/
write.rs
950 lines (836 loc) · 34.1 KB
/
write.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
// Copyright 2013-2014 The Rust Project Developers. See the COPYRIGHT
// file at the top-level directory of this distribution and at
// http://rust-lang.org/COPYRIGHT.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
use back::lto;
use back::link::{get_cc_prog, remove};
use driver::driver::{CrateTranslation, ModuleTranslation, OutputFilenames};
use driver::config::NoDebugInfo;
use driver::session::Session;
use driver::config;
use llvm;
use llvm::{ModuleRef, TargetMachineRef, PassManagerRef};
use util::common::time;
use syntax::abi;
use syntax::codemap;
use syntax::diagnostic;
use syntax::diagnostic::{Emitter, Handler, Level, mk_handler};
use std::c_str::{ToCStr, CString};
use std::io::Command;
use std::io::fs;
use std::iter::Unfold;
use std::ptr;
use std::str;
use std::sync::{Arc, Mutex};
use std::task::TaskBuilder;
use libc::{c_uint, c_int};
#[deriving(Clone, PartialEq, PartialOrd, Ord, Eq)]
pub enum OutputType {
OutputTypeBitcode,
OutputTypeAssembly,
OutputTypeLlvmAssembly,
OutputTypeObject,
OutputTypeExe,
}
pub fn llvm_err(handler: &diagnostic::Handler, msg: String) -> ! {
unsafe {
let cstr = llvm::LLVMRustGetLastError();
if cstr == ptr::null() {
handler.fatal(msg.as_slice());
} else {
let err = CString::new(cstr, true);
let err = String::from_utf8_lossy(err.as_bytes());
handler.fatal(format!("{}: {}",
msg.as_slice(),
err.as_slice()).as_slice());
}
}
}
pub fn write_output_file(
handler: &diagnostic::Handler,
target: llvm::TargetMachineRef,
pm: llvm::PassManagerRef,
m: ModuleRef,
output: &Path,
file_type: llvm::FileType) {
unsafe {
output.with_c_str(|output| {
let result = llvm::LLVMRustWriteOutputFile(
target, pm, m, output, file_type);
if !result {
llvm_err(handler, "could not write output".to_string());
}
})
}
}
struct Diagnostic {
msg: String,
code: Option<String>,
lvl: Level,
}
// We use an Arc instead of just returning a list of diagnostics from the
// child task because we need to make sure that the messages are seen even
// if the child task fails (for example, when `fatal` is called).
#[deriving(Clone)]
struct SharedEmitter {
buffer: Arc<Mutex<Vec<Diagnostic>>>,
}
impl SharedEmitter {
fn new() -> SharedEmitter {
SharedEmitter {
buffer: Arc::new(Mutex::new(Vec::new())),
}
}
fn dump(&mut self, handler: &Handler) {
let mut buffer = self.buffer.lock();
for diag in buffer.iter() {
match diag.code {
Some(ref code) => {
handler.emit_with_code(None,
diag.msg.as_slice(),
code.as_slice(),
diag.lvl);
},
None => {
handler.emit(None,
diag.msg.as_slice(),
diag.lvl);
},
}
}
buffer.clear();
}
}
impl Emitter for SharedEmitter {
fn emit(&mut self, cmsp: Option<(&codemap::CodeMap, codemap::Span)>,
msg: &str, code: Option<&str>, lvl: Level) {
assert!(cmsp.is_none(), "SharedEmitter doesn't support spans");
self.buffer.lock().push(Diagnostic {
msg: msg.to_string(),
code: code.map(|s| s.to_string()),
lvl: lvl,
});
}
fn custom_emit(&mut self, _cm: &codemap::CodeMap,
_sp: diagnostic::RenderSpan, _msg: &str, _lvl: Level) {
fail!("SharedEmitter doesn't support custom_emit");
}
}
// On android, we by default compile for armv7 processors. This enables
// things like double word CAS instructions (rather than emulating them)
// which are *far* more efficient. This is obviously undesirable in some
// cases, so if any sort of target feature is specified we don't append v7
// to the feature list.
//
// On iOS only armv7 and newer are supported. So it is useful to
// get all hardware potential via VFP3 (hardware floating point)
// and NEON (SIMD) instructions supported by LLVM.
// Note that without those flags various linking errors might
// arise as some of intrinsics are converted into function calls
// and nobody provides implementations those functions
fn target_feature<'a>(sess: &'a Session) -> &'a str {
match sess.targ_cfg.os {
abi::OsAndroid => {
if "" == sess.opts.cg.target_feature.as_slice() {
"+v7"
} else {
sess.opts.cg.target_feature.as_slice()
}
},
abi::OsiOS if sess.targ_cfg.arch == abi::Arm => {
"+v7,+thumb2,+vfp3,+neon"
},
_ => sess.opts.cg.target_feature.as_slice()
}
}
fn get_llvm_opt_level(optimize: config::OptLevel) -> llvm::CodeGenOptLevel {
match optimize {
config::No => llvm::CodeGenLevelNone,
config::Less => llvm::CodeGenLevelLess,
config::Default => llvm::CodeGenLevelDefault,
config::Aggressive => llvm::CodeGenLevelAggressive,
}
}
fn create_target_machine(sess: &Session) -> TargetMachineRef {
let reloc_model = match sess.opts.cg.relocation_model.as_slice() {
"pic" => llvm::RelocPIC,
"static" => llvm::RelocStatic,
"default" => llvm::RelocDefault,
"dynamic-no-pic" => llvm::RelocDynamicNoPic,
_ => {
sess.err(format!("{} is not a valid relocation mode",
sess.opts
.cg
.relocation_model).as_slice());
sess.abort_if_errors();
unreachable!();
}
};
let opt_level = get_llvm_opt_level(sess.opts.optimize);
let use_softfp = sess.opts.cg.soft_float;
// FIXME: #11906: Omitting frame pointers breaks retrieving the value of a parameter.
// FIXME: #11954: mac64 unwinding may not work with fp elim
let no_fp_elim = (sess.opts.debuginfo != NoDebugInfo) ||
(sess.targ_cfg.os == abi::OsMacos &&
sess.targ_cfg.arch == abi::X86_64);
// OSX has -dead_strip, which doesn't rely on ffunction_sections
// FIXME(#13846) this should be enabled for windows
let ffunction_sections = sess.targ_cfg.os != abi::OsMacos &&
sess.targ_cfg.os != abi::OsWindows;
let fdata_sections = ffunction_sections;
let code_model = match sess.opts.cg.code_model.as_slice() {
"default" => llvm::CodeModelDefault,
"small" => llvm::CodeModelSmall,
"kernel" => llvm::CodeModelKernel,
"medium" => llvm::CodeModelMedium,
"large" => llvm::CodeModelLarge,
_ => {
sess.err(format!("{} is not a valid code model",
sess.opts
.cg
.code_model).as_slice());
sess.abort_if_errors();
unreachable!();
}
};
unsafe {
sess.targ_cfg
.target_strs
.target_triple
.as_slice()
.with_c_str(|t| {
sess.opts.cg.target_cpu.as_slice().with_c_str(|cpu| {
target_feature(sess).with_c_str(|features| {
llvm::LLVMRustCreateTargetMachine(
t, cpu, features,
code_model,
reloc_model,
opt_level,
true /* EnableSegstk */,
use_softfp,
no_fp_elim,
ffunction_sections,
fdata_sections,
)
})
})
})
}
}
/// Module-specific configuration for `optimize_and_codegen`.
#[deriving(Clone)]
struct ModuleConfig {
/// LLVM TargetMachine to use for codegen.
tm: TargetMachineRef,
/// Names of additional optimization passes to run.
passes: Vec<String>,
/// Some(level) to optimize at a certain level, or None to run
/// absolutely no optimizations (used for the metadata module).
opt_level: Option<llvm::CodeGenOptLevel>,
// Flags indicating which outputs to produce.
emit_no_opt_bc: bool,
emit_bc: bool,
emit_lto_bc: bool,
emit_ir: bool,
emit_asm: bool,
emit_obj: bool,
// Miscellaneous flags. These are mostly copied from command-line
// options.
no_verify: bool,
no_prepopulate_passes: bool,
no_builtins: bool,
time_passes: bool,
}
impl ModuleConfig {
fn new(tm: TargetMachineRef, passes: Vec<String>) -> ModuleConfig {
ModuleConfig {
tm: tm,
passes: passes,
opt_level: None,
emit_no_opt_bc: false,
emit_bc: false,
emit_lto_bc: false,
emit_ir: false,
emit_asm: false,
emit_obj: false,
no_verify: false,
no_prepopulate_passes: false,
no_builtins: false,
time_passes: false,
}
}
fn set_flags(&mut self, sess: &Session, trans: &CrateTranslation) {
self.no_verify = sess.no_verify();
self.no_prepopulate_passes = sess.opts.cg.no_prepopulate_passes;
self.no_builtins = trans.no_builtins;
self.time_passes = sess.time_passes();
}
}
/// Additional resources used by optimize_and_codegen (not module specific)
struct CodegenContext<'a> {
// Extra resources used for LTO: (sess, reachable). This will be `None`
// when running in a worker thread.
lto_ctxt: Option<(&'a Session, &'a [String])>,
// Handler to use for diagnostics produced during codegen.
handler: &'a Handler,
}
impl<'a> CodegenContext<'a> {
fn new(handler: &'a Handler) -> CodegenContext<'a> {
CodegenContext {
lto_ctxt: None,
handler: handler,
}
}
fn new_with_session(sess: &'a Session, reachable: &'a [String]) -> CodegenContext<'a> {
CodegenContext {
lto_ctxt: Some((sess, reachable)),
handler: sess.diagnostic().handler(),
}
}
}
// Unsafe due to LLVM calls.
unsafe fn optimize_and_codegen(cgcx: &CodegenContext,
mtrans: ModuleTranslation,
config: ModuleConfig,
name_extra: String,
output_names: OutputFilenames) {
let ModuleTranslation { llmod, llcx } = mtrans;
let tm = config.tm;
if config.emit_no_opt_bc {
let ext = format!("{}.no-opt.bc", name_extra);
output_names.with_extension(ext.as_slice()).with_c_str(|buf| {
llvm::LLVMWriteBitcodeToFile(llmod, buf);
})
}
match config.opt_level {
Some(opt_level) => {
// Create the two optimizing pass managers. These mirror what clang
// does, and are by populated by LLVM's default PassManagerBuilder.
// Each manager has a different set of passes, but they also share
// some common passes.
let fpm = llvm::LLVMCreateFunctionPassManagerForModule(llmod);
let mpm = llvm::LLVMCreatePassManager();
// If we're verifying or linting, add them to the function pass
// manager.
let addpass = |pass: &str| {
pass.as_slice().with_c_str(|s| llvm::LLVMRustAddPass(fpm, s))
};
if !config.no_verify { assert!(addpass("verify")); }
if !config.no_prepopulate_passes {
llvm::LLVMRustAddAnalysisPasses(tm, fpm, llmod);
llvm::LLVMRustAddAnalysisPasses(tm, mpm, llmod);
populate_llvm_passes(fpm, mpm, llmod, opt_level,
config.no_builtins);
}
for pass in config.passes.iter() {
pass.as_slice().with_c_str(|s| {
if !llvm::LLVMRustAddPass(mpm, s) {
cgcx.handler.warn(format!("unknown pass {}, ignoring",
*pass).as_slice());
}
})
}
// Finally, run the actual optimization passes
time(config.time_passes, "llvm function passes", (), |()|
llvm::LLVMRustRunFunctionPassManager(fpm, llmod));
time(config.time_passes, "llvm module passes", (), |()|
llvm::LLVMRunPassManager(mpm, llmod));
// Deallocate managers that we're now done with
llvm::LLVMDisposePassManager(fpm);
llvm::LLVMDisposePassManager(mpm);
match cgcx.lto_ctxt {
Some((sess, reachable)) if sess.lto() => {
time(sess.time_passes(), "all lto passes", (), |()|
lto::run(sess, llmod, tm, reachable));
if config.emit_lto_bc {
let name = format!("{}.lto.bc", name_extra);
output_names.with_extension(name.as_slice()).with_c_str(|buf| {
llvm::LLVMWriteBitcodeToFile(llmod, buf);
})
}
},
_ => {},
}
},
None => {},
}
// A codegen-specific pass manager is used to generate object
// files for an LLVM module.
//
// Apparently each of these pass managers is a one-shot kind of
// thing, so we create a new one for each type of output. The
// pass manager passed to the closure should be ensured to not
// escape the closure itself, and the manager should only be
// used once.
unsafe fn with_codegen(tm: TargetMachineRef, llmod: ModuleRef,
no_builtins: bool, f: |PassManagerRef|) {
let cpm = llvm::LLVMCreatePassManager();
llvm::LLVMRustAddAnalysisPasses(tm, cpm, llmod);
llvm::LLVMRustAddLibraryInfo(cpm, llmod, no_builtins);
f(cpm);
llvm::LLVMDisposePassManager(cpm);
}
if config.emit_bc {
let ext = format!("{}.bc", name_extra);
output_names.with_extension(ext.as_slice()).with_c_str(|buf| {
llvm::LLVMWriteBitcodeToFile(llmod, buf);
})
}
time(config.time_passes, "codegen passes", (), |()| {
if config.emit_ir {
let ext = format!("{}.ll", name_extra);
output_names.with_extension(ext.as_slice()).with_c_str(|output| {
with_codegen(tm, llmod, config.no_builtins, |cpm| {
llvm::LLVMRustPrintModule(cpm, llmod, output);
})
})
}
if config.emit_asm {
let path = output_names.with_extension(format!("{}.s", name_extra).as_slice());
with_codegen(tm, llmod, config.no_builtins, |cpm| {
write_output_file(cgcx.handler, tm, cpm, llmod, &path, llvm::AssemblyFile);
});
}
if config.emit_obj {
let path = output_names.with_extension(format!("{}.o", name_extra).as_slice());
with_codegen(tm, llmod, config.no_builtins, |cpm| {
write_output_file(cgcx.handler, tm, cpm, llmod, &path, llvm::ObjectFile);
});
}
});
llvm::LLVMDisposeModule(llmod);
llvm::LLVMContextDispose(llcx);
llvm::LLVMRustDisposeTargetMachine(tm);
}
pub fn run_passes(sess: &Session,
trans: &CrateTranslation,
output_types: &[OutputType],
crate_output: &OutputFilenames) {
// It's possible that we have `codegen_units > 1` but only one item in
// `trans.modules`. We could theoretically proceed and do LTO in that
// case, but it would be confusing to have the validity of
// `-Z lto -C codegen-units=2` depend on details of the crate being
// compiled, so we complain regardless.
if sess.lto() && sess.opts.cg.codegen_units > 1 {
// This case is impossible to handle because LTO expects to be able
// to combine the entire crate and all its dependencies into a
// single compilation unit, but each codegen unit is in a separate
// LLVM context, so they can't easily be combined.
sess.fatal("can't perform LTO when using multiple codegen units");
}
unsafe {
configure_llvm(sess);
}
let tm = create_target_machine(sess);
// Figure out what we actually need to build.
let mut modules_config = ModuleConfig::new(tm, sess.opts.cg.passes.clone());
let mut metadata_config = ModuleConfig::new(tm, vec!());
modules_config.opt_level = Some(get_llvm_opt_level(sess.opts.optimize));
// Save all versions of the bytecode if we're saving our temporaries.
if sess.opts.cg.save_temps {
modules_config.emit_no_opt_bc = true;
modules_config.emit_bc = true;
modules_config.emit_lto_bc = true;
metadata_config.emit_bc = true;
}
// Emit a bitcode file for the crate if we're emitting an rlib.
// Whenever an rlib is created, the bitcode is inserted into the
// archive in order to allow LTO against it.
let needs_crate_bitcode =
sess.crate_types.borrow().contains(&config::CrateTypeRlib) &&
sess.opts.output_types.contains(&OutputTypeExe) &&
sess.opts.cg.codegen_units == 1;
if needs_crate_bitcode {
modules_config.emit_bc = true;
}
for output_type in output_types.iter() {
match *output_type {
OutputTypeBitcode => { modules_config.emit_bc = true; },
OutputTypeLlvmAssembly => { modules_config.emit_ir = true; },
OutputTypeAssembly => {
modules_config.emit_asm = true;
// If we're not using the LLVM assembler, this function
// could be invoked specially with output_type_assembly, so
// in this case we still want the metadata object file.
if !sess.opts.output_types.contains(&OutputTypeAssembly) {
metadata_config.emit_obj = true;
}
},
OutputTypeObject => { modules_config.emit_obj = true; },
OutputTypeExe => {
modules_config.emit_obj = true;
metadata_config.emit_obj = true;
},
}
}
modules_config.set_flags(sess, trans);
metadata_config.set_flags(sess, trans);
// Populate a buffer with a list of codegen tasks. Items are processed in
// LIFO order, just because it's a tiny bit simpler that way. (The order
// doesn't actually matter.)
let mut work_items = Vec::with_capacity(1 + trans.modules.len());
{
let work = build_work_item(sess,
trans.metadata_module,
metadata_config.clone(),
crate_output.clone(),
"metadata".to_string());
work_items.push(work);
}
for (index, mtrans) in trans.modules.iter().enumerate() {
let work = build_work_item(sess,
*mtrans,
modules_config.clone(),
crate_output.clone(),
format!("{}", index));
work_items.push(work);
}
// Process the work items, optionally using worker threads.
if sess.opts.cg.codegen_units == 1 {
run_work_singlethreaded(sess, trans.reachable.as_slice(), work_items);
if needs_crate_bitcode {
// The only bitcode file produced (aside from metadata) was
// "crate.0.bc". Rename to "crate.bc" since that's what
// `link_rlib` expects to find.
fs::copy(&crate_output.with_extension("0.bc"),
&crate_output.temp_path(OutputTypeBitcode)).unwrap();
}
} else {
run_work_multithreaded(sess, work_items, sess.opts.cg.codegen_units);
assert!(!needs_crate_bitcode,
"can't produce a crate bitcode file from multiple compilation units");
}
// All codegen is finished.
unsafe {
llvm::LLVMRustDisposeTargetMachine(tm);
}
// Produce final compile outputs.
let copy_if_one_unit = |ext: &str, output_type: OutputType| {
// Three cases:
if sess.opts.cg.codegen_units == 1 {
// 1) Only one codegen unit. In this case it's no difficulty
// to copy `foo.0.x` to `foo.x`.
fs::copy(&crate_output.with_extension(ext),
&crate_output.path(output_type)).unwrap();
if !sess.opts.cg.save_temps {
// The user just wants `foo.x`, not `foo.0.x`.
remove(sess, &crate_output.with_extension(ext));
}
} else {
if crate_output.single_output_file.is_some() {
// 2) Multiple codegen units, with `-o some_name`. We have
// no good solution for this case, so warn the user.
sess.warn(format!("ignoring -o because multiple .{} files were produced",
ext).as_slice());
} else {
// 3) Multiple codegen units, but no `-o some_name`. We
// just leave the `foo.0.x` files in place.
// (We don't have to do any work in this case.)
}
}
};
let link_obj = |output_path: &Path| {
// Some builds of MinGW GCC will pass --force-exe-suffix to ld, which
// will automatically add a .exe extension if the extension is not
// already .exe or .dll. To ensure consistent behavior on Windows, we
// add the .exe suffix explicitly and then rename the output file to
// the desired path. This will give the correct behavior whether or
// not GCC adds --force-exe-suffix.
let windows_output_path =
if sess.targ_cfg.os == abi::OsWindows {
Some(output_path.with_extension("o.exe"))
} else {
None
};
let pname = get_cc_prog(sess);
let mut cmd = Command::new(pname.as_slice());
cmd.args(sess.targ_cfg.target_strs.cc_args.as_slice());
cmd.arg("-nostdlib");
for index in range(0, trans.modules.len()) {
cmd.arg(crate_output.with_extension(format!("{}.o", index).as_slice()));
}
cmd.arg("-r")
.arg("-o")
.arg(windows_output_path.as_ref().unwrap_or(output_path));
if (sess.opts.debugging_opts & config::PRINT_LINK_ARGS) != 0 {
println!("{}", &cmd);
}
cmd.stdin(::std::io::process::Ignored)
.stdout(::std::io::process::InheritFd(1))
.stderr(::std::io::process::InheritFd(2));
match cmd.status() {
Ok(_) => {},
Err(e) => {
sess.err(format!("could not exec the linker `{}`: {}",
pname,
e).as_slice());
sess.abort_if_errors();
},
}
match windows_output_path {
Some(ref windows_path) => {
fs::rename(windows_path, output_path).unwrap();
},
None => {
// The file is already named according to `output_path`.
}
}
};
// Flag to indicate whether the user explicitly requested bitcode.
// Otherwise, we produced it only as a temporary output, and will need
// to get rid of it.
// FIXME: Since we don't support LTO anyway, maybe we can avoid
// producing the temporary .0.bc's in the first place?
let mut save_bitcode = false;
for output_type in output_types.iter() {
match *output_type {
OutputTypeBitcode => {
save_bitcode = true;
copy_if_one_unit("0.bc", OutputTypeBitcode);
},
OutputTypeLlvmAssembly => { copy_if_one_unit("0.ll", OutputTypeLlvmAssembly); },
OutputTypeAssembly => { copy_if_one_unit("0.s", OutputTypeAssembly); },
OutputTypeObject => { link_obj(&crate_output.path(OutputTypeObject)); },
OutputTypeExe => {
// If OutputTypeObject is already in the list, then
// `crate.o` will be handled by the OutputTypeObject case.
// Otherwise, we need to create the temporary object so we
// can run the linker.
if !sess.opts.output_types.contains(&OutputTypeObject) {
link_obj(&crate_output.temp_path(OutputTypeObject));
}
},
}
}
let save_bitcode = save_bitcode;
// Clean up unwanted temporary files.
// We create the following files by default:
// - crate.0.bc
// - crate.0.o
// - crate.metadata.bc
// - crate.metadata.o
// - crate.o (linked from crate.##.o)
// - crate.bc (copied from crate.0.bc)
// We may create additional files if requested by the user (through
// `-C save-temps` or `--emit=` flags).
if !sess.opts.cg.save_temps {
// Remove the temporary .0.o objects. If the user didn't
// explicitly request bitcode (with --emit=bc), we must remove
// .0.bc as well. (We don't touch the crate.bc that may have been
// produced earlier.)
for i in range(0, trans.modules.len()) {
if modules_config.emit_obj {
let ext = format!("{}.o", i);
remove(sess, &crate_output.with_extension(ext.as_slice()));
}
if modules_config.emit_bc && !save_bitcode {
let ext = format!("{}.bc", i);
remove(sess, &crate_output.with_extension(ext.as_slice()));
}
}
if metadata_config.emit_bc && !save_bitcode {
remove(sess, &crate_output.with_extension("metadata.bc"));
}
}
// We leave the following files around by default:
// - crate.o
// - crate.metadata.o
// - crate.bc
// These are used in linking steps and will be cleaned up afterward.
// FIXME: time_llvm_passes support - does this use a global context or
// something?
//if sess.time_llvm_passes() { llvm::LLVMRustPrintPassTimings(); }
}
type WorkItem = proc(&CodegenContext):Send;
fn build_work_item(sess: &Session,
mtrans: ModuleTranslation,
config: ModuleConfig,
output_names: OutputFilenames,
name_extra: String) -> WorkItem {
let mut config = config;
config.tm = create_target_machine(sess);
proc(cgcx) unsafe {
optimize_and_codegen(cgcx, mtrans, config, name_extra, output_names);
}
}
fn run_work_singlethreaded(sess: &Session,
reachable: &[String],
work_items: Vec<WorkItem>) {
let cgcx = CodegenContext::new_with_session(sess, reachable);
let mut work_items = work_items;
// Since we're running single-threaded, we can pass the session to
// the proc, allowing `optimize_and_codegen` to perform LTO.
for work in Unfold::new((), |_| work_items.pop()) {
work(&cgcx);
}
}
fn run_work_multithreaded(sess: &Session,
work_items: Vec<WorkItem>,
num_workers: uint) {
// Run some workers to process the work items.
let work_items_arc = Arc::new(Mutex::new(work_items));
let mut diag_emitter = SharedEmitter::new();
let mut futures = Vec::with_capacity(num_workers);
for i in range(0, num_workers) {
let work_items_arc = work_items_arc.clone();
let diag_emitter = diag_emitter.clone();
let future = TaskBuilder::new().named(format!("codegen-{}", i)).try_future(proc() {
let diag_handler = mk_handler(box diag_emitter);
// Must construct cgcx inside the proc because it has non-Send
// fields.
let cgcx = CodegenContext::new(&diag_handler);
loop {
// Avoid holding the lock for the entire duration of the match.
let maybe_work = work_items_arc.lock().pop();
match maybe_work {
Some(work) => {
work(&cgcx);
// Make sure to fail the worker so the main thread can
// tell that there were errors.
cgcx.handler.abort_if_errors();
}
None => break,
}
}
});
futures.push(future);
}
let mut failed = false;
for future in futures.move_iter() {
match future.unwrap() {
Ok(()) => {},
Err(_) => {
failed = true;
},
}
// Display any new diagnostics.
diag_emitter.dump(sess.diagnostic().handler());
}
if failed {
sess.fatal("aborting due to worker thread failure");
}
}
pub fn run_assembler(sess: &Session, outputs: &OutputFilenames) {
let pname = get_cc_prog(sess);
let mut cmd = Command::new(pname.as_slice());
cmd.arg("-c").arg("-o").arg(outputs.path(OutputTypeObject))
.arg(outputs.temp_path(OutputTypeAssembly));
debug!("{}", &cmd);
match cmd.output() {
Ok(prog) => {
if !prog.status.success() {
sess.err(format!("linking with `{}` failed: {}",
pname,
prog.status).as_slice());
sess.note(format!("{}", &cmd).as_slice());
let mut note = prog.error.clone();
note.push_all(prog.output.as_slice());
sess.note(str::from_utf8(note.as_slice()).unwrap());
sess.abort_if_errors();
}
},
Err(e) => {
sess.err(format!("could not exec the linker `{}`: {}",
pname,
e).as_slice());
sess.abort_if_errors();
}
}
}
unsafe fn configure_llvm(sess: &Session) {
use std::sync::{Once, ONCE_INIT};
static mut INIT: Once = ONCE_INIT;
// Copy what clang does by turning on loop vectorization at O2 and
// slp vectorization at O3
let vectorize_loop = !sess.opts.cg.no_vectorize_loops &&
(sess.opts.optimize == config::Default ||
sess.opts.optimize == config::Aggressive);
let vectorize_slp = !sess.opts.cg.no_vectorize_slp &&
sess.opts.optimize == config::Aggressive;
let mut llvm_c_strs = Vec::new();
let mut llvm_args = Vec::new();
{
let add = |arg: &str| {
let s = arg.to_c_str();
llvm_args.push(s.as_ptr());
llvm_c_strs.push(s);
};
add("rustc"); // fake program name
if vectorize_loop { add("-vectorize-loops"); }
if vectorize_slp { add("-vectorize-slp"); }
if sess.time_llvm_passes() { add("-time-passes"); }
if sess.print_llvm_passes() { add("-debug-pass=Structure"); }
for arg in sess.opts.cg.llvm_args.iter() {
add((*arg).as_slice());
}
}
INIT.doit(|| {
llvm::LLVMInitializePasses();
// Only initialize the platforms supported by Rust here, because
// using --llvm-root will have multiple platforms that rustllvm
// doesn't actually link to and it's pointless to put target info
// into the registry that Rust cannot generate machine code for.
llvm::LLVMInitializeX86TargetInfo();
llvm::LLVMInitializeX86Target();
llvm::LLVMInitializeX86TargetMC();
llvm::LLVMInitializeX86AsmPrinter();
llvm::LLVMInitializeX86AsmParser();
llvm::LLVMInitializeARMTargetInfo();
llvm::LLVMInitializeARMTarget();
llvm::LLVMInitializeARMTargetMC();
llvm::LLVMInitializeARMAsmPrinter();
llvm::LLVMInitializeARMAsmParser();
llvm::LLVMInitializeMipsTargetInfo();
llvm::LLVMInitializeMipsTarget();
llvm::LLVMInitializeMipsTargetMC();
llvm::LLVMInitializeMipsAsmPrinter();
llvm::LLVMInitializeMipsAsmParser();
llvm::LLVMRustSetLLVMOptions(llvm_args.len() as c_int,
llvm_args.as_ptr());
});
}
unsafe fn populate_llvm_passes(fpm: llvm::PassManagerRef,
mpm: llvm::PassManagerRef,
llmod: ModuleRef,
opt: llvm::CodeGenOptLevel,
no_builtins: bool) {
// Create the PassManagerBuilder for LLVM. We configure it with
// reasonable defaults and prepare it to actually populate the pass
// manager.
let builder = llvm::LLVMPassManagerBuilderCreate();
match opt {
llvm::CodeGenLevelNone => {
// Don't add lifetime intrinsics at O0
llvm::LLVMRustAddAlwaysInlinePass(builder, false);
}
llvm::CodeGenLevelLess => {
llvm::LLVMRustAddAlwaysInlinePass(builder, true);
}
// numeric values copied from clang
llvm::CodeGenLevelDefault => {
llvm::LLVMPassManagerBuilderUseInlinerWithThreshold(builder,
225);
}
llvm::CodeGenLevelAggressive => {
llvm::LLVMPassManagerBuilderUseInlinerWithThreshold(builder,
275);
}
}
llvm::LLVMPassManagerBuilderSetOptLevel(builder, opt as c_uint);
llvm::LLVMRustAddBuilderLibraryInfo(builder, llmod, no_builtins);
// Use the builder to populate the function/module pass managers.
llvm::LLVMPassManagerBuilderPopulateFunctionPassManager(builder, fpm);
llvm::LLVMPassManagerBuilderPopulateModulePassManager(builder, mpm);
llvm::LLVMPassManagerBuilderDispose(builder);
match opt {
llvm::CodeGenLevelDefault | llvm::CodeGenLevelAggressive => {
"mergefunc".with_c_str(|s| llvm::LLVMRustAddPass(mpm, s));
}
_ => {}
};
}