Skip to content

Commit e3fb9f0

Browse files
fhahnfhahn
committed
!fixup add test with align attribute
1 parent a6b632a commit e3fb9f0

File tree

1 file changed

+186
-0
lines changed

1 file changed

+186
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-variable-size.ll

Lines changed: 186 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -369,3 +369,189 @@ loop.latch:
369369
exit:
370370
ret void
371371
}
372+
373+
; %a is known dereferenceable via assume for the whole loop, alignment is known via function attribute.
374+
define void @deref_assumption_in_preheader_non_constant_trip_count_access_i32_align_attribute(ptr noalias noundef align 4 %a, ptr noalias %b, ptr noalias %c, i64 %n) nofree nosync {
375+
; CHECK-LABEL: define void @deref_assumption_in_preheader_non_constant_trip_count_access_i32_align_attribute(
376+
; CHECK-SAME: ptr noalias noundef align 4 [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]], i64 [[N:%.*]]) #[[ATTR1]] {
377+
; CHECK-NEXT: [[ENTRY:.*]]:
378+
; CHECK-NEXT: [[MUL:%.*]] = mul nuw nsw i64 [[N]], 4
379+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[A]], i64 [[MUL]]) ]
380+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2
381+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
382+
; CHECK: [[VECTOR_PH]]:
383+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2
384+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
385+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
386+
; CHECK: [[VECTOR_BODY]]:
387+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
388+
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
389+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]]
390+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0
391+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP2]], align 4
392+
; CHECK-NEXT: [[TMP3:%.*]] = icmp sge <2 x i32> [[WIDE_LOAD]], zeroinitializer
393+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i32, ptr [[TMP0]], i32 0
394+
; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i32>, ptr [[TMP4]], align 4
395+
; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP3]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[WIDE_LOAD1]]
396+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
397+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[TMP5]], i32 0
398+
; CHECK-NEXT: store <2 x i32> [[PREDPHI]], ptr [[TMP6]], align 1
399+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
400+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
401+
; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
402+
; CHECK: [[MIDDLE_BLOCK]]:
403+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
404+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
405+
; CHECK: [[SCALAR_PH]]:
406+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
407+
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
408+
; CHECK: [[LOOP_HEADER]]:
409+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
410+
; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]]
411+
; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV]]
412+
; CHECK-NEXT: [[L_B:%.*]] = load i32, ptr [[GEP_B]], align 4
413+
; CHECK-NEXT: [[C_1:%.*]] = icmp sge i32 [[L_B]], 0
414+
; CHECK-NEXT: br i1 [[C_1]], label %[[LOOP_LATCH]], label %[[LOOP_THEN:.*]]
415+
; CHECK: [[LOOP_THEN]]:
416+
; CHECK-NEXT: [[L_A:%.*]] = load i32, ptr [[GEP_A]], align 4
417+
; CHECK-NEXT: br label %[[LOOP_LATCH]]
418+
; CHECK: [[LOOP_LATCH]]:
419+
; CHECK-NEXT: [[MERGE:%.*]] = phi i32 [ [[L_A]], %[[LOOP_THEN]] ], [ [[L_B]], %[[LOOP_HEADER]] ]
420+
; CHECK-NEXT: [[GEP_C:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IV]]
421+
; CHECK-NEXT: store i32 [[MERGE]], ptr [[GEP_C]], align 1
422+
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
423+
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
424+
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP11:![0-9]+]]
425+
; CHECK: [[EXIT]]:
426+
; CHECK-NEXT: ret void
427+
;
428+
entry:
429+
%mul = mul nsw nuw i64 %n, 4
430+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %a, i64 %mul) ]
431+
br label %loop.header
432+
433+
loop.header:
434+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
435+
%gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
436+
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
437+
%l.b = load i32, ptr %gep.b, align 4
438+
%c.1 = icmp sge i32 %l.b, 0
439+
br i1 %c.1, label %loop.latch, label %loop.then
440+
441+
loop.then:
442+
%l.a = load i32, ptr %gep.a, align 4
443+
br label %loop.latch
444+
445+
loop.latch:
446+
%merge = phi i32 [ %l.a, %loop.then ], [ %l.b, %loop.header ]
447+
%gep.c = getelementptr inbounds i32, ptr %c, i64 %iv
448+
store i32 %merge, ptr %gep.c, align 1
449+
%iv.next = add nuw nsw i64 %iv, 1
450+
%ec = icmp eq i64 %iv.next, %n
451+
br i1 %ec, label %exit, label %loop.header
452+
453+
exit:
454+
ret void
455+
}
456+
457+
; Alignment via argument attribute is too small (1 but needs 4).
458+
define void @deref_assumption_in_preheader_non_constant_trip_count_access_i32_align_attribute_too_small(ptr noalias noundef align 1 %a, ptr noalias %b, ptr noalias %c, i64 %n) nofree nosync {
459+
; CHECK-LABEL: define void @deref_assumption_in_preheader_non_constant_trip_count_access_i32_align_attribute_too_small(
460+
; CHECK-SAME: ptr noalias noundef align 1 [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]], i64 [[N:%.*]]) #[[ATTR1]] {
461+
; CHECK-NEXT: [[ENTRY:.*]]:
462+
; CHECK-NEXT: [[MUL:%.*]] = mul nuw nsw i64 [[N]], 4
463+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[A]], i64 [[MUL]]) ]
464+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2
465+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
466+
; CHECK: [[VECTOR_PH]]:
467+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2
468+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
469+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
470+
; CHECK: [[VECTOR_BODY]]:
471+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
472+
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]]
473+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[TMP0]], i32 0
474+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4
475+
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <2 x i32> [[WIDE_LOAD]], zeroinitializer
476+
; CHECK-NEXT: [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], splat (i1 true)
477+
; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i1> [[TMP3]], i32 0
478+
; CHECK-NEXT: br i1 [[TMP4]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
479+
; CHECK: [[PRED_LOAD_IF]]:
480+
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 0
481+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP5]]
482+
; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4
483+
; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> poison, i32 [[TMP7]], i32 0
484+
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE]]
485+
; CHECK: [[PRED_LOAD_CONTINUE]]:
486+
; CHECK-NEXT: [[TMP9:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP8]], %[[PRED_LOAD_IF]] ]
487+
; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP3]], i32 1
488+
; CHECK-NEXT: br i1 [[TMP10]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
489+
; CHECK: [[PRED_LOAD_IF1]]:
490+
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 1
491+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP11]]
492+
; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
493+
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x i32> [[TMP9]], i32 [[TMP13]], i32 1
494+
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE2]]
495+
; CHECK: [[PRED_LOAD_CONTINUE2]]:
496+
; CHECK-NEXT: [[TMP15:%.*]] = phi <2 x i32> [ [[TMP9]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP14]], %[[PRED_LOAD_IF1]] ]
497+
; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP15]]
498+
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
499+
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[TMP16]], i32 0
500+
; CHECK-NEXT: store <2 x i32> [[PREDPHI]], ptr [[TMP17]], align 1
501+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
502+
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
503+
; CHECK-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
504+
; CHECK: [[MIDDLE_BLOCK]]:
505+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
506+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
507+
; CHECK: [[SCALAR_PH]]:
508+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
509+
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
510+
; CHECK: [[LOOP_HEADER]]:
511+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
512+
; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IV]]
513+
; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IV]]
514+
; CHECK-NEXT: [[L_B:%.*]] = load i32, ptr [[GEP_B]], align 4
515+
; CHECK-NEXT: [[C_1:%.*]] = icmp sge i32 [[L_B]], 0
516+
; CHECK-NEXT: br i1 [[C_1]], label %[[LOOP_LATCH]], label %[[LOOP_THEN:.*]]
517+
; CHECK: [[LOOP_THEN]]:
518+
; CHECK-NEXT: [[L_A:%.*]] = load i32, ptr [[GEP_A]], align 4
519+
; CHECK-NEXT: br label %[[LOOP_LATCH]]
520+
; CHECK: [[LOOP_LATCH]]:
521+
; CHECK-NEXT: [[MERGE:%.*]] = phi i32 [ [[L_A]], %[[LOOP_THEN]] ], [ [[L_B]], %[[LOOP_HEADER]] ]
522+
; CHECK-NEXT: [[GEP_C:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IV]]
523+
; CHECK-NEXT: store i32 [[MERGE]], ptr [[GEP_C]], align 1
524+
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
525+
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
526+
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP13:![0-9]+]]
527+
; CHECK: [[EXIT]]:
528+
; CHECK-NEXT: ret void
529+
;
530+
entry:
531+
%mul = mul nsw nuw i64 %n, 4
532+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %a, i64 %mul) ]
533+
br label %loop.header
534+
535+
loop.header:
536+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
537+
%gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
538+
%gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
539+
%l.b = load i32, ptr %gep.b, align 4
540+
%c.1 = icmp sge i32 %l.b, 0
541+
br i1 %c.1, label %loop.latch, label %loop.then
542+
543+
loop.then:
544+
%l.a = load i32, ptr %gep.a, align 4
545+
br label %loop.latch
546+
547+
loop.latch:
548+
%merge = phi i32 [ %l.a, %loop.then ], [ %l.b, %loop.header ]
549+
%gep.c = getelementptr inbounds i32, ptr %c, i64 %iv
550+
store i32 %merge, ptr %gep.c, align 1
551+
%iv.next = add nuw nsw i64 %iv, 1
552+
%ec = icmp eq i64 %iv.next, %n
553+
br i1 %ec, label %exit, label %loop.header
554+
555+
exit:
556+
ret void
557+
}

0 commit comments

Comments
 (0)