forked from OSchip/llvm-project
18 lines
562 B
LLVM
18 lines
562 B
LLVM
; This entire chain of computation should be optimized away, but
|
|
; wasn't because the two multiplies were not detected as being identical.
|
|
;
|
|
; RUN: llvm-as < %s | opt -gcse -instcombine -dce | llvm-dis | not grep sub
|
|
|
|
implementation ; Functions:
|
|
|
|
uint "vnum_test4"(uint* %data) {
|
|
%idx1 = getelementptr uint* %data, uint 1
|
|
%idx2 = getelementptr uint* %data, uint 3
|
|
%reg1101 = load uint* %idx1
|
|
%reg1111 = load uint* %idx2
|
|
%reg109 = mul uint %reg1101, %reg1111
|
|
%reg108 = mul uint %reg1111, %reg1101
|
|
%reg121 = sub uint %reg108, %reg109
|
|
ret uint %reg121
|
|
}
|