[LLVMbugs] [Bug 16063] New: Multiplying vector type by a floating point constant results in an invalid result

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=16063

            Bug ID: 16063
           Summary: Multiplying vector type by a floating point constant
                    results in an invalid result
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: normal
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: georgellvmb at oss3d.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Please check the code below. A float vector is multiplied by a double constant.
When the vector size is changed to 4, the code even fails to compile. However
for size 2 it compiles and gives the following wrong results.

>From my analysis of the code, a bitcast instruction is used in LLVM code
generator, which (I presume) simply reinterprets double constant instead of
converting it.

===== EXPECTED OUTPUT:
ORIGINAL: 1.00,1.00
MULTIPLIED: 0.50,0.50

===== ACTUAL OUTPUT:
ORIGINAL: 1.00,1.00
MULTIPLIED: 0.00,1.75

===== CODE:

#include <stdio.h>
#include <math.h>

typedef float tfloat __attribute__((ext_vector_type(2)))  
__attribute__((aligned(8))) ;

int main()
{
    tfloat orig=1.0;


    printf("ORIGINAL: %.2lf,%.2lf\n",(double)orig[0],(double)orig[1]);

    orig = orig*0.5;

    printf("MULTIPLIED: %.2lf,%.2lf\n",(double)orig[0],(double)orig[1]);

    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/20130518/d16aa239/attachment.html>