[LLVMdev] Wrong float value stored in LLVM IR code

[Adrian Ortega]

Originally was the latter, but after carefully debugging it turned out 
the former. In summary, I was storing the floating point value in a 
string and using ConstantFP::get() to get a
float value in two different lines in my code. The problem was that in 
one line for debugging reasons I was converting the string value to a 
C++ float type, and then again back to string thus I lost some precision 
there and a few digits were lost, and two different constants were 
generated (I know, I looked at the code and seems I needed a cup of 
coffee... :) ). Solution was to call ConstantFP::get() with the same 
string in both places with no castings to other type.

However, the information you provided was really helpful as I looked 
over my code and detected some bugs related to floats precision I was 
generating related to the explanation below (aside that silly bug from 
above).

Thanks everyone for your help.

On 18/06/14 07:29, Daniel Sanders wrote:
> Is the problem that the store and the fcmp have different single-precision constants to eachother, or is it that neither constant is what you expected?
>
> If it's the latter, then the constants look correct to me. Hexadecimal floating point values in LLVM-IR are in double precision format for both single and double precision values. When used for a single-precision constant, the double-precision constant must be convertible to single-precision without rounding. See the paragraph starting 'The one non-intuitive notation ...' at http://llvm.org/docs/LangRef.html#simple-constants.
>
> I'm not sure why the store and fcmp constants differ though. 3.141592 isn't exactly representable in single precision but I would have expected both constants to be rounded the same way.
>
>> -----Original Message-----
>> From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu]
>> On Behalf Of Adrian Ortega
>> Sent: 18 June 2014 04:59
>> To: llvmdev at cs.uiuc.edu
>> Subject: [LLVMdev] Wrong float value stored in LLVM IR code
>>
>> Hi everyone,
>>
>> I'm learning how to use LLVM API and JIT engine and I've come across with an
>> issue I haven't been able to figure out.
>>
>> The problem I'm having is that the wrong float is being stored  in a float global
>> variable.
>>
>> The code snippet I use to generate the float value is as follow:
>>
>>       llvm::Type* type = // initialize with the global variable type;
>>       std::string real_value = // initialized with the float value from std input.
>>       ...
>>       return ConstantFP::get(type,real_value);
>>
>> I am using the code snippet above to store a float and double value, the
>> double value works, the float value not. This can be seen with the following
>> LLVM IR I am generating:
>>
>> What the functions do is just:
>>    1. backup current value of global variable
>>    2. store new value
>>    3. call a C function which returns the value of the global variable in that same
>> module
>>    4. Compare against the value used in step 2
>>    5. restore the initial value
>>    6. return true or false if the values were equal.
>>
>> /**
>>     float gfloat = 3.141592;
>> */
>> define i32 @test_getgfloat_0() {
>> block_test_getgfloat_0:
>>     %0 = load float* @gfloat
>>     store float 0x400921FB00000000, float* @gfloat // wrong value stored
>>     %1 = call float @getgfloat()
>>     %2 = fcmp oeq float %1, 0x400921FA00000000 // wrong value stored
>>     %3 = zext i1 %2 to i32
>>     store float %0, float* @gfloat
>>     ret i32 %3
>> }
>>
>> /**
>>     double gdouble = 2.52340;
>>    */
>> define i32 @test_getdouble_0() {
>> block_test_getdouble_0:
>>     %0 = load double* @gdouble
>>     store double 2.523400e+00, double* @gdouble // correct value stored
>>     %1 = call double @getdouble()
>>     %2 = fcmp oeq double %1, 2.523400e+00 // correct value stored
>>     %3 = zext i1 %2 to i32
>>     store double %0, double* @gdouble
>>     ret i32 %3
>> }
>>
>> Do you have any idea of why I'm having this problem?
>>
>> Thank you.
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140618/2cb10963/attachment.html>