[PATCH] D20685: SLPVectorizer to handle GEP with differing constant index types

Michael Zolotukhin via llvm-commits llvm-commits at lists.llvm.org
Fri May 27 14:25:43 PDT 2016


mzolotukhin added a comment.

Hi Jesper,

This test doesn't work to me. Could you please verify that this patch is for the current trunk?

Also, some more specific remarks inline.

Thanks,
Michael


================
Comment at: test/Transforms/SLPVectorizer/X86/gep_mismatch.ll:1
@@ +1,2 @@
+; RUN: opt < %s -S -mcpu=x86-64 -mtriple=x86_64-linux -indvars -slp-vectorizer -O1
+
----------------
We usually try to minimize number of passes running on a test. For instance, if the assertion happened in SLPVectorizer, we probably only need slp-vectorizer: `-indvars` and `-O1` can be removed (i.e. you run `opt -indvars` on the test and the output is the new test that doesn't need `-indvars` in the command line).

================
Comment at: test/Transforms/SLPVectorizer/X86/gep_mismatch.ll:6
@@ +5,3 @@
+
+target datalayout = "E-B16-p:16:16-p20:32:16-p21:32:16-p40:32:16-F40-i1:16-i8:16-i32:16-i64:16-f32:16-f64:16-v32:16:16-v64:16:16-v128:16:16-a:0:16-i24:16-i40:16-n16:32"
+target triple = "x86_64-unknown-linux-gnu"
----------------
I got `LLVM ERROR: Unknown specifier in datalayout string` on this. Could you please check if it's correct?

================
Comment at: test/Transforms/SLPVectorizer/X86/gep_mismatch.ll:39
@@ +38,2 @@
+
+declare void @TEST__MAIN() #3
----------------
`#3` means that we need attribute #3, but the test contains no attributes at all. Were they lost when you were reducing the test?


http://reviews.llvm.org/D20685





More information about the llvm-commits mailing list