis $(TMPDIR)/test.bc

src/test/run-make-fulldeps/issue-84395-lto-embed-bitcode/test.rs

@@ -0,0 +1,3 @@
+fn main() {
+    println!("Hello World!");
+}

src/test/rustdoc-ui/z-help.stdout

@@ -36,6 +36,7 @@
     -Z                       dump-mir-spanview=val -- in addition to `.mir` files, create `.html` files to view spans for all `statement`s (including terminators), only `terminator` spans, or computed `block` spans (one span encompassing a block's terminator and all statements). If `-Z instrument-coverage` is also enabled, create an additional `.html` file showing the computed coverage spans.
     -Z                           dwarf-version=val -- version of DWARF debug information to emit (default: 2 or 4, depending on platform)
     -Z                        emit-stack-sizes=val -- emit a section containing stack size metadata (default: no)
+    -Z                           emit-thin-lto=val -- emit the bc module with thin LTO info (default: yes)
     -Z                             fewer-names=val -- reduce memory use by retaining fewer names within compilation artifacts (LLVM-IR) (default: no)
     -Z              force-unstable-if-unmarked=val -- force all crates to be `rustc_private` unstable (default: no)
     -Z                                    fuel=val -- set the optimization fuel quota for a crate