blob: f9798922f2697ad71ee4c111f53af09cdab8dac8 [file] [edit]
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph" --version 6
; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s
; A byte load does not make a wider load at the same pointer safe.
define void @wider_predicated_load(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @wider_predicated_load(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE6:.*]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i64, ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP8:%.*]] = load i8, ptr [[TMP4]], align 1
; CHECK-NEXT: [[TMP9:%.*]] = load i8, ptr [[TMP5]], align 1
; CHECK-NEXT: [[TMP10:%.*]] = load i8, ptr [[TMP6]], align 1
; CHECK-NEXT: [[TMP11:%.*]] = load i8, ptr [[TMP7]], align 1
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i8> poison, i8 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i8> [[TMP12]], i8 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i8> [[TMP14]], i8 [[TMP11]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i8> [[TMP15]], zeroinitializer
; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP16]], i64 0
; CHECK-NEXT: br i1 [[TMP17]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
; CHECK: [[PRED_LOAD_IF]]:
; CHECK-NEXT: [[TMP18:%.*]] = load i64, ptr [[TMP4]], align 1
; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i64> poison, i64 [[TMP18]], i64 0
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE]]
; CHECK: [[PRED_LOAD_CONTINUE]]:
; CHECK-NEXT: [[TMP20:%.*]] = phi <4 x i64> [ poison, %[[VECTOR_BODY]] ], [ [[TMP19]], %[[PRED_LOAD_IF]] ]
; CHECK-NEXT: [[TMP21:%.*]] = extractelement <4 x i1> [[TMP16]], i64 1
; CHECK-NEXT: br i1 [[TMP21]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2:.*]]
; CHECK: [[PRED_LOAD_IF1]]:
; CHECK-NEXT: [[TMP22:%.*]] = load i64, ptr [[TMP5]], align 1
; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i64> [[TMP20]], i64 [[TMP22]], i64 1
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE2]]
; CHECK: [[PRED_LOAD_CONTINUE2]]:
; CHECK-NEXT: [[TMP24:%.*]] = phi <4 x i64> [ [[TMP20]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP23]], %[[PRED_LOAD_IF1]] ]
; CHECK-NEXT: [[TMP25:%.*]] = extractelement <4 x i1> [[TMP16]], i64 2
; CHECK-NEXT: br i1 [[TMP25]], label %[[PRED_LOAD_IF3:.*]], label %[[PRED_LOAD_CONTINUE4:.*]]
; CHECK: [[PRED_LOAD_IF3]]:
; CHECK-NEXT: [[TMP26:%.*]] = load i64, ptr [[TMP6]], align 1
; CHECK-NEXT: [[TMP27:%.*]] = insertelement <4 x i64> [[TMP24]], i64 [[TMP26]], i64 2
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE4]]
; CHECK: [[PRED_LOAD_CONTINUE4]]:
; CHECK-NEXT: [[TMP28:%.*]] = phi <4 x i64> [ [[TMP24]], %[[PRED_LOAD_CONTINUE2]] ], [ [[TMP27]], %[[PRED_LOAD_IF3]] ]
; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i1> [[TMP16]], i64 3
; CHECK-NEXT: br i1 [[TMP29]], label %[[PRED_LOAD_IF5:.*]], label %[[PRED_LOAD_CONTINUE6]]
; CHECK: [[PRED_LOAD_IF5]]:
; CHECK-NEXT: [[TMP30:%.*]] = load i64, ptr [[TMP7]], align 1
; CHECK-NEXT: [[TMP31:%.*]] = insertelement <4 x i64> [[TMP28]], i64 [[TMP30]], i64 3
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE6]]
; CHECK: [[PRED_LOAD_CONTINUE6]]:
; CHECK-NEXT: [[TMP32:%.*]] = phi <4 x i64> [ [[TMP28]], %[[PRED_LOAD_CONTINUE4]] ], [ [[TMP31]], %[[PRED_LOAD_IF5]] ]
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x i64> [[TMP32]], <4 x i64> zeroinitializer
; CHECK-NEXT: [[TMP33:%.*]] = getelementptr i64, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x i64> [[PREDPHI]], ptr [[TMP33]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP34]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr i64, ptr %src, i64 %iv
%guard = load i8, ptr %p, align 1
%keep = icmp ne i8 %guard, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load i64, ptr %p, align 1
br label %loop.latch
loop.latch:
%result = phi i64 [ %value, %read ], [ 0, %loop.header ]
%out = getelementptr i64, ptr %dst, i64 %iv
store i64 %result, ptr %out, align 1
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; The same number of accessible bytes permits a load with a different type.
define void @same_width_different_type(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @same_width_different_type(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i32> [[TMP12]], i32 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i32> [[TMP13]], i32 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP11]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
; CHECK-NEXT: [[TMP17:%.*]] = load float, ptr [[TMP4]], align 4
; CHECK-NEXT: [[TMP18:%.*]] = load float, ptr [[TMP5]], align 4
; CHECK-NEXT: [[TMP19:%.*]] = load float, ptr [[TMP6]], align 4
; CHECK-NEXT: [[TMP20:%.*]] = load float, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x float> poison, float [[TMP17]], i64 0
; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x float> [[TMP21]], float [[TMP18]], i64 1
; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x float> [[TMP22]], float [[TMP19]], i64 2
; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x float> [[TMP23]], float [[TMP20]], i64 3
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x float> [[TMP24]], <4 x float> zeroinitializer
; CHECK-NEXT: [[TMP25:%.*]] = getelementptr float, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x float> [[PREDPHI]], ptr [[TMP25]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP26]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr [4 x i32], ptr %src, i64 %iv
%guard = load i32, ptr %p, align 4
%keep = icmp ne i32 %guard, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load float, ptr %p, align 4
br label %loop.latch
loop.latch:
%result = phi float [ %value, %read ], [ 0.000000e+00, %loop.header ]
%out = getelementptr float, ptr %dst, i64 %iv
store float %result, ptr %out, align 4
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; A wider access also makes a narrower access at the same pointer safe.
define void @narrower_predicated_load(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @narrower_predicated_load(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP8:%.*]] = load i64, ptr [[TMP4]], align 8
; CHECK-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP5]], align 8
; CHECK-NEXT: [[TMP10:%.*]] = load i64, ptr [[TMP6]], align 8
; CHECK-NEXT: [[TMP11:%.*]] = load i64, ptr [[TMP7]], align 8
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i64> poison, i64 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i64> [[TMP12]], i64 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i64> [[TMP13]], i64 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i64> [[TMP14]], i64 [[TMP11]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i64> [[TMP15]], zeroinitializer
; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i32> poison, i32 [[TMP17]], i64 0
; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i32> [[TMP21]], i32 [[TMP18]], i64 1
; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> [[TMP22]], i32 [[TMP19]], i64 2
; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i64 3
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x i32> [[TMP24]], <4 x i32> zeroinitializer
; CHECK-NEXT: [[TMP25:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x i32> [[PREDPHI]], ptr [[TMP25]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP26]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr [2 x i64], ptr %src, i64 %iv
%guard = load i64, ptr %p, align 8
%keep = icmp ne i64 %guard, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load i32, ptr %p, align 4
br label %loop.latch
loop.latch:
%result = phi i32 [ %value, %read ], [ 0, %loop.header ]
%out = getelementptr i32, ptr %dst, i64 %iv
store i32 %result, ptr %out, align 4
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; Size and alignment can be established by separate unconditional accesses.
define void @separate_size_and_alignment(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @separate_size_and_alignment(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr [2 x i64], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP8:%.*]] = load i64, ptr [[TMP4]], align 1
; CHECK-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP5]], align 1
; CHECK-NEXT: [[TMP10:%.*]] = load i64, ptr [[TMP6]], align 1
; CHECK-NEXT: [[TMP11:%.*]] = load i64, ptr [[TMP7]], align 1
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i64> poison, i64 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i64> [[TMP12]], i64 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i64> [[TMP13]], i64 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i64> [[TMP14]], i64 [[TMP11]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i32> poison, i32 [[TMP16]], i64 0
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i32> [[TMP20]], i32 [[TMP17]], i64 1
; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i32> [[TMP21]], i32 [[TMP18]], i64 2
; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> [[TMP22]], i32 [[TMP19]], i64 3
; CHECK-NEXT: [[TMP24:%.*]] = trunc <4 x i64> [[TMP15]] to <4 x i32>
; CHECK-NEXT: [[TMP25:%.*]] = add <4 x i32> [[TMP24]], [[TMP23]]
; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <4 x i32> [[TMP25]], zeroinitializer
; CHECK-NEXT: [[TMP27:%.*]] = load i64, ptr [[TMP4]], align 16
; CHECK-NEXT: [[TMP28:%.*]] = load i64, ptr [[TMP5]], align 16
; CHECK-NEXT: [[TMP29:%.*]] = load i64, ptr [[TMP6]], align 16
; CHECK-NEXT: [[TMP30:%.*]] = load i64, ptr [[TMP7]], align 16
; CHECK-NEXT: [[TMP31:%.*]] = insertelement <4 x i64> poison, i64 [[TMP27]], i64 0
; CHECK-NEXT: [[TMP32:%.*]] = insertelement <4 x i64> [[TMP31]], i64 [[TMP28]], i64 1
; CHECK-NEXT: [[TMP33:%.*]] = insertelement <4 x i64> [[TMP32]], i64 [[TMP29]], i64 2
; CHECK-NEXT: [[TMP34:%.*]] = insertelement <4 x i64> [[TMP33]], i64 [[TMP30]], i64 3
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP26]], <4 x i64> [[TMP34]], <4 x i64> zeroinitializer
; CHECK-NEXT: [[TMP35:%.*]] = getelementptr i64, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x i64> [[PREDPHI]], ptr [[TMP35]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP36]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr [2 x i64], ptr %src, i64 %iv
%wide = load i64, ptr %p, align 1
%aligned = load i32, ptr %p, align 16
%lower = trunc i64 %wide to i32
%sum = add i32 %lower, %aligned
%keep = icmp ne i32 %sum, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load i64, ptr %p, align 16
br label %loop.latch
loop.latch:
%result = phi i64 [ %value, %read ], [ 0, %loop.header ]
%out = getelementptr i64, ptr %dst, i64 %iv
store i64 %result, ptr %out, align 8
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; A weaker alignment does not justify the predicated load alignment.
define void @stronger_predicated_alignment(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @stronger_predicated_alignment(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE6:.*]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i32> [[TMP12]], i32 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i32> [[TMP13]], i32 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP11]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP16]], i64 0
; CHECK-NEXT: br i1 [[TMP17]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
; CHECK: [[PRED_LOAD_IF]]:
; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i32> poison, i32 [[TMP18]], i64 0
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE]]
; CHECK: [[PRED_LOAD_CONTINUE]]:
; CHECK-NEXT: [[TMP20:%.*]] = phi <4 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP21]], %[[PRED_LOAD_IF]] ]
; CHECK-NEXT: [[TMP24:%.*]] = extractelement <4 x i1> [[TMP16]], i64 1
; CHECK-NEXT: br i1 [[TMP24]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2:.*]]
; CHECK: [[PRED_LOAD_IF1]]:
; CHECK-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT: [[TMP27:%.*]] = insertelement <4 x i32> [[TMP20]], i32 [[TMP22]], i64 1
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE2]]
; CHECK: [[PRED_LOAD_CONTINUE2]]:
; CHECK-NEXT: [[TMP23:%.*]] = phi <4 x i32> [ [[TMP20]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP27]], %[[PRED_LOAD_IF1]] ]
; CHECK-NEXT: [[TMP25:%.*]] = extractelement <4 x i1> [[TMP16]], i64 2
; CHECK-NEXT: br i1 [[TMP25]], label %[[PRED_LOAD_IF3:.*]], label %[[PRED_LOAD_CONTINUE4:.*]]
; CHECK: [[PRED_LOAD_IF3]]:
; CHECK-NEXT: [[TMP26:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT: [[TMP28:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP26]], i64 2
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE4]]
; CHECK: [[PRED_LOAD_CONTINUE4]]:
; CHECK-NEXT: [[TMP32:%.*]] = phi <4 x i32> [ [[TMP23]], %[[PRED_LOAD_CONTINUE2]] ], [ [[TMP28]], %[[PRED_LOAD_IF3]] ]
; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i1> [[TMP16]], i64 3
; CHECK-NEXT: br i1 [[TMP29]], label %[[PRED_LOAD_IF5:.*]], label %[[PRED_LOAD_CONTINUE6]]
; CHECK: [[PRED_LOAD_IF5]]:
; CHECK-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT: [[TMP35:%.*]] = insertelement <4 x i32> [[TMP32]], i32 [[TMP30]], i64 3
; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE6]]
; CHECK: [[PRED_LOAD_CONTINUE6]]:
; CHECK-NEXT: [[TMP31:%.*]] = phi <4 x i32> [ [[TMP32]], %[[PRED_LOAD_CONTINUE4]] ], [ [[TMP35]], %[[PRED_LOAD_IF5]] ]
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP16]], <4 x i32> [[TMP31]], <4 x i32> zeroinitializer
; CHECK-NEXT: [[TMP33:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x i32> [[PREDPHI]], ptr [[TMP33]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP34]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr [4 x i32], ptr %src, i64 %iv
%guard = load i32, ptr %p, align 4
%keep = icmp ne i32 %guard, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load i32, ptr %p, align 16
br label %loop.latch
loop.latch:
%result = phi i32 [ %value, %read ], [ 0, %loop.header ]
%out = getelementptr i32, ptr %dst, i64 %iv
store i32 %result, ptr %out, align 1
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; Keep the strongest known alignment when several accesses share a pointer.
define void @sufficient_alignment(ptr noalias %src, ptr noalias %dst, i64 %n) {
; CHECK-LABEL: define void @sufficient_alignment(
; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = and i64 [[N]], 3
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[TMP0]]
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr [4 x i32], ptr [[SRC]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP34:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP8]], i64 0
; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x i32> [[TMP12]], i32 [[TMP9]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = insertelement <4 x i32> [[TMP13]], i32 [[TMP10]], i64 2
; CHECK-NEXT: [[TMP33:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP11]], i64 3
; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP34]], align 1
; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[TMP4]], align 4
; CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP6]], align 4
; CHECK-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i32> poison, i32 [[TMP17]], i64 0
; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i32> [[TMP21]], i32 [[TMP18]], i64 1
; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> [[TMP22]], i32 [[TMP19]], i64 2
; CHECK-NEXT: [[TMP26:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i64 3
; CHECK-NEXT: [[TMP24:%.*]] = add <4 x i32> [[TMP33]], [[TMP26]]
; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <4 x i32> [[TMP24]], zeroinitializer
; CHECK-NEXT: [[TMP27:%.*]] = load i32, ptr [[TMP4]], align 16
; CHECK-NEXT: [[TMP28:%.*]] = load i32, ptr [[TMP5]], align 16
; CHECK-NEXT: [[TMP29:%.*]] = load i32, ptr [[TMP6]], align 16
; CHECK-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP7]], align 16
; CHECK-NEXT: [[TMP31:%.*]] = insertelement <4 x i32> poison, i32 [[TMP27]], i64 0
; CHECK-NEXT: [[TMP32:%.*]] = insertelement <4 x i32> [[TMP31]], i32 [[TMP28]], i64 1
; CHECK-NEXT: [[TMP36:%.*]] = insertelement <4 x i32> [[TMP32]], i32 [[TMP29]], i64 2
; CHECK-NEXT: [[TMP37:%.*]] = insertelement <4 x i32> [[TMP36]], i32 [[TMP30]], i64 3
; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP25]], <4 x i32> [[TMP37]], <4 x i32> zeroinitializer
; CHECK-NEXT: store <4 x i32> [[PREDPHI]], ptr [[TMP34]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP35]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr [4 x i32], ptr %src, i64 %iv
%out = getelementptr i32, ptr %dst, i64 %iv
%guard = load i32, ptr %p, align 16
store i32 0, ptr %out, align 1
%weak = load i32, ptr %p, align 4
%sum = add i32 %guard, %weak
%keep = icmp ne i32 %sum, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load i32, ptr %p, align 16
br label %loop.latch
loop.latch:
%result = phi i32 [ %value, %read ], [ 0, %loop.header ]
store i32 %result, ptr %out, align 1
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}
; Existing scalable accesses must not establish safety for predicated loads.
; The loop remains unvectorized because its vector-valued loads cannot be widened.
define void @scalable_accesses(ptr %src, ptr %dst, i64 %n) {
; CHECK-LABEL: define void @scalable_accesses(
; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
; CHECK: [[LOOP_HEADER]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
; CHECK-NEXT: [[P:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[SRC]], i64 [[IV]]
; CHECK-NEXT: [[OUT:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[DST]], i64 [[IV]]
; CHECK-NEXT: [[WIDE:%.*]] = load <vscale x 2 x i64>, ptr [[P]], align 8
; CHECK-NEXT: store <vscale x 2 x i64> [[WIDE]], ptr [[OUT]], align 8
; CHECK-NEXT: [[GUARD:%.*]] = load i32, ptr [[P]], align 4
; CHECK-NEXT: [[KEEP:%.*]] = icmp ne i32 [[GUARD]], 0
; CHECK-NEXT: br i1 [[KEEP]], label %[[READ:.*]], label %[[LOOP_LATCH]]
; CHECK: [[READ]]:
; CHECK-NEXT: [[VALUE:%.*]] = load <vscale x 2 x i32>, ptr [[P]], align 4
; CHECK-NEXT: store <vscale x 2 x i32> [[VALUE]], ptr [[OUT]], align 4
; CHECK-NEXT: br label %[[LOOP_LATCH]]
; CHECK: [[LOOP_LATCH]]:
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[DONE:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; CHECK-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP_HEADER]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
entry:
br label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%p = getelementptr <vscale x 2 x i64>, ptr %src, i64 %iv
%out = getelementptr <vscale x 2 x i64>, ptr %dst, i64 %iv
%wide = load <vscale x 2 x i64>, ptr %p, align 8
store <vscale x 2 x i64> %wide, ptr %out, align 8
%guard = load i32, ptr %p, align 4
%keep = icmp ne i32 %guard, 0
br i1 %keep, label %read, label %loop.latch
read:
%value = load <vscale x 2 x i32>, ptr %p, align 4
store <vscale x 2 x i32> %value, ptr %out, align 4
br label %loop.latch
loop.latch:
%iv.next = add nuw i64 %iv, 1
%done = icmp eq i64 %iv.next, %n
br i1 %done, label %exit, label %loop.header
exit:
ret void
}