ARM32 vector lowering: fabs, scalarize remaining arithmetic operations.

pydir/crosstest.py

Index: pydir/crosstest.py
diff --git a/pydir/crosstest.py b/pydir/crosstest.py
index 20cc349c568e29fc2168e436e26befa5d2fe32cd..2fe0208c64f93196fc06e9ec50edf6ecacd0d85f 100755
--- a/pydir/crosstest.py
+++ b/pydir/crosstest.py
@@ -29,13 +29,10 @@ def main():
                  'arm32': targets.ARM32Target }
     arch_sz_flags = { 'x8632': [],
                       'x8664': [],
-                      # TODO(jvoung): remove skip-unimplemented when
-                      # implemented.
                       # For ARM, test a large stack offset as well, until we
                       # are more confident. +/- 4095 is the limit, so test

[Jim Stichnoth, 2016/02/17 20:59:51]

While you're at it, I would reword this comment a bit.

I think we're plenty confident by now, but we might as well retain this flag to
get better test coverage.  So maybe this:

# For ARM, test a large stack offset as well, to increase coverage on that code
path.  +/- 4096 is ...

[Eric Holk, 2016/02/17 22:31:56]

Done.

                       # somewhere near that boundary.
-                      'arm32': ['--skip-unimplemented',
-                                '--test-stack-extra', '4084']
+                      'arm32': ['--test-stack-extra', '4084']
     }
     arch_llc_flags_extra = {
         # Use sse2 instructions regardless of input -mattr