Auto merge of #113040 - Kobzol:llvm-remark-streamer, r=tmiasko

Add `-Zremark-dir` unstable flag to write LLVM optimization remarks to YAML

This PR adds an option for `rustc` to emit LLVM optimization remarks to a set of YAML files, which can then be digested by existing tools, like https://github.com/OfekShilon/optview2. When `-Cremark-dir` is passed, and remarks are enabled (`-Cremark=all`), the remarks will be now written to the specified directory, **instead** of being printed to standard error output.  The files are named based on the CGU from which they are being generated.

Currently, the remarks are written using the LLVM streaming machinery, directly in the diagnostics handler. It seemed easier than going back to Rust and then form there back to C++ to use the streamer from the diagnostics handler. But there are many ways to implement this, of course, so I'm open to suggestions :)

I included some comments with questions into the code. Also, I'm not sure how to test this.

r? `@tmiasko`

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -268,6 +268,16 @@ pub(crate) fn save_temp_bitcode(
     }
 }
 
+/// In what context is a dignostic handler being attached to a codegen unit?
+pub enum CodegenDiagnosticsStage {
+    /// Prelink optimization stage.
+    Opt,
+    /// LTO/ThinLTO postlink optimization stage.
+    LTO,
+    /// Code generation.
+    Codegen,
+}
+
 pub struct DiagnosticHandlers<'a> {
     data: *mut (&'a CodegenContext<LlvmCodegenBackend>, &'a Handler),
     llcx: &'a llvm::Context,
@@ -279,6 +289,8 @@ impl<'a> DiagnosticHandlers<'a> {
         cgcx: &'a CodegenContext<LlvmCodegenBackend>,
         handler: &'a Handler,
         llcx: &'a llvm::Context,
+        module: &ModuleCodegen<ModuleLlvm>,
+        stage: CodegenDiagnosticsStage,
     ) -> Self {
         let remark_passes_all: bool;
         let remark_passes: Vec<CString>;
@@ -295,6 +307,20 @@ impl<'a> DiagnosticHandlers<'a> {
         };
         let remark_passes: Vec<*const c_char> =
             remark_passes.iter().map(|name: &CString| name.as_ptr()).collect();
+        let remark_file = cgcx
+            .remark_dir
+            .as_ref()
+            // Use the .opt.yaml file suffix, which is supported by LLVM's opt-viewer.
+            .map(|dir| {
+                let stage_suffix = match stage {
+                    CodegenDiagnosticsStage::Codegen => "codegen",
+                    CodegenDiagnosticsStage::Opt => "opt",
+                    CodegenDiagnosticsStage::LTO => "lto",
+                };
+                dir.join(format!("{}.{stage_suffix}.opt.yaml", module.name))
+            })
+            .and_then(|dir| dir.to_str().and_then(|p| CString::new(p).ok()));
+
         let data = Box::into_raw(Box::new((cgcx, handler)));
         unsafe {
             let old_handler = llvm::LLVMRustContextGetDiagnosticHandler(llcx);
@@ -305,6 +331,9 @@ impl<'a> DiagnosticHandlers<'a> {
                 remark_passes_all,
                 remark_passes.as_ptr(),
                 remark_passes.len(),
+                // The `as_ref()` is important here, otherwise the `CString` will be dropped
+                // too soon!
+                remark_file.as_ref().map(|dir| dir.as_ptr()).unwrap_or(std::ptr::null()),
             );
             DiagnosticHandlers { data, llcx, old_handler }
         }
@@ -523,7 +552,8 @@ pub(crate) unsafe fn optimize(
 
     let llmod = module.module_llvm.llmod();
     let llcx = &*module.module_llvm.llcx;
-    let _handlers = DiagnosticHandlers::new(cgcx, diag_handler, llcx);
+    let _handlers =
+        DiagnosticHandlers::new(cgcx, diag_handler, llcx, module, CodegenDiagnosticsStage::Opt);
 
     let module_name = module.name.clone();
     let module_name = Some(&module_name[..]);
@@ -582,7 +612,13 @@ pub(crate) unsafe fn codegen(
         let tm = &*module.module_llvm.tm;
         let module_name = module.name.clone();
         let module_name = Some(&module_name[..]);
-        let handlers = DiagnosticHandlers::new(cgcx, diag_handler, llcx);
+        let _handlers = DiagnosticHandlers::new(
+            cgcx,
+            diag_handler,
+            llcx,
+            &module,
+            CodegenDiagnosticsStage::Codegen,
+        );
 
         if cgcx.msvc_imps_needed {
             create_msvc_imps(cgcx, llcx, llmod);
@@ -775,7 +811,6 @@ pub(crate) unsafe fn codegen(
         }
 
         record_llvm_cgu_instructions_stats(&cgcx.prof, llmod);
-        drop(handlers);
     }
 
     // `.dwo` files are only emitted if: