[PATCH] D57050: [CodeGenPrepare] Handle all debug calls in dupRetToEnableTailCallOpts()

Wed Jan 23 00:57:39 PST 2019

jonpa added inline comments.

================
Comment at: test/CodeGen/SystemZ/debuginstr-cgp.mir:54
+    %call59.pn = phi %"class.xercesc_2_5::ModifierToken.18.41.64.87.110.133.156.179.248.455"* [ %call59, %if.then55 ], [ undef, %if.then69 ]
+    %tok.0 = getelementptr inbounds %"class.xercesc_2_5::ModifierToken.18.41.64.87.110.133.156.179.248.455", %"class.xercesc_2_5::ModifierToken.18.41.64.87.110.133.156.179.248.455"* %call59.pn, i64 0, i32 0, !dbg !1564
+    call void @llvm.dbg.value(metadata %"class.xercesc_2_5::Token.4.27.50.73.96.119.142.165.234.441"* %tok.0, metadata !1554, metadata !DIExpression()), !dbg !1565
----------------
wolfgangp wrote:
> I'm a bit confused about this. The change was about dbg.value instructions following bitcasts, but here we have the dbg.value following a getelementptr. Perhaps I'm missing something?
Ah.. that's what Bugpoint did. It seems that CodeGenPrepare replaced that GEP (with only 0 indexes) with a bitcast before reaching the point of interest. I replaced it with a Bitcast in the test case now for clarity.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57050/new/

https://reviews.llvm.org/D57050