[llvm-commits] New classes for PR1255: Should enhance LLVM switch instruction to take case ranges.

[Stepan Dyatkovskiy]

Hi Chris. I commited new classes in r155464. I want to realize points summarized here:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20120305/138780.html

And more generally I propose to start very-very carefully (watch for buildbots, take in account all remarks and so on) go within this way:

0. Implement new classes. Classes doesn't affect anything. They still work with ConstantInt base values at this stage.
1. Fictitious replacement of current ConstantInt case values with ConstantRangesSet. Case ranges set will still hold single value, and ConstantInt *getCaseValue() will return it. But additionally implement new method in SwitchInst that allows to work with case ranges. Currenly I think it should be some wrapper that returns either single value or ConstantRangesSet object.
2. Step-by-step replacement of old "ConstantInt* getCaseValue()" with new alternative. Modify algorithms for all passes that works with SwitchInst. But don't modify LLParser and BitcodeReader/Writer. Still hold single value in each ConstantRangesSet object. On this stage some parts of LLVM will use old-style methods, and some ones new-style.
3. After all getCaseValue() usages will removed and whole LLVM and its clients will work in new style - modify LLParser, Reader and Writer. Remove getCaseValue().
4. Replace ConstantInt*-based case ranges set items with APInt ones.

Can you give me your approval for that?
Thanks.

-Stepan.

26.03.2012, 18:26, "Stepan Dyatkovskiy" <STPWORLD at narod.ru>:
> ping
>
> 23.03.2012, 16:40, "Stepan Dyatkovskiy" <stpworld at narod.ru>:
>
>>  Hi Chris.
>>
>>  // Some range class it may be ConstantRange as Duncan proposed.
>>  class Range {
>>     APInt Low;
>>     APInt High;
>>
>>     public:
>>
>>     // Some interface here (getLow, getHigh, <, >, isInRange and so on).
>>  }
>>
>>  class ConstantRangesSet {
>>     std::vector<Range> Ranges;
>>
>>     /// Checks is the given constant satisfies this case.
>>     bool isSatisfies(const APInt& N);
>>
>>     /// Returns number of ranges.
>>     unsigned getNumItems() const;
>>
>>     /// Returns set item with given index.
>>     const Range& getItem(unsigned idx);
>>  }
>>
>>  SingleNumber is APInt instead.
>>
>>  -Stepan.
>>
>>  Chris Lattner wrote:
>>>   On Mar 11, 2012, at 1:00 AM, Stepan Dyatkovskiy wrote:
>>>>   Hi Chris.
>>>>
>>>>   Summary of previous discussion:
>>>>   1. Finally ConstantInt should be replaced with APInt.
>>>>   2. How to store case ranges. I remind your last proposal:
>>>>
>>>>   [quote]
>>>>
>>>>   The mapping of the range to entries in CaseValues could look like this:
>>>>
>>>>   [N, N+1] ->  Single element in CaseValues indicates the case value for the destination.
>>>>   [MAXINT, MININT] ->  The entire case is "default"
>>>>   Otherwise, the pair indicates a series of ranges (inclusive).  If one of the ranges is MAXINT/MININT, then it is the default case.  For example [0, 4, 6,6]  would handle 0,1,2,3,4,6.  [4,6, MAXINT,MININT] would handle 4,5,6 and default.
>>>>
>>>>   [/quote]
>>>>
>>>>   My proposal was to create ConstantRangesSet that will store all numbers and ranges for complex case. I still propose to use this schema.
>>>>   You can ask what about most probable case, when for most of successors will be the only case value?
>>>>   OK. We can create collection for "classic" cases  vector<pair<SingleNumber,Success>  >  too. Then finally on the top level we have 2 collections:
>>>>   class SwitchInst {
>>>>   ..
>>>>     vector<pair<SingleNumber, BasicBlock*>  >  ClassicCases;
>>>>     vector<pair<ConstantRangesSet, BasicBlock>  >  ComplexCases;
>>>>   ..
>>>>   }
>>>>
>>>>   So, what do you think about this?
>>>   This approach seems sensible to me.  Can you sketch out what SingleNumber and ConstantRangesSet look like?
>>>
>>>   -Chris
>>  _______________________________________________
>>  llvm-commits mailing list
>>  llvm-commits at cs.uiuc.edu
>>  http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits