1 IRgen optimization opportunities. 2 3 //===---------------------------------------------------------------------===// 4 5 The common pattern of 6 -- 7 short x; // or char, etc 8 (x == 10) 9 -- 10 generates an zext/sext of x which can easily be avoided. 11 12 //===---------------------------------------------------------------------===// 13 14 Bitfields accesses can be shifted to simplify masking and sign 15 extension. For example, if the bitfield width is 8 and it is 16 appropriately aligned then is is a lot shorter to just load the char 17 directly. 18 19 //===---------------------------------------------------------------------===// 20 21 It may be worth avoiding creation of alloca's for formal arguments 22 for the common situation where the argument is never written to or has 23 its address taken. The idea would be to begin generating code by using 24 the argument directly and if its address is taken or it is stored to 25 then generate the alloca and patch up the existing code. 26 27 In theory, the same optimization could be a win for block local 28 variables as long as the declaration dominates all statements in the 29 block. 30 31 NOTE: The main case we care about this for is for -O0 -g compile time 32 performance, and in that scenario we will need to emit the alloca 33 anyway currently to emit proper debug info. So this is blocked by 34 being able to emit debug information which refers to an LLVM 35 temporary, not an alloca. 36 37 //===---------------------------------------------------------------------===// 38 39 We should try and avoid generating basic blocks which only contain 40 jumps. At -O0, this penalizes us all the way from IRgen (malloc & 41 instruction overhead), all the way down through code generation and 42 assembly time. 43 44 On 176.gcc:expr.ll, it looks like over 12% of basic blocks are just 45 direct branches! 46 47 //===---------------------------------------------------------------------===// 48
