add LTO support

The CLI gains -flto and -fno-lto options to override the default.
However, the cool thing about this is that the defaults are great! In
general when you use build-exe in release mode, Zig will enable LTO if
it would work and it would help.

zig cc supports detecting and honoring the -flto and -fno-lto flags as
well. The linkWithLld functions are improved to all be the same with
regards to copying the artifact instead of trying to pass single objects
through LLD with -r. There is possibly a future improvement here as
well; see the respective TODOs.

stage1 is updated to support outputting LLVM bitcode instead of machine
code when lto is enabled. This allows LLVM to optimize across the Zig and
C/C++ code boundary.

closes #2845

Author: andrewrk

src/Compilation.zig

@@ -444,6 +444,7 @@ pub const InitOptions = struct {
     want_valgrind: ?bool = null,
     want_tsan: ?bool = null,
     want_compiler_rt: ?bool = null,
+    want_lto: ?bool = null,
     use_llvm: ?bool = null,
     use_lld: ?bool = null,
     use_clang: ?bool = null,
@@ -602,6 +603,12 @@ pub fn create(gpa: *Allocator, options: InitOptions) !*Compilation {
             if (ofmt == .c)
                 break :blk false;
 
+            if (options.want_lto) |lto| {
+                if (lto) {
+                    break :blk true;
+                }
+            }
+
             // Our linker can't handle objects or most advanced options yet.
             if (options.link_objects.len != 0 or
                 options.c_source_files.len != 0 or
@@ -647,6 +654,26 @@ pub fn create(gpa: *Allocator, options: InitOptions) !*Compilation {
             break :outer opts;
         } else .{};
 
+        const lto = blk: {
+            if (options.want_lto) |explicit| {
+                if (!use_lld)
+                    return error.LtoUnavailableWithoutLld;
+                break :blk explicit;
+            } else if (!use_lld) {
+                break :blk false;
+            } else if (options.c_source_files.len == 0) {
+                break :blk false;
+            } else if (darwin_options.system_linker_hack) {
+                break :blk false;
+            } else switch (options.output_mode) {
+                .Lib, .Obj => break :blk false,
+                .Exe => switch (options.optimize_mode) {
+                    .Debug => break :blk false,
+                    .ReleaseSafe, .ReleaseFast, .ReleaseSmall => break :blk true,
+                },
+            }
+        };
+
         const tsan = options.want_tsan orelse false;
 
         const link_libc = options.link_libc or target_util.osRequiresLibC(options.target) or tsan;
@@ -821,6 +848,7 @@ pub fn create(gpa: *Allocator, options: InitOptions) !*Compilation {
         cache.hash.add(ofmt);
         cache.hash.add(pic);
         cache.hash.add(pie);
+        cache.hash.add(lto);
         cache.hash.add(tsan);
         cache.hash.add(stack_check);
         cache.hash.add(red_zone);
@@ -1022,6 +1050,7 @@ pub fn create(gpa: *Allocator, options: InitOptions) !*Compilation {
             .libc_installation = libc_dirs.libc_installation,
             .pic = pic,
             .pie = pie,
+            .lto = lto,
             .valgrind = valgrind,
             .tsan = tsan,
             .stack_check = stack_check,
@@ -2233,6 +2262,9 @@ pub fn addCCArgs(
                 "-nostdinc",
                 "-fno-spell-checking",
             });
+            if (comp.bin_file.options.lto) {
+                try argv.append("-flto");
+            }
 
             // According to Rich Felker libc headers are supposed to go before C language headers.
             // However as noted by @dimenus, appending libc headers before c_headers breaks intrinsics
@@ -3255,6 +3287,7 @@ fn updateStage1Module(comp: *Compilation, main_progress_node: *std.Progress.Node
         .err_color = @enumToInt(comp.color),
         .pic = comp.bin_file.options.pic,
         .pie = comp.bin_file.options.pie,
+        .lto = comp.bin_file.options.lto,
         .link_libc = comp.bin_file.options.link_libc,
         .link_libcpp = comp.bin_file.options.link_libcpp,
         .strip = comp.bin_file.options.strip,
@@ -3415,6 +3448,10 @@ pub fn build_crt_file(
         .want_tsan = false,
         .want_pic = comp.bin_file.options.pic,
         .want_pie = comp.bin_file.options.pie,
+        .want_lto = switch (output_mode) {
+            .Lib => comp.bin_file.options.lto,
+            .Obj, .Exe => false,
+        },
         .emit_h = null,
         .strip = comp.compilerRtStrip(),
         .is_native_os = comp.bin_file.options.is_native_os,

src/clang_options_data.zig

@@ -2732,7 +2732,14 @@ flagpd1("fkeep-static-consts"),
 flagpd1("flat_namespace"),
 flagpd1("flax-vector-conversions"),
 flagpd1("flimit-debug-info"),
-flagpd1("flto"),
+.{
+    .name = "flto",
+    .syntax = .flag,
+    .zig_equivalent = .lto,
+    .pd1 = true,
+    .pd2 = false,
+    .psl = false,
+},
 flagpd1("flto-unit"),
 flagpd1("flto-visibility-public-std"),
 sepd1("fmacro-backtrace-limit"),
@@ -2942,7 +2949,14 @@ flagpd1("fno-jump-tables"),
 flagpd1("fno-keep-static-consts"),
 flagpd1("fno-lax-vector-conversions"),
 flagpd1("fno-limit-debug-info"),
-flagpd1("fno-lto"),
+.{
+    .name = "fno-lto",
+    .syntax = .flag,
+    .zig_equivalent = .no_lto,
+    .pd1 = true,
+    .pd2 = false,
+    .psl = false,
+},
 flagpd1("fno-lto-unit"),
 flagpd1("fno-math-builtin"),
 flagpd1("fno-math-errno"),
@@ -5638,7 +5652,14 @@ jspd1("Ttext"),
     .pd2 = true,
     .psl = false,
 },
-joinpd1("flto="),
+.{
+    .name = "flto=",
+    .syntax = .joined,
+    .zig_equivalent = .lto,
+    .pd1 = true,
+    .pd2 = false,
+    .psl = false,
+},
 joinpd1("gcoff"),
 joinpd1("mabi="),
 joinpd1("mabs="),

src/codegen/llvm/bindings.zig

@@ -243,7 +243,13 @@ pub const TargetMachine = opaque {
     extern fn LLVMDisposeTargetMachine(T: *const TargetMachine) void;
 
     pub const emitToFile = LLVMTargetMachineEmitToFile;
-    extern fn LLVMTargetMachineEmitToFile(*const TargetMachine, M: *const Module, Filename: [*:0]const u8, codegen: CodeGenFileType, ErrorMessage: *[*:0]const u8) LLVMBool;
+    extern fn LLVMTargetMachineEmitToFile(
+        *const TargetMachine,
+        M: *const Module,
+        Filename: [*:0]const u8,
+        codegen: CodeGenFileType,
+        ErrorMessage: *[*:0]const u8,
+    ) LLVMBool;
 };
 
 pub const CodeMode = extern enum {

src/link.zig

@@ -74,6 +74,7 @@ pub const Options = struct {
     is_native_abi: bool,
     pic: bool,
     pie: bool,
+    lto: bool,
     valgrind: bool,
     tsan: bool,
     stack_check: bool,

src/link/Coff.zig

@@ -945,6 +945,13 @@ fn linkWithLLD(self: *Coff, comp: *Compilation) !void {
         if (!self.base.options.strip) {
             try argv.append("-DEBUG");
         }
+        if (self.base.options.lto) {
+            switch (self.base.options.optimize_mode) {
+                .Debug => {},
+                .ReleaseSmall => try argv.append("-OPT:lldlto=2"),
+                .ReleaseFast, .ReleaseSafe => try argv.append("-OPT:lldlto=3"),
+            }
+        }
         if (self.base.options.output_mode == .Exe) {
             const stack_size = self.base.options.stack_size_override orelse 16777216;
             try argv.append(try allocPrint(arena, "-STACK:{d}", .{stack_size}));