Problem recognizing nested select operations

Hello all,

I just noticed that LLVM has trouble recognizing nested simple select operations.

My testcase is as follows:
-------------------------- snip ------------------------------
#include <stdio.h>

int main(int argc, char **argv)
{
  int a;

  scanf("%d", &a);

  a = a > 255 ? 255 : a < 0 ? 0 : a;

  printf("A: %d\n", a);

  return 0;
}
-------------------------- snip ------------------------------

Compiling using 'clang -emit-llvm -O3 -S test.c -o -' results in the following:

-------------------------- snip ------------------------------
define i32 @main(i32 %argc, i8** nocapture %argv) nounwind {
entry:
   %a = alloca i32, align 4
   %call = call i32 (i8*, ...)* @__isoc99_scanf(i8* getelementptr inbounds ([3 x i8]* @.str, i32 0, i32 0), i32* %a) nounwind
   %0 = load i32* %a, align 4, !tbaa !0
   %cmp = icmp sgt i32 %0, 255
   br i1 %cmp, label %cond.end4, label %cond.false

cond.false: ; preds = %entry
   %cmp1 = icmp slt i32 %0, 0
   %cond = select i1 %cmp1, i32 0, i32 %0
   br label %cond.end4

cond.end4: ; preds = %cond.false, %entry
   %cond5 = phi i32 [ 255, %entry ], [ %cond, %cond.false ]
   store i32 %cond5, i32* %a, align 4, !tbaa !0
   %call6 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([7 x i8]* @.str1, i32 0, i32 0), i32 %cond5) nounwind
   ret i32 0
}
-------------------------- snip ------------------------------

But the following output is much more efficient in this case:
-------------------------- snip ------------------------------
define i32 @main(i32 %argc, i8** nocapture %argv) nounwind {
entry:
   %a = alloca i32, align 4
   %call = call i32 (i8*, ...)* @__isoc99_scanf(i8* getelementptr inbounds ([3 x i8]* @.str, i32 0, i32 0), i32* %a) nounwind
   %0 = load i32* %a, align 4, !tbaa !0
   %cmp = icmp sgt i32 %0, 255
   %cond = select i1 %cmp, i32 255, i32 %0
   %cmp1 = icmp slt i32 %cond, 0
   %cond5 = select i1 %cmp1, i32 0, i32 %cond
   store i32 %cond5, i32* %a, align 4, !tbaa !0
   %call6 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([7 x i8]* @.str1, i32 0, i32 0), i32 %cond5) nounwind
   ret i32 0
}
-------------------------- snip ------------------------------

Is this something I should expect LLVM to solve or is there a reason for lazy evaluation of the inner nested select? I suppose that lazy evaluation is in the C standard and may be expected but in this case it would be perfectly safe to speculatively evaluate the inner nested select and transform the phi statement into a second select... Is there a transformation available capable of such a speculative move?

Cheers,
  Roel

HiRoel

In SimplifyCFG.cpp, SpeculativelyExecuteBB will turn a BB with a single instruction into a select. Your case has the two instructions, icmp and select, so it skips that BB.

You could try change that code to allow your case but I have no idea if its a good idea performance wise on all the test suite.

Cheers,
Pete