[llvm-commits] [llvm] r113996 - in /llvm/trunk: include/llvm/MC/MCInst.h lib/Target/ARM/ARMMCInstLower.cpp lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp

[Jim Grosbach]

On Sep 15, 2010, at 2:47 PM, Stephen Canon wrote:

> On Sep 15, 2010, at 2:41 PM, Jim Grosbach wrote:
> 
>> On Sep 15, 2010, at 1:52 PM, Stephen Canon wrote:
>> 
>>> On Sep 15, 2010, at 1:05 PM, Dale Johannesen wrote:
>>> 
>>>> On Sep 15, 2010, at 11:47 AMPDT, Jim Grosbach wrote:
>>>> 
>>>>> Author: grosbach
>>>>> Date: Wed Sep 15 13:47:08 2010
>>>>> New Revision: 113996
>>>>> 
>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=113996&view=rev
>>>>> Log:
>>>>> Add support for floating point immediates to MC instruction printing. ARM
>>>>> VFP instructions use it for loading some constants, so implement that
>>>>> handling.
>>>>> 
>>>>> Not thrilled with adding a member to MCOperand, but not sure there's much of
>>>>> a better option that's not pretty fragile (like putting a double in the
>>>>> union instead and just assuming that's good enough). Suggestions welcome...
>>>> 
>>>> This doesn't seem to handle NaNs and infinities.  Also, I wonder whether you're guaranteed to get enough precision to represent the value accurately with << (printf does not guarantee this).  Will the assembler take a hex format?  That would be better.
>>> 
>>> My reading of the C++ standard is that << does not guarantee that the value will be printed exactly.  You need 7 digits after the decimal point in order to print all expressible immediates exactly, but you only get 6 digits.  A test on the immediate value 0x1.1p-3 seems to confirm this.
>> 
>> I get #1.328125e-01 for 0x1.1p-3, which seems right? Note that LLVM's output streamer is a raw_ostream, not standard C++ streams.
> 
> Yes, that's the correct value.  With "std::cout << 0x1.1p-3" I was seeing "0.132812".

OK, cool. LLVM uses the equivalent to the printf() style format string "%e" for floating point, which should, if I follow properly, handle the necessary value range. Standard C++ streams use the equivalent to "%f" IIRC.

Thanks for mentioning this. It's very good to explicitly check for this sort of thing. I'll add a testcase with the above value in it as a literal.

-Jim