Skip to content

Commit 012e574

Browse files
fhahnfhahn
committed
[LV] Add FindLastIV test with truncated IV and epilogue vectorization.
This adds missing test coverage for #132691.
1 parent 7c4013d commit 012e574

File tree

1 file changed

+126
-0
lines changed

1 file changed

+126
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/AArch64/epilog-iv-select-cmp.ll

Lines changed: 126 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -120,3 +120,129 @@ loop:
120120
exit:
121121
ret i8 %sel
122122
}
123+
124+
define i32 @select_icmp_var_start_iv_trunc(i32 %N, i32 %start) #0 {
125+
; CHECK-LABEL: define i32 @select_icmp_var_start_iv_trunc(
126+
; CHECK-SAME: i32 [[N:%.*]], i32 [[START:%.*]]) #[[ATTR0:[0-9]+]] {
127+
; CHECK-NEXT: [[ITER_CHECK:.*]]:
128+
; CHECK-NEXT: [[N_POS:%.*]] = icmp sgt i32 [[N]], 0
129+
; CHECK-NEXT: call void @llvm.assume(i1 [[N_POS]])
130+
; CHECK-NEXT: [[N_EXT:%.*]] = zext i32 [[N]] to i64
131+
; CHECK-NEXT: [[TMP0:%.*]] = add nuw nsw i64 [[N_EXT]], 1
132+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4
133+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]
134+
; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:
135+
; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP0]], 16
136+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
137+
; CHECK: [[VECTOR_PH]]:
138+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 16
139+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
140+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[START]], i64 0
141+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
142+
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq <4 x i32> [[BROADCAST_SPLAT]], zeroinitializer
143+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
144+
; CHECK: [[VECTOR_BODY]]:
145+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
146+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ splat (i32 -2147483648), %[[VECTOR_PH]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]
147+
; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ splat (i32 -2147483648), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ]
148+
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ splat (i32 -2147483648), %[[VECTOR_PH]] ], [ [[TMP5:%.*]], %[[VECTOR_BODY]] ]
149+
; CHECK-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ splat (i32 -2147483648), %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
150+
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
151+
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], splat (i32 4)
152+
; CHECK-NEXT: [[STEP_ADD_2:%.*]] = add <4 x i32> [[STEP_ADD]], splat (i32 4)
153+
; CHECK-NEXT: [[STEP_ADD_3:%.*]] = add <4 x i32> [[STEP_ADD_2]], splat (i32 4)
154+
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <4 x i1> [[TMP1]], i32 0
155+
; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], <4 x i32> [[VEC_IND]], <4 x i32> [[VEC_PHI]]
156+
; CHECK-NEXT: [[TMP4]] = select i1 [[TMP2]], <4 x i32> [[STEP_ADD]], <4 x i32> [[VEC_PHI2]]
157+
; CHECK-NEXT: [[TMP5]] = select i1 [[TMP2]], <4 x i32> [[STEP_ADD_2]], <4 x i32> [[VEC_PHI3]]
158+
; CHECK-NEXT: [[TMP6]] = select i1 [[TMP2]], <4 x i32> [[STEP_ADD_3]], <4 x i32> [[VEC_PHI4]]
159+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
160+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[STEP_ADD_3]], splat (i32 4)
161+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
162+
; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
163+
; CHECK: [[MIDDLE_BLOCK]]:
164+
; CHECK-NEXT: [[RDX_MINMAX:%.*]] = call <4 x i32> @llvm.smax.v4i32(<4 x i32> [[TMP3]], <4 x i32> [[TMP4]])
165+
; CHECK-NEXT: [[RDX_MINMAX5:%.*]] = call <4 x i32> @llvm.smax.v4i32(<4 x i32> [[RDX_MINMAX]], <4 x i32> [[TMP5]])
166+
; CHECK-NEXT: [[RDX_MINMAX6:%.*]] = call <4 x i32> @llvm.smax.v4i32(<4 x i32> [[RDX_MINMAX5]], <4 x i32> [[TMP6]])
167+
; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[RDX_MINMAX6]])
168+
; CHECK-NEXT: [[RDX_SELECT_CMP:%.*]] = icmp ne i32 [[TMP8]], -2147483648
169+
; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[RDX_SELECT_CMP]], i32 [[TMP8]], i32 [[START]]
170+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
171+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]
172+
; CHECK: [[VEC_EPILOG_ITER_CHECK]]:
173+
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP0]], [[N_VEC]]
174+
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4
175+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]]
176+
; CHECK: [[VEC_EPILOG_PH]]:
177+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
178+
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[RDX_SELECT]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
179+
; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i32 [[BC_MERGE_RDX]], [[START]]
180+
; CHECK-NEXT: [[TMP10:%.*]] = select i1 [[TMP9]], i32 -2147483648, i32 [[BC_MERGE_RDX]]
181+
; CHECK-NEXT: [[N_MOD_VF7:%.*]] = urem i64 [[TMP0]], 4
182+
; CHECK-NEXT: [[N_VEC8:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF7]]
183+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT9:%.*]] = insertelement <4 x i32> poison, i32 [[START]], i64 0
184+
; CHECK-NEXT: [[BROADCAST_SPLAT10:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT9]], <4 x i32> poison, <4 x i32> zeroinitializer
185+
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq <4 x i32> [[BROADCAST_SPLAT10]], zeroinitializer
186+
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP10]], i64 0
187+
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
188+
; CHECK-NEXT: [[TMP12:%.*]] = trunc i64 [[BC_RESUME_VAL]] to i32
189+
; CHECK-NEXT: [[DOTSPLATINSERT13:%.*]] = insertelement <4 x i32> poison, i32 [[TMP12]], i64 0
190+
; CHECK-NEXT: [[DOTSPLAT14:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT13]], <4 x i32> poison, <4 x i32> zeroinitializer
191+
; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i32> [[DOTSPLAT14]], <i32 0, i32 1, i32 2, i32 3>
192+
; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
193+
; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:
194+
; CHECK-NEXT: [[INDEX11:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT17:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
195+
; CHECK-NEXT: [[VEC_PHI12:%.*]] = phi <4 x i32> [ [[DOTSPLAT]], %[[VEC_EPILOG_PH]] ], [ [[TMP14:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
196+
; CHECK-NEXT: [[VEC_IND15:%.*]] = phi <4 x i32> [ [[INDUCTION]], %[[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT16:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
197+
; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[TMP11]], i32 0
198+
; CHECK-NEXT: [[TMP14]] = select i1 [[TMP13]], <4 x i32> [[VEC_IND15]], <4 x i32> [[VEC_PHI12]]
199+
; CHECK-NEXT: [[INDEX_NEXT17]] = add nuw i64 [[INDEX11]], 4
200+
; CHECK-NEXT: [[VEC_IND_NEXT16]] = add <4 x i32> [[VEC_IND15]], splat (i32 4)
201+
; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT17]], [[N_VEC8]]
202+
; CHECK-NEXT: br i1 [[TMP15]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
203+
; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:
204+
; CHECK-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP14]])
205+
; CHECK-NEXT: [[RDX_SELECT_CMP18:%.*]] = icmp ne i32 [[TMP16]], -2147483648
206+
; CHECK-NEXT: [[RDX_SELECT19:%.*]] = select i1 [[RDX_SELECT_CMP18]], i32 [[TMP16]], i32 [[START]]
207+
; CHECK-NEXT: [[CMP_N20:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC8]]
208+
; CHECK-NEXT: br i1 [[CMP_N20]], label %[[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]]
209+
; CHECK: [[VEC_EPILOG_SCALAR_PH]]:
210+
; CHECK-NEXT: [[BC_RESUME_VAL21:%.*]] = phi i64 [ [[N_VEC8]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]
211+
; CHECK-NEXT: [[BC_MERGE_RDX22:%.*]] = phi i32 [ [[RDX_SELECT19]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[RDX_SELECT]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], %[[ITER_CHECK]] ]
212+
; CHECK-NEXT: br label %[[LOOP:.*]]
213+
; CHECK: [[LOOP]]:
214+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL21]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
215+
; CHECK-NEXT: [[RED:%.*]] = phi i32 [ [[BC_MERGE_RDX22]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[RED_NEXT:%.*]], %[[LOOP]] ]
216+
; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[START]], 0
217+
; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i32
218+
; CHECK-NEXT: [[RED_NEXT]] = select i1 [[C]], i32 [[IV_TRUNC]], i32 [[RED]]
219+
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
220+
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], [[N_EXT]]
221+
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP7:![0-9]+]]
222+
; CHECK: [[EXIT]]:
223+
; CHECK-NEXT: [[RED_NEXT_LCSSA:%.*]] = phi i32 [ [[RED_NEXT]], %[[LOOP]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[RDX_SELECT19]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ]
224+
; CHECK-NEXT: ret i32 [[RED_NEXT_LCSSA]]
225+
;
226+
entry:
227+
%N.pos = icmp sgt i32 %N, 0
228+
call void @llvm.assume(i1 %N.pos)
229+
%N.ext = zext i32 %N to i64
230+
br label %loop
231+
232+
loop:
233+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
234+
%red = phi i32 [ %start, %entry ], [ %red.next, %loop ]
235+
%c = icmp eq i32 %start, 0
236+
%iv.trunc = trunc i64 %iv to i32
237+
%red.next = select i1 %c, i32 %iv.trunc, i32 %red
238+
%iv.next = add i64 %iv, 1
239+
%ec = icmp eq i64 %iv, %N.ext
240+
br i1 %ec, label %exit, label %loop
241+
242+
exit:
243+
ret i32 %red.next
244+
}
245+
246+
declare void @llvm.assume(i1 noundef)
247+
248+
attributes #0 = { "target-cpu"="apple-m1" }

0 commit comments

Comments
 (0)