@@ -227,90 +227,9 @@ define void @udiv_urem_feeding_gep(i64 %x, ptr %dst, i64 %N) {
227
227
; CHECK-NEXT: [[ENTRY:.*]]:
228
228
; CHECK-NEXT: [[MUL_1_I:%.*]] = mul i64 [[X]], [[X]]
229
229
; CHECK-NEXT: [[MUL_2_I:%.*]] = mul i64 [[MUL_1_I]], [[X]]
230
- ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1
231
- ; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
232
- ; CHECK: [[VECTOR_SCEVCHECK]]:
233
- ; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[N]] to i32
234
- ; CHECK-NEXT: [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0
235
- ; CHECK-NEXT: [[TMP3:%.*]] = icmp ugt i64 [[N]], 4294967295
236
- ; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP2]], [[TMP3]]
237
- ; CHECK-NEXT: br i1 [[TMP4]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
238
- ; CHECK: [[VECTOR_PH]]:
239
- ; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 [[TMP0]], 3
240
- ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], 4
241
- ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
242
- ; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i64 [[TMP0]], 1
243
- ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0
244
- ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
245
- ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
246
- ; CHECK: [[VECTOR_BODY]]:
247
- ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_UDIV_CONTINUE6:.*]] ]
248
- ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_UDIV_CONTINUE6]] ]
249
- ; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 0
250
- ; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
251
- ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0
252
- ; CHECK-NEXT: br i1 [[TMP6]], label %[[PRED_UDIV_IF:.*]], label %[[PRED_UDIV_CONTINUE:.*]]
253
- ; CHECK: [[PRED_UDIV_IF]]:
254
- ; CHECK-NEXT: [[TMP8:%.*]] = udiv i64 [[TMP7]], [[MUL_2_I]]
255
- ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x i64> poison, i64 [[TMP8]], i32 0
256
- ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE]]
257
- ; CHECK: [[PRED_UDIV_CONTINUE]]:
258
- ; CHECK-NEXT: [[TMP15:%.*]] = phi <4 x i64> [ poison, %[[VECTOR_BODY]] ], [ [[TMP9]], %[[PRED_UDIV_IF]] ]
259
- ; CHECK-NEXT: [[TMP21:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1
260
- ; CHECK-NEXT: br i1 [[TMP21]], label %[[PRED_UDIV_IF1:.*]], label %[[PRED_UDIV_CONTINUE2:.*]]
261
- ; CHECK: [[PRED_UDIV_IF1]]:
262
- ; CHECK-NEXT: [[TMP22:%.*]] = add i64 [[INDEX]], 1
263
- ; CHECK-NEXT: [[TMP23:%.*]] = udiv i64 [[TMP22]], [[MUL_2_I]]
264
- ; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i64> [[TMP15]], i64 [[TMP23]], i32 1
265
- ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE2]]
266
- ; CHECK: [[PRED_UDIV_CONTINUE2]]:
267
- ; CHECK-NEXT: [[TMP30:%.*]] = phi <4 x i64> [ [[TMP15]], %[[PRED_UDIV_CONTINUE]] ], [ [[TMP24]], %[[PRED_UDIV_IF1]] ]
268
- ; CHECK-NEXT: [[TMP36:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2
269
- ; CHECK-NEXT: br i1 [[TMP36]], label %[[PRED_UDIV_IF3:.*]], label %[[PRED_UDIV_CONTINUE4:.*]]
270
- ; CHECK: [[PRED_UDIV_IF3]]:
271
- ; CHECK-NEXT: [[TMP37:%.*]] = add i64 [[INDEX]], 2
272
- ; CHECK-NEXT: [[TMP38:%.*]] = udiv i64 [[TMP37]], [[MUL_2_I]]
273
- ; CHECK-NEXT: [[TMP39:%.*]] = insertelement <4 x i64> [[TMP30]], i64 [[TMP38]], i32 2
274
- ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE4]]
275
- ; CHECK: [[PRED_UDIV_CONTINUE4]]:
276
- ; CHECK-NEXT: [[TMP45:%.*]] = phi <4 x i64> [ [[TMP30]], %[[PRED_UDIV_CONTINUE2]] ], [ [[TMP39]], %[[PRED_UDIV_IF3]] ]
277
- ; CHECK-NEXT: [[TMP51:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3
278
- ; CHECK-NEXT: br i1 [[TMP51]], label %[[PRED_UDIV_IF5:.*]], label %[[PRED_UDIV_CONTINUE6]]
279
- ; CHECK: [[PRED_UDIV_IF5]]:
280
- ; CHECK-NEXT: [[TMP52:%.*]] = add i64 [[INDEX]], 3
281
- ; CHECK-NEXT: [[TMP53:%.*]] = udiv i64 [[TMP52]], [[MUL_2_I]]
282
- ; CHECK-NEXT: [[TMP54:%.*]] = insertelement <4 x i64> [[TMP45]], i64 [[TMP53]], i32 3
283
- ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE6]]
284
- ; CHECK: [[PRED_UDIV_CONTINUE6]]:
285
- ; CHECK-NEXT: [[TMP60:%.*]] = phi <4 x i64> [ [[TMP45]], %[[PRED_UDIV_CONTINUE4]] ], [ [[TMP54]], %[[PRED_UDIV_IF5]] ]
286
- ; CHECK-NEXT: [[TMP55:%.*]] = urem i64 [[TMP7]], [[MUL_2_I]]
287
- ; CHECK-NEXT: [[TMP56:%.*]] = udiv i64 [[TMP55]], [[MUL_1_I]]
288
- ; CHECK-NEXT: [[TMP57:%.*]] = urem i64 [[TMP55]], [[MUL_1_I]]
289
- ; CHECK-NEXT: [[TMP58:%.*]] = udiv i64 [[TMP57]], [[X]]
290
- ; CHECK-NEXT: [[TMP59:%.*]] = urem i64 [[TMP57]], [[X]]
291
- ; CHECK-NEXT: [[TMP66:%.*]] = extractelement <4 x i64> [[TMP60]], i32 0
292
- ; CHECK-NEXT: [[TMP67:%.*]] = mul i64 [[X]], [[TMP66]]
293
- ; CHECK-NEXT: [[TMP68:%.*]] = add i64 [[TMP67]], [[TMP56]]
294
- ; CHECK-NEXT: [[TMP69:%.*]] = mul i64 [[TMP68]], [[X]]
295
- ; CHECK-NEXT: [[TMP70:%.*]] = add i64 [[TMP69]], [[TMP58]]
296
- ; CHECK-NEXT: [[TMP71:%.*]] = mul i64 [[TMP70]], [[X]]
297
- ; CHECK-NEXT: [[TMP72:%.*]] = add i64 [[TMP71]], [[TMP59]]
298
- ; CHECK-NEXT: [[TMP73:%.*]] = shl i64 [[TMP72]], 32
299
- ; CHECK-NEXT: [[TMP74:%.*]] = ashr i64 [[TMP73]], 32
300
- ; CHECK-NEXT: [[TMP75:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP74]]
301
- ; CHECK-NEXT: [[TMP76:%.*]] = getelementptr i64, ptr [[TMP75]], i32 0
302
- ; CHECK-NEXT: call void @llvm.masked.store.v4i64.p0(<4 x i64> [[TMP60]], ptr [[TMP76]], i32 4, <4 x i1> [[TMP5]])
303
- ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
304
- ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], <i64 4, i64 4, i64 4, i64 4>
305
- ; CHECK-NEXT: [[TMP77:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
306
- ; CHECK-NEXT: br i1 [[TMP77]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
307
- ; CHECK: [[MIDDLE_BLOCK]]:
308
- ; CHECK-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]]
309
- ; CHECK: [[SCALAR_PH]]:
310
- ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ], [ 0, %[[VECTOR_SCEVCHECK]] ]
311
230
; CHECK-NEXT: br label %[[LOOP:.*]]
312
231
; CHECK: [[LOOP]]:
313
- ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]] , %[[SCALAR_PH ]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
232
+ ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0 , %[[ENTRY ]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
314
233
; CHECK-NEXT: [[DIV_I:%.*]] = udiv i64 [[IV]], [[MUL_2_I]]
315
234
; CHECK-NEXT: [[REM_I:%.*]] = urem i64 [[IV]], [[MUL_2_I]]
316
235
; CHECK-NEXT: [[DIV_1_I:%.*]] = udiv i64 [[REM_I]], [[MUL_1_I]]
@@ -329,7 +248,7 @@ define void @udiv_urem_feeding_gep(i64 %x, ptr %dst, i64 %N) {
329
248
; CHECK-NEXT: store i64 [[DIV_I]], ptr [[GEP]], align 4
330
249
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
331
250
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV]], [[N]]
332
- ; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP7:![0-9]+ ]]
251
+ ; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.* ]], label %[[LOOP]]
333
252
; CHECK: [[EXIT]]:
334
253
; CHECK-NEXT: ret void
335
254
;
@@ -370,6 +289,4 @@ exit:
370
289
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
371
290
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
372
291
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]]}
373
- ; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
374
- ; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]]}
375
292
;.
0 commit comments