Auto merge of #67502 - Mark-Simulacrum:opt-catch, r=<try>

Optimize catch_unwind to match C++ try/catch

This refactors the implementation of catching unwinds to allow LLVM to inline the "try" closure directly into the happy path, avoiding indirection. This means that the catch_unwind implementation is (after this PR) zero-cost unless a panic is thrown.

https://rust.godbolt.org/z/cZcUSB is an example of the current codegen in a simple case. Notably, the codegen is *exactly the same* if `-Cpanic=abort` is passed, which is clearly not great.

This PR, on the other hand, generates the following assembly:

```asm
# -Cpanic=unwind:
	push   rbx
	mov    ebx,0x2a
	call   QWORD PTR [rip+0x1c53c]        # <happy>
	mov    eax,ebx
	pop    rbx
	ret
	mov    rdi,rax
	call   QWORD PTR [rip+0x1c537]        # cleanup function call
	call   QWORD PTR [rip+0x1c539]        # <unfortunate>
	mov    ebx,0xd
	mov    eax,ebx
	pop    rbx
	ret

# -Cpanic=abort:
	push   rax
	call   QWORD PTR [rip+0x20a1]        # <happy>
	mov    eax,0x2a
	pop    rcx
	ret
```

Fixes #64224, and resolves #64222.

Author: bors

Diff for: .gitmodules

@@ -39,8 +39,8 @@
 	url = https://github.com/rust-lang/edition-guide.git
 [submodule "src/llvm-project"]
 	path = src/llvm-project
-	url = https://github.com/rust-lang/llvm-project.git
-	branch = rustc/9.0-2019-12-19
+	url = https://github.com/Amanieu/llvm-project.git
+	branch = opt-catch
 [submodule "src/doc/embedded-book"]
 	path = src/doc/embedded-book
 	url = https://github.com/rust-embedded/book.git

Diff for: Cargo.lock

@@ -2308,6 +2308,7 @@ dependencies = [
 name = "panic_abort"
 version = "0.0.0"
 dependencies = [
+ "cfg-if",
  "compiler_builtins",
  "core",
  "libc",

Diff for: src/ci/azure-pipelines/try.yml

@@ -6,17 +6,17 @@ variables:
 - group: prod-credentials
 
 jobs:
-- job: Linux
-  timeoutInMinutes: 600
-  pool:
-    vmImage: ubuntu-16.04
-  steps:
-  - template: steps/run.yml
-  strategy:
-    matrix:
-      dist-x86_64-linux: {}
-      dist-x86_64-linux-alt:
-        IMAGE: dist-x86_64-linux
+    # - job: Linux
+    #   timeoutInMinutes: 600
+    #   pool:
+    #     vmImage: ubuntu-16.04
+    #   steps:
+    #   - template: steps/run.yml
+    #   strategy:
+    #     matrix:
+    #       dist-x86_64-linux: {}
+    #       dist-x86_64-linux-alt:
+    #         IMAGE: dist-x86_64-linux
 
 # The macOS and Windows builds here are currently disabled due to them not being
 # overly necessary on `try` builds. We also don't actually have anything that
@@ -49,25 +49,20 @@ jobs:
 #         NO_LLVM_ASSERTIONS: 1
 #         NO_DEBUG_ASSERTIONS: 1
 #
-# - job: Windows
-#   timeoutInMinutes: 600
-#   pool:
-#     vmImage: 'vs2017-win2016'
-#   steps:
-#   - template: steps/run.yml
-#   strategy:
-#     matrix:
-#       dist-x86_64-msvc:
-#         RUST_CONFIGURE_ARGS: >
-#           --build=x86_64-pc-windows-msvc
-#           --target=x86_64-pc-windows-msvc,aarch64-pc-windows-msvc
-#           --enable-full-tools
-#           --enable-profiler
-#         SCRIPT: python x.py dist
-#         DIST_REQUIRE_ALL_TOOLS: 1
-#         DEPLOY: 1
-#
-#       dist-x86_64-msvc-alt:
-#         RUST_CONFIGURE_ARGS: --build=x86_64-pc-windows-msvc --enable-extended --enable-profiler
-#         SCRIPT: python x.py dist
-#         DEPLOY_ALT: 1
+- job: Windows
+  timeoutInMinutes: 600
+  pool:
+    vmImage: 'vs2017-win2016'
+  steps:
+  - template: steps/run.yml
+  strategy:
+    matrix:
+      i686-mingw-2:
+        RUST_CONFIGURE_ARGS: --build=i686-pc-windows-gnu
+        SCRIPT: make ci-mingw-subset-2
+        CUSTOM_MINGW: 1
+      dist-i686-mingw:
+        RUST_CONFIGURE_ARGS: --build=i686-pc-windows-gnu --enable-debug-assertions --enable-optimize --debuginfo-level=1
+        SCRIPT: python x.py dist
+        CUSTOM_MINGW: 1
+        N_DIST_REQUIRE_ALL_TOOLS: 1

Diff for: src/doc/unstable-book/src/language-features/lang-items.md

@@ -52,7 +52,6 @@ fn main(_argc: isize, _argv: *const *const u8) -> isize {
 
 #[lang = "eh_personality"] extern fn rust_eh_personality() {}
 #[lang = "panic_impl"] extern fn rust_begin_panic(info: &PanicInfo) -> ! { unsafe { intrinsics::abort() } }
-#[lang = "eh_unwind_resume"] extern fn rust_eh_unwind_resume() {}
 #[no_mangle] pub extern fn rust_eh_register_frames () {}
 #[no_mangle] pub extern fn rust_eh_unregister_frames () {}
 ```
@@ -67,7 +66,7 @@ Other features provided by lang items include:
   marked with lang items; those specific four are `eq`, `ord`,
   `deref`, and `add` respectively.
 - stack unwinding and general failure; the `eh_personality`,
-  `eh_unwind_resume`, `fail` and `fail_bounds_checks` lang items.
+  `fail` and `fail_bounds_checks` lang items.
 - the traits in `std::marker` used to indicate types of
   various kinds; lang items `send`, `sync` and `copy`.
 - the marker types and variance indicators found in
@@ -130,12 +129,6 @@ fn start(_argc: isize, _argv: *const *const u8) -> isize {
 pub extern fn rust_eh_personality() {
 }
 
-// This function may be needed based on the compilation target.
-#[lang = "eh_unwind_resume"]
-#[no_mangle]
-pub extern fn rust_eh_unwind_resume() {
-}
-
 #[lang = "panic_impl"]
 #[no_mangle]
 pub extern fn rust_begin_panic(info: &PanicInfo) -> ! {
@@ -173,12 +166,6 @@ pub extern fn main(_argc: i32, _argv: *const *const u8) -> i32 {
 pub extern fn rust_eh_personality() {
 }
 
-// This function may be needed based on the compilation target.
-#[lang = "eh_unwind_resume"]
-#[no_mangle]
-pub extern fn rust_eh_unwind_resume() {
-}
-
 #[lang = "panic_impl"]
 #[no_mangle]
 pub extern fn rust_begin_panic(info: &PanicInfo) -> ! {
@@ -213,8 +200,8 @@ the screen. While the language item's name is `panic_impl`, the symbol name is
 
 A third function, `rust_eh_unwind_resume`, is also needed if the `custom_unwind_resume`
 flag is set in the options of the compilation target. It allows customizing the
-process of resuming unwind at the end of the landing pads. The language item's name
-is `eh_unwind_resume`.
+process of resuming unwind at the end of the landing pads. Since this function
+must be defined in assembly code it does not have an associated language item.
 
 ## List of all language items
 
@@ -247,9 +234,8 @@ the source code.
   - `eh_personality`: `libpanic_unwind/emcc.rs` (EMCC)
   - `eh_personality`: `libpanic_unwind/gcc.rs` (GNU)
   - `eh_personality`: `libpanic_unwind/seh.rs` (SEH)
-  - `eh_unwind_resume`: `libpanic_unwind/gcc.rs` (GCC)
-  - `eh_catch_typeinfo`: `libpanic_unwind/seh.rs` (SEH)
   - `eh_catch_typeinfo`: `libpanic_unwind/emcc.rs` (EMCC)
+  - `rust_eh_unwind_resume`: `libpanic_unwind/gcc.rs` (GCC)
   - `panic`: `libcore/panicking.rs`
   - `panic_bounds_check`: `libcore/panicking.rs`
   - `panic_impl`: `libcore/panicking.rs`

Diff for: src/libpanic_abort/Cargo.toml

@@ -14,3 +14,4 @@ doc = false
 core = { path = "../libcore" }
 libc = { version = "0.2", default-features = false }
 compiler_builtins = "0.1.0"
+cfg-if = "0.1.8"

Diff for: src/libpanic_abort/lib.rs

@@ -18,17 +18,14 @@
 #![feature(staged_api)]
 #![feature(rustc_attrs)]
 
-// Rust's "try" function, but if we're aborting on panics we just call the
-// function as there's nothing else we need to do here.
+use core::any::Any;
+
+// We need the definition of TryPayload for __rust_panic_cleanup.
+include!("../libpanic_unwind/payload.rs");
+
 #[rustc_std_internal_symbol]
-pub unsafe extern "C" fn __rust_maybe_catch_panic(
-    f: fn(*mut u8),
-    data: *mut u8,
-    _data_ptr: *mut usize,
-    _vtable_ptr: *mut usize,
-) -> u32 {
-    f(data);
-    0
+pub unsafe extern "C" fn __rust_panic_cleanup(_: TryPayload) -> *mut (dyn Any + Send + 'static) {
+    unreachable!()
 }
 
 // "Leak" the payload and shim to the relevant abort on the platform in

Diff for: src/libpanic_unwind/dummy.rs

@@ -6,10 +6,6 @@ use alloc::boxed::Box;
 use core::any::Any;
 use core::intrinsics;
 
-pub fn payload() -> *mut u8 {
-    core::ptr::null_mut()
-}
-
 pub unsafe fn cleanup(_ptr: *mut u8) -> Box<dyn Any + Send> {
     intrinsics::abort()
 }

Diff for: src/libpanic_unwind/emcc.rs

@@ -48,10 +48,6 @@ static EXCEPTION_TYPE_INFO: TypeInfo = TypeInfo {
     name: b"rust_panic\0".as_ptr(),
 };
 
-pub fn payload() -> *mut u8 {
-    ptr::null_mut()
-}
-
 struct Exception {
     // This needs to be an Option because the object's lifetime follows C++
     // semantics: when catch_unwind moves the Box out of the exception it must

Diff for: src/libpanic_unwind/gcc.rs

@@ -48,7 +48,6 @@
 
 use alloc::boxed::Box;
 use core::any::Any;
-use core::ptr;
 
 use crate::dwarf::eh::{self, EHAction, EHContext};
 use libc::{c_int, uintptr_t};
@@ -83,10 +82,6 @@ pub unsafe fn panic(data: Box<dyn Any + Send>) -> u32 {
     }
 }
 
-pub fn payload() -> *mut u8 {
-    ptr::null_mut()
-}
-
 pub unsafe fn cleanup(ptr: *mut u8) -> Box<dyn Any + Send> {
     let exception = Box::from_raw(ptr as *mut Exception);
     exception.cause
@@ -331,6 +326,27 @@ unsafe fn find_eh_action(
 
 // See docs in the `unwind` module.
 #[cfg(all(
+    not(bootstrap),
+    target_os = "windows",
+    any(target_arch = "x86", target_arch = "x86_64"),
+    target_env = "gnu"
+))]
+global_asm! {
+    r#"
+    .def _rust_eh_unwind_resume;
+    .scl 2;
+    .type 32;
+    .endef
+    .globl _rust_eh_unwind_resume
+    .p2align 4, 0x90
+_rust_eh_unwind_resume:
+    .cfi_startproc
+    jmp __Unwind_Resume
+    .cfi_endproc
+    "#
+}
+#[cfg(all(
+    bootstrap,
     target_os = "windows",
     any(target_arch = "x86", target_arch = "x86_64"),
     target_env = "gnu"
@@ -341,6 +357,7 @@ unsafe extern "C" fn rust_eh_unwind_resume(panic_ctx: *mut u8) -> ! {
     uw::_Unwind_Resume(panic_ctx as *mut uw::_Unwind_Exception);
 }
 
+
 // Frame unwind info registration
 //
 // Each module's image contains a frame unwind info section (usually

Diff for: src/libpanic_unwind/hermit.rs

@@ -6,10 +6,6 @@ use alloc::boxed::Box;
 use core::any::Any;
 use core::ptr;
 
-pub fn payload() -> *mut u8 {
-    ptr::null_mut()
-}
-
 pub unsafe fn cleanup(_ptr: *mut u8) -> Box<dyn Any + Send> {
     extern "C" {
         pub fn __rust_abort() -> !;

Diff for: src/libpanic_unwind/lib.rs

@@ -22,20 +22,24 @@
 #![feature(libc)]
 #![feature(nll)]
 #![feature(panic_unwind)]
-#![feature(raw)]
 #![feature(staged_api)]
 #![feature(std_internals)]
 #![feature(unwind_attributes)]
 #![feature(abi_thiscall)]
+#![feature(rustc_attrs)]
+#![feature(raw)]
+#![feature(asm)]
+#![feature(naked_functions)]
 #![panic_runtime]
 #![feature(panic_runtime)]
+#![feature(global_asm)]
 
 use alloc::boxed::Box;
-use core::intrinsics;
-use core::mem;
+use core::any::Any;
 use core::panic::BoxMeUp;
-use core::raw;
 
+// If adding to this list, you should also look at the list of TryPayload types
+// defined in payload.rs and likely add to there as well.
 cfg_if::cfg_if! {
     if #[cfg(target_os = "emscripten")] {
         #[path = "emcc.rs"]
@@ -61,6 +65,8 @@ cfg_if::cfg_if! {
     }
 }
 
+include!("payload.rs");
+
 extern "C" {
     /// Handler in libstd called when a panic object is dropped outside of
     /// `catch_unwind`.
@@ -69,28 +75,11 @@ extern "C" {
 
 mod dwarf;
 
-// Entry point for catching an exception, implemented using the `try` intrinsic
-// in the compiler.
-//
-// The interaction between the `payload` function and the compiler is pretty
-// hairy and tightly coupled, for more information see the compiler's
-// implementation of this.
 #[no_mangle]
-pub unsafe extern "C" fn __rust_maybe_catch_panic(
-    f: fn(*mut u8),
-    data: *mut u8,
-    data_ptr: *mut usize,
-    vtable_ptr: *mut usize,
-) -> u32 {
-    let mut payload = imp::payload();
-    if intrinsics::r#try(f, data, &mut payload as *mut _ as *mut _) == 0 {
-        0
-    } else {
-        let obj = mem::transmute::<_, raw::TraitObject>(imp::cleanup(payload));
-        *data_ptr = obj.data as usize;
-        *vtable_ptr = obj.vtable as usize;
-        1
-    }
+pub unsafe extern "C" fn __rust_panic_cleanup(
+    payload: TryPayload,
+) -> *mut (dyn Any + Send + 'static) {
+    Box::into_raw(imp::cleanup(payload))
 }
 
 // Entry point for raising an exception, just delegates to the platform-specific

Diff for: src/libpanic_unwind/payload.rs

@@ -0,0 +1,21 @@
+// Type definition for the payload argument of the try intrinsic.
+//
+// This must be kept in sync with the implementations of the try intrinsic.
+//
+// This file is included by both panic runtimes and libstd. It is part of the
+// panic runtime ABI.
+cfg_if::cfg_if! {
+    if #[cfg(target_os = "emscripten")] {
+        type TryPayload = *mut u8;
+    } else if #[cfg(target_arch = "wasm32")] {
+        type TryPayload = *mut u8;
+    } else if #[cfg(target_os = "hermit")] {
+        type TryPayload = *mut u8;
+    } else if #[cfg(all(target_env = "msvc", target_arch = "aarch64"))] {
+        type TryPayload = *mut u8;
+    } else if #[cfg(target_env = "msvc")] {
+        type TryPayload = [u64; 2];
+    } else {
+        type TryPayload = *mut u8;
+    }
+}