|  | implementation | 
|  |  | 
|  | ; edgefailure - This function illustrates how SCCP is not doing it's job.  This | 
|  | ; function should be optimized almost completely away: the loop should be | 
|  | ; analyzed to detect that the body executes exactly once, and thus the branch | 
|  | ; can be eliminated and code becomes trivially dead.  This is distilled from a | 
|  | ; real benchmark (mst from Olden benchmark, MakeGraph function).  When SCCP is | 
|  | ; fixed, this should be eliminated by a single SCCP application.  TODO | 
|  | ; | 
|  | int *"edgefailure"() | 
|  | begin | 
|  | bb1: | 
|  | %A = malloc int | 
|  | br label %bb2 | 
|  | bb2: | 
|  | %i = phi int [ %i2, %bb2 ], [ 0, %bb1 ]   ;; Always 0 | 
|  | %i2 = add int %i, 1                       ;; Always 1 | 
|  | store int %i, int *%A | 
|  | %loop = setle int %i2, 0                  ;; Always false | 
|  | br bool %loop, label %bb2, label %bb3 | 
|  |  | 
|  | bb3: | 
|  | ret int * %A | 
|  | end | 
|  |  | 
|  |  | 
|  |  | 
|  | int "test function"(int %i0, int %j0) | 
|  | %i1 = const int 1 | 
|  | %j1 = const int 1 | 
|  | %k1 = const int 0 | 
|  | begin | 
|  | BB1: | 
|  | br label %BB2 | 
|  | BB2: | 
|  | %j2 = phi int [%j4, %BB7], [%j1, %BB1] | 
|  | %k2 = phi int [%k4, %BB7], [%k1, %BB1] | 
|  | %kcond = setlt int %k2, 100 | 
|  | br bool %kcond, label %BB3, label %BB4 | 
|  |  | 
|  | BB3: | 
|  | %jcond = setlt int %j2, 20 | 
|  | br bool %jcond, label %BB5, label %BB6 | 
|  |  | 
|  | BB4: | 
|  | ret int %j2 | 
|  |  | 
|  | BB5: | 
|  | %k3 = add int %k2, 1 | 
|  | br label %BB7 | 
|  |  | 
|  | BB6: | 
|  | %k5 = add int %k2, 1 | 
|  | br label %BB7 | 
|  |  | 
|  | BB7: | 
|  | %j4 = phi int [%i1, %BB5], [%k2, %BB6] | 
|  | %k4 = phi int [%k3, %BB5], [%k5, %BB6] | 
|  | br label %BB2 | 
|  | end | 
|  |  |