Hey Louis
I don't know arm well enough to comment on all the patch, but the results are very impressive.
You did change the compiler to handle i64 variants. Can you add tests for those too?
Thanks
Pete
Sent from my iPhone
> On May 2, 2014, at 5:43 PM, Louis Gerbarg <lgg at apple.com> wrote:
>
> The attached patch provides custom lowering of overflow intrinsics for the ARM architecture. This greatly improves the codeine when those intrinsics are present. For example, the following function:
>
> define i32 @sadd_overflow(i32 %a, i32 %b) #0 {
> %sadd = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
> %1 = extractvalue { i32, i1 } %sadd, 1
> %2 = zext i1 %1 to i32
> ret i32 %2
> }
>
> Currently compiles to this:
>
> _sadd_overflow: @ @sadd_overflow
> @ BB#0:
> cmp r1, #0
> mov r3, #0
> mov r2, #0
> mov r12, #0
> movge r3, #1
> cmp r0, #0
> add r0, r0, r1
> movge r2, #1
> cmp r2, r3
> mov r3, #0
> moveq r3, #1
> cmp r0, #0
> mov r0, #0
> movge r0, #1
> cmp r2, r0
> movne r12, #1
> and r0, r3, r12
> mov pc, lr
>
> with the patch it compiles to this:
>
> _sadd_overflow: @ @sadd_overflow
> @ BB#0:
> add r2, r0, r1
> mov r1, #1
> cmp r2, r0
> movvc r1, #0
> mov r0, r1
> mov pc, lr
>
> Louis
>
> <0001-Add-custome-lowering-for-some-add-sub-with-overflow-.patch>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits