Missing Optimization Opportunities

Project Infrastructure LLVM Dev List Archives
1

Hi,

I'm using LLVM 2.7 right now, and I found "opt -std-compile-opts" has
missed some opportunities for optimization:

define void @spa.main() readonly {
entry:
  %tmp = load i32* @dst-ip ; <i32> [#uses=3]
  %tmp1 = and i32 %tmp, -16777216 ; <i32> [#uses=1]
  %tmp2 = icmp eq i32 %tmp1, 167772160 ; <i1> [#uses=2]
  %tmp3 = and i32 %tmp, -65536 ; <i32> [#uses=2]
  %tmp4 = icmp ne i32 %tmp3, 168296448 ; <i1> [#uses=1]
  %tmp5 = and i1 %tmp2, %tmp4 ; <i1> [#uses=1]
  %tmp6 = and i32 %tmp, -256 ; <i32> [#uses=2]
  %tmp7 = icmp eq i32 %tmp6, 168296704 ; <i1> [#uses=1]
  %tmp8 = icmp eq i32 %tmp3, 168296448 ; <i1> [#uses=2]
  %tmp9 = icmp ne i32 %tmp6, 168296704 ; <i1> [#uses=1]
  %tmp10 = and i1 %tmp8, %tmp9 ; <i1> [#uses=1]
  %tmp11 = load i32* @src-ip ; <i32> [#uses=1]
  %tmp12 = and i32 %tmp11, -16777216 ; <i32> [#uses=1]
  %tmp13 = icmp eq i32 %tmp12, 721420288 ; <i1> [#uses=3]
  %tmp14 = and i1 %tmp2, %tmp5 ; <i1> [#uses=1]
  %tmp15 = and i1 %tmp13, %tmp14 ; <i1> [#uses=1]
  tail call void @spa.assert(i1 %tmp15)
  %tmp16 = and i1 %tmp8, %tmp10 ; <i1> [#uses=1]
  %tmp17 = and i1 %tmp13, %tmp16 ; <i1> [#uses=1]
  tail call void @spa.assert(i1 %tmp17)
  %tmp18 = and i1 %tmp13, %tmp7 ; <i1> [#uses=1]
  tail call void @spa.assert(i1 %tmp18)
  ret void
}

Please notice the following code sequences are not optimal:

%tmp5 = and i1 %tmp2, %tmp4 ; <i1> [#uses=1]
%tmp14 = and i1 %tmp2, %tmp5 ; <i1> [#uses=1]

Is it a problem of scalar evolution or I just need to call another pass?

Thanks.

Haohui

2

The reassociate pass used to do this but it was disabled for "i1" values because
it pessimized some code (r95333). I pushed a fix (r113651) to trunk which adds support
for this optimization to the InstructionSimplify analysis (which is run, among others,
by the Instcombine pass).

Thanks for bringing this up. :slight_smile: