[PATCH] Schedule discriminators pass.

[Diego Novillo]

On Tue, Feb 25, 2014 at 8:29 PM, Eric Christopher <echristo at gmail.com> wrote:
>
>   Comments inline.
>
>
> ================
> Comment at: lib/CodeGen/CGDebugInfo.cpp:2581
> @@ -2581,1 +2580,3 @@
> +                                getColumnNumber(CurLoc),
> +                                0);
>    llvm::MDNode *DN = D;
> ----------------
> This can go in before, it just means that nothing will deal with it yes? In addition, it would be good to test that nothing changes if you pass 0 here and don't actually run the pass.

Split into http://llvm-reviews.chandlerc.com/D2891

>
> ================
> Comment at: test/CodeGenCXX/linetable-eh.cpp:9
> @@ -8,3 +8,3 @@
>  // CHECK: unwind label %{{.*}}, !dbg ![[DBG1]]
> -// CHECK: store i64 %{{.*}}, i64* %current_address, align 8, !dbg ![[DBG1]]
> +// CHECK: store i64 %{{.*}}, i64* %current_address, align 8, !dbg ![[DBG4:.*]]
>  // CHECK-NEXT: call void @llvm.dbg.declare(metadata !{{{.*}}}, metadata ![[FOUND_IT:.*]]), !dbg ![[DBG2:.*]]
> ----------------
> This is with the pass running?

Yes. When the pass runs, a new lexical block is created in this test
because the code needs a new discriminator value.


Diego.