assigned to @lattner

Reduced testcase:

int %foo(int %t, int %C) {
entry:
br label %cond_true

cond_true: ; preds = %cond_true, %entry
%t_addr.0.0 = phi int [ %t, %entry ], [ %tmp7, %cond_true ] ; [#uses=2]
%tmp7 = add int %t_addr.0.0, 1 ; [#uses=1]
%tmp = setgt int %C, 39 ; [#uses=1]
br bool %tmp, label %bb12, label %cond_true

bb12: ; preds = %cond_true
ret int %t_addr.0.0
}

_foo:
movl 8(%esp), %ecx
movl 4(%esp), %edx
LBB1_1: #cond_true
movl %edx, %eax
movl %eax, %edx
incl %edx
cmpl $40, %ecx
jl LBB1_1 #cond_true
LBB1_2: #bb12
ret

Actually, I was wrong above, eax isn't dead, it's used by the ret outside the loop.

This fix for this one case is pretty trivial (I'm testing it now). One question, should we modify it so that
the general case is also caught? Something like this:

LBB1_1: #cond_true
movl %edx, %eax
;; Only uses of %eax
movl %eax, %edx

Is transformed into:

LBB1_1: #cond_true
movl %edx, %eax
;; Only uses of %eax

?

-bw

Fixed. After thinking about it some more, doing a check for the "general" case would be at least O(N^2)
and hairy. If it's something we want to perform in the future, we should create a new report for it.

This is definitely an improvement, but not enough. The generated code now looks like this:

_foo:
movl 8(%esp), %ecx
movl 4(%esp), %edx
LBB1_1: #cond_true
movl %edx, %eax
incl %edx
cmpl $40, %ecx
jl LBB1_1 #cond_true
LBB1_2: #bb12
ret

This is clearly better than having two movs, but it would be better to have:

_foo:
movl 8(%esp), %ecx
movl 4(%esp), %eax
LBB1_1: #cond_true
incl %eax
cmpl $40, %ecx
jl LBB1_1 #cond_true
LBB1_2: #bb12
ret

I think that this is a coallescing problem, the PHI live-range isn't getting coallesced with the copy into
EAX at the end properly or something.

-Chris

Ah! Yes. I see what you mean. Okay, I'll delve more deeply into the coalescer.

No, I'm confused. My suggestion would produce a different result. However, it does seem better to solve
this at the interval level if possible. I'll take a quick look and if not, I'll close it.

Fixed. Testcase here: X86/2006-08-21-ExtraMovInst.ll

Patches here:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20060821/037017.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20060821/037018.html

-Chris