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