Skip to content

Commit 04b5c63

Browse files
fhahnfhahn
committed
[LV] Add inbounds to interleave test.
In preparation for #127543
1 parent 3b9f964 commit 04b5c63

File tree

1 file changed

+27
-63
lines changed

1 file changed

+27
-63
lines changed

‎llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll

Lines changed: 27 additions & 63 deletions
Original file line numberDiff line numberDiff line change
@@ -114,44 +114,8 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
114114
; CHECK-SAME: ptr [[ARG:%.*]], i64 [[ARG1:%.*]], ptr [[ARG2:%.*]]) #[[ATTR0:[0-9]+]] {
115115
; CHECK-NEXT: [[ENTRY:.*]]:
116116
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[ARG1]], 1
117-
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 54
118-
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
119-
; CHECK: [[VECTOR_SCEVCHECK]]:
120-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[ARG2]], i64 8
121-
; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[ARG1]])
122-
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
123-
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
124-
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
125-
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT]]
126-
; CHECK-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[SCEVGEP]]
127-
; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
128-
; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[ARG2]], i64 12
129-
; CHECK-NEXT: [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[ARG1]])
130-
; CHECK-NEXT: [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
131-
; CHECK-NEXT: [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
132-
; CHECK-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT3]]
133-
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SCEVGEP1]], i64 [[MUL_RESULT3]]
134-
; CHECK-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[SCEVGEP1]]
135-
; CHECK-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW4]]
136-
; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr i8, ptr [[ARG2]], i64 4
137-
; CHECK-NEXT: [[MUL6:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[ARG1]])
138-
; CHECK-NEXT: [[MUL_RESULT7:%.*]] = extractvalue { i64, i1 } [[MUL6]], 0
139-
; CHECK-NEXT: [[MUL_OVERFLOW8:%.*]] = extractvalue { i64, i1 } [[MUL6]], 1
140-
; CHECK-NEXT: [[TMP9:%.*]] = sub i64 0, [[MUL_RESULT7]]
141-
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[SCEVGEP5]], i64 [[MUL_RESULT7]]
142-
; CHECK-NEXT: [[TMP11:%.*]] = icmp ult ptr [[TMP10]], [[SCEVGEP5]]
143-
; CHECK-NEXT: [[TMP12:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW8]]
144-
; CHECK-NEXT: [[MUL9:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[ARG1]])
145-
; CHECK-NEXT: [[MUL_RESULT10:%.*]] = extractvalue { i64, i1 } [[MUL9]], 0
146-
; CHECK-NEXT: [[MUL_OVERFLOW11:%.*]] = extractvalue { i64, i1 } [[MUL9]], 1
147-
; CHECK-NEXT: [[TMP13:%.*]] = sub i64 0, [[MUL_RESULT10]]
148-
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[ARG2]], i64 [[MUL_RESULT10]]
149-
; CHECK-NEXT: [[TMP15:%.*]] = icmp ult ptr [[TMP14]], [[ARG2]]
150-
; CHECK-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW11]]
151-
; CHECK-NEXT: [[TMP17:%.*]] = or i1 [[TMP4]], [[TMP8]]
152-
; CHECK-NEXT: [[TMP18:%.*]] = or i1 [[TMP17]], [[TMP12]]
153-
; CHECK-NEXT: [[TMP19:%.*]] = or i1 [[TMP18]], [[TMP16]]
154-
; CHECK-NEXT: br i1 [[TMP19]], label %[[SCALAR_PH]], label %[[VECTOR_MEMCHECK:.*]]
117+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8
118+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
155119
; CHECK: [[VECTOR_MEMCHECK]]:
156120
; CHECK-NEXT: [[TMP20:%.*]] = shl i64 [[ARG1]], 4
157121
; CHECK-NEXT: [[TMP21:%.*]] = add i64 [[TMP20]], 16
@@ -171,9 +135,9 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
171135
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
172136
; CHECK-NEXT: [[TMP24:%.*]] = add i64 [[INDEX]], 0
173137
; CHECK-NEXT: [[TMP25:%.*]] = shl i64 [[TMP24]], 5
174-
; CHECK-NEXT: [[TMP26:%.*]] = getelementptr i8, ptr [[ARG]], i64 [[TMP25]]
138+
; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 [[TMP25]]
175139
; CHECK-NEXT: [[TMP27:%.*]] = shl i64 [[TMP24]], 4
176-
; CHECK-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[ARG2]], i64 [[TMP27]]
140+
; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 [[TMP27]]
177141
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <16 x float>, ptr [[TMP26]], align 4
178142
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 0, i32 8>
179143
; CHECK-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 1, i32 9>
@@ -203,44 +167,44 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
203167
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
204168
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
205169
; CHECK: [[SCALAR_PH]]:
206-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[VECTOR_MEMCHECK]] ]
170+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ], [ 0, %[[VECTOR_MEMCHECK]] ]
207171
; CHECK-NEXT: br label %[[LOOP:.*]]
208172
; CHECK: [[LOOP]]:
209173
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
210174
; CHECK-NEXT: [[SHL_IV_5:%.*]] = shl i64 [[IV]], 5
211-
; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr i8, ptr [[ARG]], i64 [[SHL_IV_5]]
175+
; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 [[SHL_IV_5]]
212176
; CHECK-NEXT: [[ADD_5:%.*]] = or disjoint i64 [[SHL_IV_5]], 16
213177
; CHECK-NEXT: [[GEP_2:%.*]] = getelementptr i8, ptr [[ARG]], i64 [[ADD_5]]
214178
; CHECK-NEXT: [[SHL_IV_4:%.*]] = shl i64 [[IV]], 4
215-
; CHECK-NEXT: [[GEP_3:%.*]] = getelementptr i8, ptr [[ARG2]], i64 [[SHL_IV_4]]
179+
; CHECK-NEXT: [[GEP_3:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 [[SHL_IV_4]]
216180
; CHECK-NEXT: [[L_1:%.*]] = load float, ptr [[GEP_1]], align 4
217181
; CHECK-NEXT: [[L_2:%.*]] = load float, ptr [[GEP_2]], align 4
218182
; CHECK-NEXT: [[ADD_1:%.*]] = fadd float [[L_1]], [[L_2]]
219183
; CHECK-NEXT: [[MUL_1:%.*]] = fmul float [[ADD_1]], 0.000000e+00
220184
; CHECK-NEXT: store float [[MUL_1]], ptr [[GEP_3]], align 4
221-
; CHECK-NEXT: [[GEP_4:%.*]] = getelementptr i8, ptr [[GEP_1]], i64 4
185+
; CHECK-NEXT: [[GEP_4:%.*]] = getelementptr inbounds i8, ptr [[GEP_1]], i64 4
222186
; CHECK-NEXT: [[L_3:%.*]] = load float, ptr [[GEP_4]], align 4
223-
; CHECK-NEXT: [[GEP_5:%.*]] = getelementptr i8, ptr [[GEP_2]], i64 4
187+
; CHECK-NEXT: [[GEP_5:%.*]] = getelementptr inbounds i8, ptr [[GEP_2]], i64 4
224188
; CHECK-NEXT: [[L_4:%.*]] = load float, ptr [[GEP_5]], align 4
225189
; CHECK-NEXT: [[ADD_2:%.*]] = fadd float [[L_3]], [[L_4]]
226190
; CHECK-NEXT: [[MUL_2:%.*]] = fmul float [[ADD_2]], 0.000000e+00
227-
; CHECK-NEXT: [[GEP_6:%.*]] = getelementptr i8, ptr [[GEP_3]], i64 4
191+
; CHECK-NEXT: [[GEP_6:%.*]] = getelementptr inbounds i8, ptr [[GEP_3]], i64 4
228192
; CHECK-NEXT: store float [[MUL_2]], ptr [[GEP_6]], align 4
229-
; CHECK-NEXT: [[GEP_7:%.*]] = getelementptr i8, ptr [[GEP_1]], i64 8
193+
; CHECK-NEXT: [[GEP_7:%.*]] = getelementptr inbounds i8, ptr [[GEP_1]], i64 8
230194
; CHECK-NEXT: [[L_5:%.*]] = load float, ptr [[GEP_7]], align 4
231-
; CHECK-NEXT: [[GEP_8:%.*]] = getelementptr i8, ptr [[GEP_2]], i64 8
195+
; CHECK-NEXT: [[GEP_8:%.*]] = getelementptr inbounds i8, ptr [[GEP_2]], i64 8
232196
; CHECK-NEXT: [[L_6:%.*]] = load float, ptr [[GEP_8]], align 4
233197
; CHECK-NEXT: [[ADD_3:%.*]] = fadd float [[L_5]], [[L_6]]
234198
; CHECK-NEXT: [[MUL_3:%.*]] = fmul float [[ADD_3]], 0.000000e+00
235-
; CHECK-NEXT: [[GEP_9:%.*]] = getelementptr i8, ptr [[GEP_3]], i64 8
199+
; CHECK-NEXT: [[GEP_9:%.*]] = getelementptr inbounds i8, ptr [[GEP_3]], i64 8
236200
; CHECK-NEXT: store float [[MUL_3]], ptr [[GEP_9]], align 4
237-
; CHECK-NEXT: [[I27:%.*]] = getelementptr i8, ptr [[GEP_1]], i64 12
201+
; CHECK-NEXT: [[I27:%.*]] = getelementptr inbounds i8, ptr [[GEP_1]], i64 12
238202
; CHECK-NEXT: [[L_7:%.*]] = load float, ptr [[I27]], align 4
239-
; CHECK-NEXT: [[GEP_10:%.*]] = getelementptr i8, ptr [[GEP_2]], i64 12
203+
; CHECK-NEXT: [[GEP_10:%.*]] = getelementptr inbounds i8, ptr [[GEP_2]], i64 12
240204
; CHECK-NEXT: [[L_8:%.*]] = load float, ptr [[GEP_10]], align 4
241205
; CHECK-NEXT: [[ADD_4:%.*]] = fadd float [[L_7]], [[L_8]]
242206
; CHECK-NEXT: [[MUL_4:%.*]] = fmul float [[ADD_4]], 0.000000e+00
243-
; CHECK-NEXT: [[GEP_11:%.*]] = getelementptr i8, ptr [[GEP_3]], i64 12
207+
; CHECK-NEXT: [[GEP_11:%.*]] = getelementptr inbounds i8, ptr [[GEP_3]], i64 12
244208
; CHECK-NEXT: store float [[MUL_4]], ptr [[GEP_11]], align 4
245209
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
246210
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], [[ARG1]]
@@ -254,39 +218,39 @@ entry:
254218
loop:
255219
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
256220
%shl.iv.5 = shl i64 %iv, 5
257-
%gep.1 = getelementptr i8, ptr %arg, i64 %shl.iv.5
221+
%gep.1 = getelementptr inbounds i8, ptr %arg, i64 %shl.iv.5
258222
%add.5 = or disjoint i64 %shl.iv.5, 16
259223
%gep.2 = getelementptr i8, ptr %arg, i64 %add.5
260224
%shl.iv.4 = shl i64 %iv, 4
261-
%gep.3 = getelementptr i8, ptr %arg2, i64 %shl.iv.4
225+
%gep.3 = getelementptr inbounds i8, ptr %arg2, i64 %shl.iv.4
262226
%l.1 = load float, ptr %gep.1, align 4
263227
%l.2 = load float, ptr %gep.2, align 4
264228
%add.1 = fadd float %l.1, %l.2
265229
%mul.1 = fmul float %add.1, 0.000000e+00
266230
store float %mul.1, ptr %gep.3, align 4
267-
%gep.4 = getelementptr i8, ptr %gep.1, i64 4
231+
%gep.4 = getelementptr inbounds i8, ptr %gep.1, i64 4
268232
%l.3 = load float, ptr %gep.4, align 4
269-
%gep.5 = getelementptr i8, ptr %gep.2, i64 4
233+
%gep.5 = getelementptr inbounds i8, ptr %gep.2, i64 4
270234
%l.4 = load float, ptr %gep.5, align 4
271235
%add.2 = fadd float %l.3, %l.4
272236
%mul.2 = fmul float %add.2, 0.000000e+00
273-
%gep.6 = getelementptr i8, ptr %gep.3, i64 4
237+
%gep.6 = getelementptr inbounds i8, ptr %gep.3, i64 4
274238
store float %mul.2, ptr %gep.6, align 4
275-
%gep.7 = getelementptr i8, ptr %gep.1, i64 8
239+
%gep.7 = getelementptr inbounds i8, ptr %gep.1, i64 8
276240
%l.5 = load float, ptr %gep.7, align 4
277-
%gep.8 = getelementptr i8, ptr %gep.2, i64 8
241+
%gep.8 = getelementptr inbounds i8, ptr %gep.2, i64 8
278242
%l.6 = load float, ptr %gep.8, align 4
279243
%add.3 = fadd float %l.5, %l.6
280244
%mul.3 = fmul float %add.3, 0.000000e+00
281-
%gep.9 = getelementptr i8, ptr %gep.3, i64 8
245+
%gep.9 = getelementptr inbounds i8, ptr %gep.3, i64 8
282246
store float %mul.3, ptr %gep.9, align 4
283-
%i27 = getelementptr i8, ptr %gep.1, i64 12
247+
%i27 = getelementptr inbounds i8, ptr %gep.1, i64 12
284248
%l.7 = load float, ptr %i27, align 4
285-
%gep.10 = getelementptr i8, ptr %gep.2, i64 12
249+
%gep.10 = getelementptr inbounds i8, ptr %gep.2, i64 12
286250
%l.8 = load float, ptr %gep.10, align 4
287251
%add.4 = fadd float %l.7, %l.8
288252
%mul.4 = fmul float %add.4, 0.000000e+00
289-
%gep.11 = getelementptr i8, ptr %gep.3, i64 12
253+
%gep.11 = getelementptr inbounds i8, ptr %gep.3, i64 12
290254
store float %mul.4, ptr %gep.11, align 4
291255
%iv.next = add i64 %iv, 1
292256
%ec = icmp eq i64 %iv, %arg1

0 commit comments

Comments
 (0)