Andrew Trick | 4086bb5 | 2011-06-28 06:34:10 +0000 | [diff] [blame] | 1 | ; RUN: opt < %s -indvars -S | FileCheck %s |
Andrew Trick | 4086bb5 | 2011-06-28 06:34:10 +0000 | [diff] [blame] | 2 | ; |
Dan Gohman | 81db61a | 2009-05-12 02:17:14 +0000 | [diff] [blame] | 3 | ; PR1301 |
| 4 | |
| 5 | ; Do a bunch of analysis and prove that the loops can use an i32 trip |
| 6 | ; count without casting. |
Andrew Trick | 4086bb5 | 2011-06-28 06:34:10 +0000 | [diff] [blame] | 7 | ; |
| 8 | ; Note that all four functions should actually be converted to |
| 9 | ; memset. However, this test case validates indvars behavior. We |
| 10 | ; don't check that phis are "folded together" because that is a job |
Andrew Trick | fc933c0 | 2011-07-18 20:32:31 +0000 | [diff] [blame] | 11 | ; for loop strength reduction. But indvars must remove sext, zext, and add i8. |
Andrew Trick | 4086bb5 | 2011-06-28 06:34:10 +0000 | [diff] [blame] | 12 | ; |
Andrew Trick | fc933c0 | 2011-07-18 20:32:31 +0000 | [diff] [blame] | 13 | ; CHECK-NOT: {{sext|zext|add i8}} |
Dan Gohman | 81db61a | 2009-05-12 02:17:14 +0000 | [diff] [blame] | 14 | |
| 15 | ; ModuleID = 'ada.bc' |
Lang Hames | 68df750 | 2011-10-12 22:24:17 +0000 | [diff] [blame] | 16 | target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32" |
Dan Gohman | 81db61a | 2009-05-12 02:17:14 +0000 | [diff] [blame] | 17 | target triple = "i686-pc-linux-gnu" |
| 18 | |
| 19 | define void @kinds__sbytezero([256 x i32]* nocapture %a) nounwind { |
| 20 | bb.thread: |
| 21 | %tmp46 = getelementptr [256 x i32]* %a, i32 0, i32 0 ; <i32*> [#uses=1] |
| 22 | store i32 0, i32* %tmp46 |
| 23 | br label %bb |
| 24 | |
| 25 | bb: ; preds = %bb, %bb.thread |
| 26 | %i.0.reg2mem.0 = phi i8 [ -128, %bb.thread ], [ %tmp8, %bb ] ; <i8> [#uses=1] |
| 27 | %tmp8 = add i8 %i.0.reg2mem.0, 1 ; <i8> [#uses=3] |
| 28 | %tmp1 = sext i8 %tmp8 to i32 ; <i32> [#uses=1] |
| 29 | %tmp3 = add i32 %tmp1, 128 ; <i32> [#uses=1] |
| 30 | %tmp4 = getelementptr [256 x i32]* %a, i32 0, i32 %tmp3 ; <i32*> [#uses=1] |
| 31 | store i32 0, i32* %tmp4 |
| 32 | %0 = icmp eq i8 %tmp8, 127 ; <i1> [#uses=1] |
| 33 | br i1 %0, label %return, label %bb |
| 34 | |
| 35 | return: ; preds = %bb |
| 36 | ret void |
| 37 | } |
| 38 | |
| 39 | define void @kinds__ubytezero([256 x i32]* nocapture %a) nounwind { |
| 40 | bb.thread: |
| 41 | %tmp35 = getelementptr [256 x i32]* %a, i32 0, i32 0 ; <i32*> [#uses=1] |
| 42 | store i32 0, i32* %tmp35 |
| 43 | br label %bb |
| 44 | |
| 45 | bb: ; preds = %bb, %bb.thread |
| 46 | %i.0.reg2mem.0 = phi i8 [ 0, %bb.thread ], [ %tmp7, %bb ] ; <i8> [#uses=1] |
| 47 | %tmp7 = add i8 %i.0.reg2mem.0, 1 ; <i8> [#uses=3] |
| 48 | %tmp1 = zext i8 %tmp7 to i32 ; <i32> [#uses=1] |
| 49 | %tmp3 = getelementptr [256 x i32]* %a, i32 0, i32 %tmp1 ; <i32*> [#uses=1] |
| 50 | store i32 0, i32* %tmp3 |
| 51 | %0 = icmp eq i8 %tmp7, -1 ; <i1> [#uses=1] |
| 52 | br i1 %0, label %return, label %bb |
| 53 | |
| 54 | return: ; preds = %bb |
| 55 | ret void |
| 56 | } |
| 57 | |
| 58 | define void @kinds__srangezero([21 x i32]* nocapture %a) nounwind { |
| 59 | bb.thread: |
| 60 | br label %bb |
| 61 | |
| 62 | bb: ; preds = %bb, %bb.thread |
| 63 | %i.0.reg2mem.0 = phi i8 [ -10, %bb.thread ], [ %tmp7, %bb ] ; <i8> [#uses=2] |
| 64 | %tmp12 = sext i8 %i.0.reg2mem.0 to i32 ; <i32> [#uses=1] |
| 65 | %tmp4 = add i32 %tmp12, 10 ; <i32> [#uses=1] |
| 66 | %tmp5 = getelementptr [21 x i32]* %a, i32 0, i32 %tmp4 ; <i32*> [#uses=1] |
| 67 | store i32 0, i32* %tmp5 |
| 68 | %tmp7 = add i8 %i.0.reg2mem.0, 1 ; <i8> [#uses=2] |
| 69 | %0 = icmp sgt i8 %tmp7, 10 ; <i1> [#uses=1] |
| 70 | br i1 %0, label %return, label %bb |
| 71 | |
| 72 | return: ; preds = %bb |
| 73 | ret void |
| 74 | } |
| 75 | |
| 76 | define void @kinds__urangezero([21 x i32]* nocapture %a) nounwind { |
| 77 | bb.thread: |
| 78 | br label %bb |
| 79 | |
| 80 | bb: ; preds = %bb, %bb.thread |
| 81 | %i.0.reg2mem.0 = phi i8 [ 10, %bb.thread ], [ %tmp7, %bb ] ; <i8> [#uses=2] |
| 82 | %tmp12 = sext i8 %i.0.reg2mem.0 to i32 ; <i32> [#uses=1] |
| 83 | %tmp4 = add i32 %tmp12, -10 ; <i32> [#uses=1] |
| 84 | %tmp5 = getelementptr [21 x i32]* %a, i32 0, i32 %tmp4 ; <i32*> [#uses=1] |
| 85 | store i32 0, i32* %tmp5 |
| 86 | %tmp7 = add i8 %i.0.reg2mem.0, 1 ; <i8> [#uses=2] |
| 87 | %0 = icmp sgt i8 %tmp7, 30 ; <i1> [#uses=1] |
| 88 | br i1 %0, label %return, label %bb |
| 89 | |
| 90 | return: ; preds = %bb |
| 91 | ret void |
| 92 | } |