Merge commit '59a81c2ca1edc88ad3ac4b27a8e03977ffb8e73a' into subtree-update_cg_gcc_2025_01_12

2 files changed, 132 insertions, 62 deletions

diff --git a/compiler/rustc_codegen_gcc/src/back/lto.rs b/compiler/rustc_codegen_gcc/src/back/lto.rs
index f7173d4d2ff..e419bd18099 100644
--- a/compiler/rustc_codegen_gcc/src/back/lto.rs
+++ b/compiler/rustc_codegen_gcc/src/back/lto.rs
@@ -35,16 +35,13 @@ use rustc_middle::bug;
 use rustc_middle::dep_graph::WorkProduct;
 use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel};
 use rustc_session::config::{CrateType, Lto};
+use rustc_target::spec::RelocModel;
 use tempfile::{TempDir, tempdir};
 
 use crate::back::write::save_temp_bitcode;
 use crate::errors::{DynamicLinkingWithLTO, LtoBitcodeFromRlib, LtoDisallowed, LtoDylib};
 use crate::{GccCodegenBackend, GccContext, SyncContext, to_gcc_opt_level};
 
-/// We keep track of the computed LTO cache keys from the previous
-/// session to determine which CGUs we can reuse.
-//pub const THIN_LTO_KEYS_INCR_COMP_FILE_NAME: &str = "thin-lto-past-keys.bin";
-
 pub fn crate_type_allows_lto(crate_type: CrateType) -> bool {
     match crate_type {
         CrateType::Executable | CrateType::Dylib | CrateType::Staticlib | CrateType::Cdylib => true,
@@ -54,7 +51,7 @@ pub fn crate_type_allows_lto(crate_type: CrateType) -> bool {
 
 struct LtoData {
     // TODO(antoyo): use symbols_below_threshold.
-    //symbols_below_threshold: Vec<CString>,
+    //symbols_below_threshold: Vec<String>,
     upstream_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
     tmp_path: TempDir,
 }
@@ -83,7 +80,7 @@ fn prepare_lto(
 
     let symbol_filter = &|&(ref name, info): &(String, SymbolExportInfo)| {
         if info.level.is_below_threshold(export_threshold) || info.used {
-            Some(CString::new(name.as_str()).unwrap())
+            Some(name.clone())
         } else {
             None
         }
@@ -91,7 +88,7 @@ fn prepare_lto(
     let exported_symbols = cgcx.exported_symbols.as_ref().expect("needs exported symbols for LTO");
     let mut symbols_below_threshold = {
         let _timer = cgcx.prof.generic_activity("GCC_lto_generate_symbols_below_threshold");
-        exported_symbols[&LOCAL_CRATE].iter().filter_map(symbol_filter).collect::<Vec<CString>>()
+        exported_symbols[&LOCAL_CRATE].iter().filter_map(symbol_filter).collect::<Vec<String>>()
     };
     info!("{} symbols to preserve in this crate", symbols_below_threshold.len());
 
@@ -159,11 +156,7 @@ fn prepare_lto(
         }
     }
 
-    Ok(LtoData {
-        //symbols_below_threshold,
-        upstream_modules,
-        tmp_path,
-    })
+    Ok(LtoData { upstream_modules, tmp_path })
 }
 
 fn save_as_file(obj: &[u8], path: &Path) -> Result<(), LtoBitcodeFromRlib> {
@@ -191,7 +184,7 @@ pub(crate) fn run_fat(
         cached_modules,
         lto_data.upstream_modules,
         lto_data.tmp_path,
-        //&symbols_below_threshold,
+        //&lto_data.symbols_below_threshold,
     )
 }
 
@@ -202,7 +195,7 @@ fn fat_lto(
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
     mut serialized_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
     tmp_path: TempDir,
-    //symbols_below_threshold: &[*const libc::c_char],
+    //symbols_below_threshold: &[String],
 ) -> Result<LtoModuleCodegen<GccCodegenBackend>, FatalError> {
     let _timer = cgcx.prof.generic_activity("GCC_fat_lto_build_monolithic_module");
     info!("going for a fat lto");
@@ -327,6 +320,7 @@ fn fat_lto(
             ptr as *const *const libc::c_char,
             symbols_below_threshold.len() as libc::size_t,
         );*/
+
         save_temp_bitcode(cgcx, &module, "lto.after-restriction");
         //}
     }
@@ -363,8 +357,6 @@ pub(crate) fn run_thin(
     let dcx = cgcx.create_dcx();
     let dcx = dcx.handle();
     let lto_data = prepare_lto(cgcx, dcx)?;
-    /*let symbols_below_threshold =
-    symbols_below_threshold.iter().map(|c| c.as_ptr()).collect::<Vec<_>>();*/
     if cgcx.opts.cg.linker_plugin_lto.enabled() {
         unreachable!(
             "We should never reach this case if the LTO step \
@@ -377,7 +369,8 @@ pub(crate) fn run_thin(
         modules,
         lto_data.upstream_modules,
         lto_data.tmp_path,
-        cached_modules, /*, &symbols_below_threshold*/
+        cached_modules,
+        //&lto_data.symbols_below_threshold,
     )
 }
 
@@ -428,7 +421,7 @@ fn thin_lto(
     serialized_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
     tmp_path: TempDir,
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
-    //symbols_below_threshold: &[*const libc::c_char],
+    //_symbols_below_threshold: &[String],
 ) -> Result<(Vec<LtoModuleCodegen<GccCodegenBackend>>, Vec<WorkProduct>), FatalError> {
     let _timer = cgcx.prof.generic_activity("LLVM_thin_lto_global_analysis");
     info!("going for that thin, thin LTO");
@@ -640,7 +633,13 @@ pub unsafe fn optimize_thin_module(
         }
     };
     let module = ModuleCodegen {
-        module_llvm: GccContext { context, should_combine_object_files, temp_dir: None },
+        module_llvm: GccContext {
+            context,
+            should_combine_object_files,
+            // TODO(antoyo): use the correct relocation model here.
+            relocation_model: RelocModel::Pic,
+            temp_dir: None,
+        },
         name: thin_module.name().to_string(),
         kind: ModuleKind::Regular,
     };
diff --git a/compiler/rustc_codegen_gcc/src/back/write.rs b/compiler/rustc_codegen_gcc/src/back/write.rs
index 802968979c7..51c5ba73e32 100644
--- a/compiler/rustc_codegen_gcc/src/back/write.rs
+++ b/compiler/rustc_codegen_gcc/src/back/write.rs
@@ -1,6 +1,6 @@
 use std::{env, fs};
 
-use gccjit::OutputKind;
+use gccjit::{Context, OutputKind};
 use rustc_codegen_ssa::back::link::ensure_removed;
 use rustc_codegen_ssa::back::write::{BitcodeSection, CodegenContext, EmitObj, ModuleConfig};
 use rustc_codegen_ssa::{CompiledModule, ModuleCodegen};
@@ -10,6 +10,7 @@ use rustc_session::config::OutputType;
 use rustc_span::fatal_error::FatalError;
 use rustc_target::spec::SplitDebuginfo;
 
+use crate::base::add_pic_option;
 use crate::errors::CopyBitcode;
 use crate::{GccCodegenBackend, GccContext};
 
@@ -31,51 +32,87 @@ pub(crate) unsafe fn codegen(
 
         // NOTE: Only generate object files with GIMPLE when this environment variable is set for
         // now because this requires a particular setup (same gcc/lto1/lto-wrapper commit as libgccjit).
-        // TODO: remove this environment variable.
+        // TODO(antoyo): remove this environment variable.
         let fat_lto = env::var("EMBED_LTO_BITCODE").as_deref() == Ok("1");
 
         let bc_out = cgcx.output_filenames.temp_path(OutputType::Bitcode, module_name);
         let obj_out = cgcx.output_filenames.temp_path(OutputType::Object, module_name);
 
-        if config.bitcode_needed() && fat_lto {
-            let _timer = cgcx
-                .prof
-                .generic_activity_with_arg("GCC_module_codegen_make_bitcode", &*module.name);
-
-            // TODO(antoyo)
-            /*if let Some(bitcode_filename) = bc_out.file_name() {
-                cgcx.prof.artifact_size(
-                    "llvm_bitcode",
-                    bitcode_filename.to_string_lossy(),
-                    data.len() as u64,
-                );
-            }*/
-
-            if config.emit_bc || config.emit_obj == EmitObj::Bitcode {
+        if config.bitcode_needed() {
+            if fat_lto {
                 let _timer = cgcx
                     .prof
-                    .generic_activity_with_arg("GCC_module_codegen_emit_bitcode", &*module.name);
-                context.add_command_line_option("-flto=auto");
-                context.add_command_line_option("-flto-partition=one");
-                // TODO: remove since we don't want fat objects when it is for Bitcode only.
-                context.add_command_line_option("-ffat-lto-objects");
-                context
-                    .compile_to_file(OutputKind::ObjectFile, bc_out.to_str().expect("path to str"));
-            }
+                    .generic_activity_with_arg("GCC_module_codegen_make_bitcode", &*module.name);
 
-            if config.emit_obj == EmitObj::ObjectCode(BitcodeSection::Full) {
-                let _timer = cgcx
-                    .prof
-                    .generic_activity_with_arg("GCC_module_codegen_embed_bitcode", &*module.name);
-                // TODO(antoyo): maybe we should call embed_bitcode to have the proper iOS fixes?
-                //embed_bitcode(cgcx, llcx, llmod, &config.bc_cmdline, data);
-
-                context.add_command_line_option("-flto=auto");
-                context.add_command_line_option("-flto-partition=one");
-                context.add_command_line_option("-ffat-lto-objects");
-                // TODO(antoyo): Send -plugin/usr/lib/gcc/x86_64-pc-linux-gnu/11.1.0/liblto_plugin.so to linker (this should be done when specifying the appropriate rustc cli argument).
-                context
-                    .compile_to_file(OutputKind::ObjectFile, bc_out.to_str().expect("path to str"));
+                // TODO(antoyo)
+                /*if let Some(bitcode_filename) = bc_out.file_name() {
+                    cgcx.prof.artifact_size(
+                        "llvm_bitcode",
+                        bitcode_filename.to_string_lossy(),
+                        data.len() as u64,
+                    );
+                }*/
+
+                if config.emit_bc || config.emit_obj == EmitObj::Bitcode {
+                    let _timer = cgcx.prof.generic_activity_with_arg(
+                        "GCC_module_codegen_emit_bitcode",
+                        &*module.name,
+                    );
+                    context.add_command_line_option("-flto=auto");
+                    context.add_command_line_option("-flto-partition=one");
+                    // TODO(antoyo): remove since we don't want fat objects when it is for Bitcode only.
+                    context.add_command_line_option("-ffat-lto-objects");
+                    context.compile_to_file(
+                        OutputKind::ObjectFile,
+                        bc_out.to_str().expect("path to str"),
+                    );
+                }
+
+                if config.emit_obj == EmitObj::ObjectCode(BitcodeSection::Full) {
+                    let _timer = cgcx.prof.generic_activity_with_arg(
+                        "GCC_module_codegen_embed_bitcode",
+                        &*module.name,
+                    );
+                    // TODO(antoyo): maybe we should call embed_bitcode to have the proper iOS fixes?
+                    //embed_bitcode(cgcx, llcx, llmod, &config.bc_cmdline, data);
+
+                    context.add_command_line_option("-flto=auto");
+                    context.add_command_line_option("-flto-partition=one");
+                    context.add_command_line_option("-ffat-lto-objects");
+                    // TODO(antoyo): Send -plugin/usr/lib/gcc/x86_64-pc-linux-gnu/11.1.0/liblto_plugin.so to linker (this should be done when specifying the appropriate rustc cli argument).
+                    context.compile_to_file(
+                        OutputKind::ObjectFile,
+                        bc_out.to_str().expect("path to str"),
+                    );
+                }
+            } else {
+                if config.emit_bc || config.emit_obj == EmitObj::Bitcode {
+                    let _timer = cgcx.prof.generic_activity_with_arg(
+                        "GCC_module_codegen_emit_bitcode",
+                        &*module.name,
+                    );
+                    context.compile_to_file(
+                        OutputKind::ObjectFile,
+                        bc_out.to_str().expect("path to str"),
+                    );
+                }
+
+                if config.emit_obj == EmitObj::ObjectCode(BitcodeSection::Full) {
+                    // TODO(antoyo): we might want to emit to emit an error here, saying to set the
+                    // environment variable EMBED_LTO_BITCODE.
+                    let _timer = cgcx.prof.generic_activity_with_arg(
+                        "GCC_module_codegen_embed_bitcode",
+                        &*module.name,
+                    );
+                    // TODO(antoyo): maybe we should call embed_bitcode to have the proper iOS fixes?
+                    //embed_bitcode(cgcx, llcx, llmod, &config.bc_cmdline, data);
+
+                    // TODO(antoyo): Send -plugin/usr/lib/gcc/x86_64-pc-linux-gnu/11.1.0/liblto_plugin.so to linker (this should be done when specifying the appropriate rustc cli argument).
+                    context.compile_to_file(
+                        OutputKind::ObjectFile,
+                        bc_out.to_str().expect("path to str"),
+                    );
+                }
             }
         }
 
@@ -123,6 +160,8 @@ pub(crate) unsafe fn codegen(
 
                         // NOTE: without -fuse-linker-plugin, we get the following error:
                         // lto1: internal compiler error: decompressed stream: Destination buffer is too small
+                        // TODO(antoyo): since we do not do LTO when the linker is invoked anymore, perhaps
+                        // the following flag is not necessary anymore.
                         context.add_driver_option("-fuse-linker-plugin");
                     }
 
@@ -131,11 +170,43 @@ pub(crate) unsafe fn codegen(
                     // /usr/bin/ld: cannot find -lgcc_s: No such file or directory
                     context.add_driver_option("-nostdlib");
 
-                    // NOTE: this doesn't actually generate an executable. With the above flags, it combines the .o files together in another .o.
-                    context.compile_to_file(
-                        OutputKind::Executable,
-                        obj_out.to_str().expect("path to str"),
-                    );
+                    let path = obj_out.to_str().expect("path to str");
+
+                    if fat_lto {
+                        let lto_path = format!("{}.lto", path);
+                        // FIXME(antoyo): The LTO frontend generates the following warning:
+                        // ../build_sysroot/sysroot_src/library/core/src/num/dec2flt/lemire.rs:150:15: warning: type of ‘_ZN4core3num7dec2flt5table17POWER_OF_FIVE_12817ha449a68fb31379e4E’ does not match original declaration [-Wlto-type-mismatch]
+                        // 150 |     let (lo5, hi5) = POWER_OF_FIVE_128[index];
+                        //     |               ^
+                        // lto1: note: ‘_ZN4core3num7dec2flt5table17POWER_OF_FIVE_12817ha449a68fb31379e4E’ was previously declared here
+                        //
+                        // This option is to mute it to make the UI tests pass with LTO enabled.
+                        context.add_driver_option("-Wno-lto-type-mismatch");
+                        // NOTE: this doesn't actually generate an executable. With the above
+                        // flags, it combines the .o files together in another .o.
+                        context.compile_to_file(OutputKind::Executable, &lto_path);
+
+                        let context = Context::default();
+                        if cgcx.target_arch == "x86" || cgcx.target_arch == "x86_64" {
+                            // NOTE: it seems we need to use add_driver_option instead of
+                            // add_command_line_option here because we use the LTO frontend via gcc.
+                            context.add_driver_option("-masm=intel");
+                        }
+
+                        // NOTE: these two options are needed to invoke LTO to produce an object file.
+                        // We need to initiate a second compilation because the arguments "-x lto"
+                        // needs to be at the very beginning.
+                        context.add_driver_option("-x");
+                        context.add_driver_option("lto");
+                        add_pic_option(&context, module.module_llvm.relocation_model);
+                        context.add_driver_option(lto_path);
+
+                        context.compile_to_file(OutputKind::ObjectFile, path);
+                    } else {
+                        // NOTE: this doesn't actually generate an executable. With the above
+                        // flags, it combines the .o files together in another .o.
+                        context.compile_to_file(OutputKind::Executable, path);
+                    }
                 } else {
                     context.compile_to_file(
                         OutputKind::ObjectFile,