| Dan Gohman | 1f96e67 | 2009-09-17 18:05:20 +0000 | [diff] [blame^] | 1 | ; RUN: opt < %s -S -analyze -scalar-evolution -disable-output | FileCheck %s |
| 2 | |
| 3 | ; ScalarEvolution should be able to fold away the sign-extensions |
| 4 | ; on this loop with a primary induction variable incremented with |
| 5 | ; a nsw add of 2. |
| 6 | |
| 7 | target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128" |
| 8 | |
| 9 | define void @foo(i32 %n, double* nocapture %d, double* nocapture %q) nounwind { |
| 10 | entry: |
| 11 | %0 = icmp sgt i32 %n, 0 ; <i1> [#uses=1] |
| 12 | br i1 %0, label %bb.nph, label %return |
| 13 | |
| 14 | bb.nph: ; preds = %entry |
| 15 | br label %bb |
| 16 | |
| 17 | bb: ; preds = %bb.nph, %bb1 |
| 18 | %i.01 = phi i32 [ %16, %bb1 ], [ 0, %bb.nph ] ; <i32> [#uses=5] |
| 19 | |
| 20 | ; CHECK: %1 = sext i32 %i.01 to i64 |
| 21 | ; CHECK: --> {0,+,2}<bb> |
| 22 | %1 = sext i32 %i.01 to i64 ; <i64> [#uses=1] |
| 23 | |
| 24 | ; CHECK: %2 = getelementptr inbounds double* %d, i64 %1 |
| 25 | ; CHECK: --> {%d,+,16}<bb> |
| 26 | %2 = getelementptr inbounds double* %d, i64 %1 ; <double*> [#uses=1] |
| 27 | |
| 28 | %3 = load double* %2, align 8 ; <double> [#uses=1] |
| 29 | %4 = sext i32 %i.01 to i64 ; <i64> [#uses=1] |
| 30 | %5 = getelementptr inbounds double* %q, i64 %4 ; <double*> [#uses=1] |
| 31 | %6 = load double* %5, align 8 ; <double> [#uses=1] |
| 32 | %7 = or i32 %i.01, 1 ; <i32> [#uses=1] |
| 33 | |
| 34 | ; CHECK: %8 = sext i32 %7 to i64 |
| 35 | ; CHECK: --> {1,+,2}<bb> |
| 36 | %8 = sext i32 %7 to i64 ; <i64> [#uses=1] |
| 37 | |
| 38 | ; CHECK: %9 = getelementptr inbounds double* %q, i64 %8 |
| 39 | ; CHECK: {(8 + %q),+,16}<bb> |
| 40 | %9 = getelementptr inbounds double* %q, i64 %8 ; <double*> [#uses=1] |
| 41 | |
| 42 | ; Artificially repeat the above three instructions, this time using |
| 43 | ; add nsw instead of or. |
| 44 | %t7 = add nsw i32 %i.01, 1 ; <i32> [#uses=1] |
| 45 | |
| 46 | ; CHECK: %t8 = sext i32 %t7 to i64 |
| 47 | ; CHECK: --> {1,+,2}<bb> |
| 48 | %t8 = sext i32 %t7 to i64 ; <i64> [#uses=1] |
| 49 | |
| 50 | ; CHECK: %t9 = getelementptr inbounds double* %q, i64 %t8 |
| 51 | ; CHECK: {(8 + %q),+,16}<bb> |
| 52 | %t9 = getelementptr inbounds double* %q, i64 %t8 ; <double*> [#uses=1] |
| 53 | |
| 54 | %10 = load double* %9, align 8 ; <double> [#uses=1] |
| 55 | %11 = fadd double %6, %10 ; <double> [#uses=1] |
| 56 | %12 = fadd double %11, 3.200000e+00 ; <double> [#uses=1] |
| 57 | %13 = fmul double %3, %12 ; <double> [#uses=1] |
| 58 | %14 = sext i32 %i.01 to i64 ; <i64> [#uses=1] |
| 59 | %15 = getelementptr inbounds double* %d, i64 %14 ; <double*> [#uses=1] |
| 60 | store double %13, double* %15, align 8 |
| 61 | %16 = add nsw i32 %i.01, 2 ; <i32> [#uses=2] |
| 62 | br label %bb1 |
| 63 | |
| 64 | bb1: ; preds = %bb |
| 65 | %17 = icmp slt i32 %16, %n ; <i1> [#uses=1] |
| 66 | br i1 %17, label %bb, label %bb1.return_crit_edge |
| 67 | |
| 68 | bb1.return_crit_edge: ; preds = %bb1 |
| 69 | br label %return |
| 70 | |
| 71 | return: ; preds = %bb1.return_crit_edge, %entry |
| 72 | ret void |
| 73 | } |
| 74 | |
| 75 | ; CHECK: Loop bb: backedge-taken count is ((-1 + %n) /u 2) |
| 76 | ; CHECK: Loop bb: max backedge-taken count is 1073741823 |