| Eric Christopher | cee313d | 2019-04-17 04:52:47 +0000 | [diff] [blame] | 1 | ; RUN: opt -loop-distribute -enable-loop-distribute -verify-loop-info -verify-dom-info -S < %s \ |
| 2 | ; RUN: | FileCheck %s |
| 3 | |
| 4 | ; Check that definitions used outside the loop are handled correctly: (1) they |
| 5 | ; are not dropped (2) when version the loop, a phi is added to merge the value |
| 6 | ; from the non-distributed loop and the distributed loop. |
| 7 | ; |
| 8 | ; for (i = 0; i < n; i++) { |
| 9 | ; A[i + 1] = A[i] * B[i]; |
| 10 | ; ========================== |
| 11 | ; sum += C[i]; |
| 12 | ; } |
| 13 | |
| 14 | target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" |
| 15 | target triple = "x86_64-apple-macosx10.10.0" |
| 16 | |
| 17 | @B = common global i32* null, align 8 |
| 18 | @A = common global i32* null, align 8 |
| 19 | @C = common global i32* null, align 8 |
| 20 | @D = common global i32* null, align 8 |
| 21 | @E = common global i32* null, align 8 |
| 22 | @SUM = common global i32 0, align 8 |
| 23 | |
| 24 | define void @f() { |
| 25 | entry: |
| 26 | %a = load i32*, i32** @A, align 8 |
| 27 | %b = load i32*, i32** @B, align 8 |
| 28 | %c = load i32*, i32** @C, align 8 |
| 29 | %d = load i32*, i32** @D, align 8 |
| 30 | %e = load i32*, i32** @E, align 8 |
| 31 | |
| 32 | br label %for.body |
| 33 | |
| 34 | ; CHECK: for.body.ldist1: |
| 35 | ; CHECK: %mulA.ldist1 = mul i32 %loadB.ldist1, %loadA.ldist1 |
| 36 | ; CHECK: for.body.ph: |
| 37 | ; CHECK: for.body: |
| 38 | ; CHECK: %sum_add = add nuw nsw i32 %sum, %loadC |
| 39 | ; CHECK: for.end: |
| 40 | ; CHECK: %sum_add.lver = phi i32 [ %sum_add, %for.body ], [ %sum_add.lver.orig, %for.body.lver.orig ] |
| 41 | |
| 42 | for.body: ; preds = %for.body, %entry |
| 43 | %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] |
| 44 | %sum = phi i32 [ 0, %entry ], [ %sum_add, %for.body ] |
| 45 | |
| 46 | %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind |
| 47 | %loadA = load i32, i32* %arrayidxA, align 4 |
| 48 | |
| 49 | %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind |
| 50 | %loadB = load i32, i32* %arrayidxB, align 4 |
| 51 | |
| 52 | %mulA = mul i32 %loadB, %loadA |
| 53 | |
| 54 | %add = add nuw nsw i64 %ind, 1 |
| 55 | %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add |
| 56 | store i32 %mulA, i32* %arrayidxA_plus_4, align 4 |
| 57 | |
| 58 | %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind |
| 59 | %loadC = load i32, i32* %arrayidxC, align 4 |
| 60 | |
| 61 | %sum_add = add nuw nsw i32 %sum, %loadC |
| 62 | |
| 63 | %exitcond = icmp eq i64 %add, 20 |
| 64 | br i1 %exitcond, label %for.end, label %for.body |
| 65 | |
| 66 | for.end: ; preds = %for.body |
| 67 | store i32 %sum_add, i32* @SUM, align 4 |
| 68 | ret void |
| 69 | } |