[llvm-bugs] [Bug 34549] New: [miscompilation] set to 0 on __m128d ignored: garbage in high element on compare and movemask
via llvm-bugs
llvm-bugs at lists.llvm.org
Mon Sep 11 04:09:09 PDT 2017
https://bugs.llvm.org/show_bug.cgi?id=34549
Bug ID: 34549
Summary: [miscompilation] set to 0 on __m128d ignored: garbage
in high element on compare and movemask
Product: new-bugs
Version: 5.0
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: new bugs
Assignee: unassignedbugs at nondot.org
Reporter: kretz at kde.org
CC: llvm-bugs at lists.llvm.org
reduced testcase:
#include <x86intrin.h>
#include <iostream>
std::ostream &operator<<(std::ostream &s, __m128d v)
{
return s << '{' << v[0] << ", " << v[1] << '}';
}
int main()
{
for (double lo_ : {1., 1.}) {
for (double hi_ : {1., 1.}) {
for (std::size_t pos = 0; pos < 2; ++pos) {
__m128d lo = _mm_set1_pd(lo_);
__m128d hi = _mm_set1_pd(hi_);
if (0 != _mm_movemask_pd(hi < lo)) {
std::cerr << hi << ", lo: " << lo;
if (3 != _mm_movemask_pd(hi >= hi)) {
std::cerr << hi << ", lo: " << lo;
}
}
}
}
}
__m128d x = _mm_set1_pd(1.);
for (std::size_t i = 0; i < 2; ++i) {
asm("ror $64,%%rax" ::"m"(x));
x[i] = 0; // #1
}
asm("ror $64,%%rax" :"+m"(x));
if (3 != _mm_movemask_pd(x == _mm_setzero_pd())) {
std::cerr << "!!!FAILED!!!\n";
return 1;
}
return 0;
}
Compile with `clang++-5.0 -std=c++14 -O2 -msse2`. clang 4.0 does not fail.
The assignment at line #1 is skipped for i = 1. However, minimal changes to
unrelated code lead to the use of a `movsd` from memory, thus zeroing the upper
64 bits, as requested.
The above code is a reduced testcase from a unit test in
https://github.com/VcDevel/Vc.
--
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/20170911/dbabc690/attachment.html>
More information about the llvm-bugs
mailing list