[llvm-bugs] [Bug 30502] New: AVX512: incorrect code generation for simple piece of code (KNL & CNL backends affected)

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=30502

            Bug ID: 30502
           Summary: AVX512: incorrect code generation for simple piece of
                    code (KNL & CNL backends affected)
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: wenzel.jakob at epfl.ch
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

I've run into issues where the current (trunk) backend for Xeon Phi generates
incorrect code. As with similar issues before, it's got something to do with
LLVM's use of mask registers.

The snippet is given below (I tried to make it as small as possible). It makes
no use of AVX512 features whatsoever.

Try it as follows:

$ clang++ -march=nocona -std=c++11 -stdlib=libc++ -O0 -fno-rtti -fno-exceptions
-o test_works
$ ./test_works
(no crash)

$ clang++ -march=knl -std=c++11 -stdlib=libc++ -O0 -fno-rtti -fno-exceptions -o
test_broken
$ ./test_broken
[2]  91952 segmentation fault (core dumped)   ./test_broken

Looking at the diff of the assembly files generated by the two different
backends, there is only one change which is not equivalent and thus introduces
the failure:

$ diff -u assembly-nocona.s  assembly-knl.s                              [INS]
--- assembly-nocona.s    2016-09-23 11:09:30.000000000 +0200
+++ assembly-knl.s    2016-09-23 11:09:30.000000000 +0200
@@ -66,7 +66,8 @@
     movq    %rax, -24(%rbp)         # 8-byte Spill
     callq    _ZN5ArrayC2Ev
     movb    $1, -1(%rbp)
-    testb    $1, -1(%rbp)
+    movb    -1(%rbp), %cl
+    kortestw    %cl, %cl
     jne    .LBB1_2
 # BB#1:
     movq    -16(%rbp), %rdi         # 8-byte Reload


=========================================
#include <memory>

/* Some dummy data structure */
struct Data { float f; };

/* Stores a 'Data' instance via unique_ptr */
struct Array {
    using Holder = std::unique_ptr<Data[]>;

    Array() : data(new Data[1]) { }
    Array(const Array &) : data(new Data[1]) { }

    static Array Make() { Array result; return result; }

    Array &operator=(Array &value) { data[0] = value.data[0]; return *this; }

    Holder data;
};

int main(int /* argc */, char * /* argv */[]) {
    Array a = Array::Make();

    /* This next line causes a crash (nullptr dereference in __memcpy) */
    Data result = a.data[0];

    /* Quench unused variable warning */
    (void) result;

    return 0;
}
===========================

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160923/089e0a75/attachment.html>