librustc_mir: Propagate constants during copy propagation. #36639
Conversation
pcwalton
added
I-slow
| MirSource::Const(_) | MirSource::Static(..) | MirSource::Promoted(..) => return, | ||
| MirSource::Fn(function_node_id) => { | ||
| // Don't run on const functions, because constant qualification might reject the | ||
| // optimized IR. |
There was a problem hiding this comment.
Comment nit: not really a problem for const qualification (as that only looks at non-generic consts only, already handled by the previous match arm), but rather rustc_trans, which in a pre-miri world cannot evaluate MIR where mutation is used to construct ADT.
There was a problem hiding this comment.
@eddyb But when I tried to bootstrap without doing this then qualify_consts started emitting errors. So I think it is a problem with that pass, no?
There was a problem hiding this comment.
The previous match arm (MirSource::Const(_) specifically) prevents qualify_consts from freaking out. Everything else is also needed, but for a different reason (trans::mir::constant).
There was a problem hiding this comment.
Added more comments.
| } | ||
| } | ||
|
|
||
| impl<'tcx> MutVisitor<'tcx> for ConstantPropagationVisitor<'tcx> { |
There was a problem hiding this comment.
I wonder if this could be replaced with replace_all_defs_and_uses_with taking an Operand.
There was a problem hiding this comment.
At first I wanted to do that, but I realized it's not possible. That's because defs are not Operands if they appear on e.g. the LHS of an Assign.
There was a problem hiding this comment.
Right, but in that case, it'd be a bug! - or you could have a separate replace_all_uses_with.
There was a problem hiding this comment.
Seems like premature abstraction. What other passes will need that function? Note that it's mostly just useful for consts, because operands are either lvalues or consts.
| @@ -22,12 +22,12 @@ fn main() { | |||
| // START rustc.node4.PreTrans.after.mir | |||
| // bb0: { | |||
| // nop; // scope 0 at storage_ranges.rs:14:9: 14:10 | |||
| // var0 = const 0i32; // scope 0 at storage_ranges.rs:14:13: 14:14 | |||
| // nop; // scope 0 at storage_ranges.rs:14:13: 14:14 | |||
There was a problem hiding this comment.
This test should probably use the MIR before any optimization passes, just to check the way storage ranges are generated.
There was a problem hiding this comment.
Want to change this to test pre-optimization MIR? If not, r=me as-is.
pcwalton
force-pushed
the
const-prop
branch
2 times, most recently
from
bdff21a
to
cbf41b0
Compare
|
@bors: r=eddyb |
|
📌 Commit cbf41b0 has been approved by |
|
Wait, Travis failed, @bors r- |
|
Argh, those don't run on the Mac! |
pcwalton
force-pushed
the
const-prop
branch
from
cbf41b0
to
f9b73aa
Compare
|
@bors: r=eddyb I think restricting the pass to MIR opt level >= 1 will fix the debug info problems. |
|
📌 Commit f9b73aa has been approved by |
| // There must be exactly one use of the source used in a statement (not in a terminator). | ||
| let src_use_info = def_use_analysis.local_info(src_local); | ||
| let src_use_count = src_use_info.use_count(); | ||
| if src_use_count == 0 { |
There was a problem hiding this comment.
Don't you have a use of src by definition?
| debug!(" Can't copy-propagate local: no uses"); | ||
| return None | ||
| } | ||
| if src_use_count != 1 { |
|
error: this file contains an un-closed delimiter @bors r- |
pcwalton
force-pushed
the
const-prop
branch
from
f9b73aa
to
a1108d1
Compare
|
@bors: r=eddyb |
|
📌 Commit a1108d1 has been approved by |
|
⌛ Testing commit a1108d1 with merge 1cd17db... |
|
💔 Test failed - auto-win-gnu-32-opt-rustbuild |
pcwalton
force-pushed
the
const-prop
branch
from
a1108d1
to
d1366c3
Compare
|
@bors: r=eddyb |
|
📌 Commit d1366c3 has been approved by |
|
⌛ Testing commit d1366c3 with merge f1278ae... |
|
💔 Test failed - auto-win-gnu-32-opt-rustbuild |
This optimization kicks in a lot when bootstrapping the compiler.
pcwalton
force-pushed
the
const-prop
branch
from
d1366c3
to
79cb2db
Compare
|
@bors: r=eddyb |
|
📌 Commit 79cb2db has been approved by |
librustc_mir: Propagate constants during copy propagation. This optimization kicks in a lot when bootstrapping the compiler. r? @eddyb
librustc_mir: Propagate constants during copy propagation. This optimization kicks in a lot when bootstrapping the compiler. r? @eddyb
This optimization kicks in a lot when bootstrapping the compiler.
r? @eddyb