Suppress _dl_runtime_resolve_avx_slow for memcheck conditional.

glibc ld.so has an optimization when resolving a symbol that checks
whether or not the upper 128 bits of the ymm registers are zero. If
so it uses "cheaper" instructions to save/restore them using the xmm
registers. If those upper 128 bits contain undefined values memcheck
will issue an Conditional jump or move depends on uninitialised value(s)
warning whenever trying to resolve a symbol.

This triggers in our sh-mem-vecxxx test cases. Suppress the warning
by default.

https://bugs.kde.org/show_bug.cgi?id=385868

diff --git a/NEWS b/NEWS
index b043c580cd2677ebc1a0a63e8a521336e26bc5c6..50c0097d3e7a69bc089e69b8484b961404b385cc 100644 (file)
--- a/NEWS
+++ b/NEWS
@@ -65,6 +65,7 @@ n-i-bz  Fix missing workq_ops operations (macOS)
 385210  PPC64, vpermr instruction could exhaust temporary memory
 385183  PPC64, Add support for xscmpeqdp, xscmpgtdp, xscmpgedp, xsmincdp instructions
 385334  PPC64, fix vpermr, xxperm, xxpermr mask value.
+385868  glibc ld.so _dl_runtime_resolve_avx_slow conditional jump warning.
 
 Release 3.13.0 (15 June 2017)
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

diff --git a/glibc-2.X.supp.in b/glibc-2.X.supp.in
index 8edeb4a8acba0512c18c3647d941286f431a0da2..126e8b3381be5e34021e80c1aaa3bff948eb01be 100644 (file)
--- a/glibc-2.X.supp.in
+++ b/glibc-2.X.supp.in
@@ -236,3 +236,15 @@
    Memcheck:Cond
    fun:_dl_relocate_object
 }
+
+# glibc ld.so has an optimization when resolving a symbol that checks
+# whether or not the upper 128 bits of the ymm registers are zero. If
+# so it uses "cheaper" instructions to save/restore them using the xmm
+# registers. If those upper 128 bits contain undefined values memcheck
+# will issue an Conditional jump or move depends on uninitialised value(s)
+# warning whenever trying to resolve a symbol.
+{
+   dl-trampoline-sse-avx
+   Memcheck:Cond
+   fun:_dl_runtime_resolve_avx_slow
+}