1 ; RUN: opt < %s -indvars -S | FileCheck %s
3 ; Make sure that indvars can perform LFTR without a canonical IV.
5 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-n8:16:32:64"
7 ; Perform LFTR using the original pointer-type IV.
9 ; for(char* p = base; p < base + n; ++p) {
12 define void @ptriv(i8* %base, i32 %n) nounwind {
14 %idx.ext = sext i32 %n to i64
15 %add.ptr = getelementptr inbounds i8* %base, i64 %idx.ext
16 %cmp1 = icmp ult i8* %base, %add.ptr
17 br i1 %cmp1, label %for.body, label %for.end
26 %p.02 = phi i8* [ %base, %entry ], [ %incdec.ptr, %for.body ]
27 ; cruft to make the IV useful
28 %sub.ptr.lhs.cast = ptrtoint i8* %p.02 to i64
29 %sub.ptr.rhs.cast = ptrtoint i8* %base to i64
30 %sub.ptr.sub = sub i64 %sub.ptr.lhs.cast, %sub.ptr.rhs.cast
31 %conv = trunc i64 %sub.ptr.sub to i8
32 store i8 %conv, i8* %p.02
33 %incdec.ptr = getelementptr inbounds i8* %p.02, i32 1
34 %cmp = icmp ult i8* %incdec.ptr, %add.ptr
35 br i1 %cmp, label %for.body, label %for.end
41 ; This test checks that SCEVExpander can handle an outer loop that has been
42 ; simplified, and as a result the inner loop's exit test will be rewritten.
43 define void @expandOuterRecurrence(i32 %arg) nounwind {
45 %sub1 = sub nsw i32 %arg, 1
46 %cmp1 = icmp slt i32 0, %sub1
47 br i1 %cmp1, label %outer, label %exit
52 %i = phi i32 [ 0, %entry ], [ %i.inc, %outer.inc ]
53 %sub2 = sub nsw i32 %arg, %i
54 %sub3 = sub nsw i32 %sub2, 1
55 %cmp2 = icmp slt i32 0, %sub3
56 br i1 %cmp2, label %inner.ph, label %outer.inc
64 %j = phi i32 [ 0, %inner.ph ], [ %j.inc, %inner ]
65 %j.inc = add nsw i32 %j, 1
66 %cmp3 = icmp slt i32 %j.inc, %sub3
67 br i1 %cmp3, label %inner, label %outer.inc
73 %i.inc = add nsw i32 %i, 1
74 %cmp4 = icmp slt i32 %i.inc, %sub1
75 br i1 %cmp4, label %outer, label %exit
81 ; Force SCEVExpander to look for an existing well-formed phi.
82 ; Perform LFTR without generating extra preheader code.
83 define void @guardedloop([0 x double]* %matrix, [0 x double]* %vector,
84 i32 %irow, i32 %ilead) nounwind {
85 ; CHECK-LABEL: @guardedloop(
87 ; CHECK-NEXT: %[[cmp:.*]] = icmp slt i32 1, %irow
88 ; CHECK-NEXT: br i1 %[[cmp]], label %[[loop_preheader:.*]], label %[[return:.*]]
90 ; CHECK: [[loop_preheader]]:
91 ; CHECK-NEXT: %[[sext:.*]] = sext i32 %ilead to i64
92 ; CHECK-NEXT: %[[add:.*]] = add i32 %irow, -1
93 ; CHECK-NEXT: br label %[[loop:.*]]
96 ; CHECK-NEXT: %[[indvars_iv2:.*]] = phi i64
99 ; CHECK: %[[lftr_wideiv:.*]] = trunc i64 %[[indvars_iv2]] to i32
100 ; CHECK-NEXT: %[[exitcond:.*]] = icmp ne i32 %[[lftr_wideiv]], %[[add]]
101 ; CHECK-NEXT: br i1 %[[exitcond]], label %[[loop]], label
103 %cmp = icmp slt i32 1, %irow
104 br i1 %cmp, label %loop, label %return
107 %rowidx = phi i32 [ 0, %entry ], [ %row.inc, %loop ]
108 %i = phi i32 [ 0, %entry ], [ %i.inc, %loop ]
109 %diagidx = add nsw i32 %rowidx, %i
110 %diagidxw = sext i32 %diagidx to i64
111 %matrixp = getelementptr inbounds [0 x double]* %matrix, i32 0, i64 %diagidxw
112 %v1 = load double* %matrixp
113 %iw = sext i32 %i to i64
114 %vectorp = getelementptr inbounds [0 x double]* %vector, i32 0, i64 %iw
115 %v2 = load double* %vectorp
116 %row.inc = add nsw i32 %rowidx, %ilead
117 %i.inc = add nsw i32 %i, 1
118 %cmp196 = icmp slt i32 %i.inc, %irow
119 br i1 %cmp196, label %loop, label %return
125 ; Avoid generating extra code to materialize a trip count. Skip LFTR.
126 define void @unguardedloop([0 x double]* %matrix, [0 x double]* %vector,
127 i32 %irow, i32 %ilead) nounwind {
141 %rowidx = phi i32 [ 0, %entry ], [ %row.inc, %loop ]
142 %i = phi i32 [ 0, %entry ], [ %i.inc, %loop ]
143 %diagidx = add nsw i32 %rowidx, %i
144 %diagidxw = sext i32 %diagidx to i64
145 %matrixp = getelementptr inbounds [0 x double]* %matrix, i32 0, i64 %diagidxw
146 %v1 = load double* %matrixp
147 %iw = sext i32 %i to i64
148 %vectorp = getelementptr inbounds [0 x double]* %vector, i32 0, i64 %iw
149 %v2 = load double* %vectorp
150 %row.inc = add nsw i32 %rowidx, %ilead
151 %i.inc = add nsw i32 %i, 1
152 %cmp196 = icmp slt i32 %i.inc, %irow
153 br i1 %cmp196, label %loop, label %return
159 ; Remove %i which is only used by the exit test.
160 ; Verify that SCEV can still compute a backedge count from the sign
161 ; extended %n, used for pointer comparison by LFTR.
163 ; TODO: Fix for PR13371 currently makes this impossible. See
164 ; IndVarSimplify.cpp hasConcreteDef(). We may want to change to undef rules.
165 define void @geplftr(i8* %base, i32 %x, i32 %y, i32 %n) nounwind {
167 %x.ext = sext i32 %x to i64
168 %add.ptr = getelementptr inbounds i8* %base, i64 %x.ext
169 %y.ext = sext i32 %y to i64
170 %add.ptr10 = getelementptr inbounds i8* %add.ptr, i64 %y.ext
171 %lim = add i32 %x, %n
172 %cmp.ph = icmp ult i32 %x, %lim
173 br i1 %cmp.ph, label %loop, label %exit
174 ; CHECK-LABEL: @geplftr(
177 ; DISABLE-NOT: phi // This check is currently disabled
178 ; CHECK: getelementptr
180 ; DISABLE: icmp ne i8* // This check is currently disabled
183 %i = phi i32 [ %x, %entry ], [ %inc, %loop ]
184 %aptr = phi i8* [ %add.ptr10, %entry ], [ %incdec.ptr, %loop ]
185 %incdec.ptr = getelementptr inbounds i8* %aptr, i32 1
186 store i8 3, i8* %aptr
188 %cmp = icmp ult i32 %inc, %lim
189 br i1 %cmp, label %loop, label %exit
195 ; Exercise backedge taken count verification with a never-taken loop.
196 define void @nevertaken() nounwind uwtable ssp {
199 ; CHECK-LABEL: @nevertaken(
206 %i = phi i32 [ 0, %entry ], [ %inc, %loop ]
207 %inc = add nsw i32 %i, 1
208 %cmp = icmp sle i32 %inc, 0
209 br i1 %cmp, label %loop, label %exit
215 ; Test LFTR on an IV whose recurrence start is a non-unit pointer type.
216 define void @aryptriv([256 x i8]* %base, i32 %n) nounwind {
218 %ivstart = getelementptr inbounds [256 x i8]* %base, i32 0, i32 0
219 %ivend = getelementptr inbounds [256 x i8]* %base, i32 0, i32 %n
220 %cmp.ph = icmp ult i8* %ivstart, %ivend
221 br i1 %cmp.ph, label %loop, label %exit
226 ; CHECK: getelementptr
231 %aptr = phi i8* [ %ivstart, %entry ], [ %incdec.ptr, %loop ]
232 %incdec.ptr = getelementptr inbounds i8* %aptr, i32 1
233 store i8 3, i8* %aptr
234 %cmp = icmp ult i8* %incdec.ptr, %ivend
235 br i1 %cmp, label %loop, label %exit