[PATCH] D12781: PGO IR-level instrumentation infrastructure

[Rong Xu via llvm-commits]

xur marked 17 inline comments as done.
xur added a comment.

Thanks for Sean's suggestion. They indeed make the tests cleaner and more robust. I integrated his reviews in the latest patch that I'll post soon.


================
Comment at: lib/Transforms/IPO/LLVMBuild.txt:23
@@ -22,2 +22,2 @@
 library_name = ipo
-required_libraries = Analysis Core InstCombine IRReader Linker Object ProfileData Scalar Support TransformUtils Vectorize
+required_libraries = Analysis Core InstCombine IRReader Linker Object ProfileData Scalar Support TransformUtils Vectorize Instrumentation
----------------
silvas wrote:
> Why does IPO now depend on Instrumentation, but didn't previously?
> What is different about the PGO passes vs the other instrumentation passes?
thanks for catching this. This should not be there -- it is only needed for the passmangerbuilder change.

================
Comment at: test/Transforms/PGOProfile/criticaledge_use.ll:25
@@ +24,3 @@
+sw.bb:
+  %call = call i32 @_ZL3bari(i32 2)
+  br label %sw.epilog
----------------
silvas wrote:
> This test case can be simplified further. At the very least, the function `bar` is not needed.
bar is a file static function and was used to check the source name is part of the profile variable name.


http://reviews.llvm.org/D12781