5d624b5b93
In outer-loop VPlan, avoid emitting vector intrinsic calls for intrinsics without a vector form. In VPRecipeBuilder, detect missing vector intrinsic mapping and emit scalar handling instead of a vector call. Also fix assertion when `llvm.pseudoprobe` in VPlan's native path is being treated as a `WIDEN-INTRINSIC`. Reproducer: https://godbolt.org/z/GsPYobvYs
74 lines
3.2 KiB
LLVM
74 lines
3.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 6
|
|
; RUN: opt -S -passes='loop-vectorize,verify' -enable-vplan-native-path -force-vector-width=2 < %s | FileCheck %s
|
|
|
|
; The noalias.scope.decl intrinsic declares a noalias scope valid for a single
|
|
; iteration. During outer loop vectorization, emitting it as a single-scalar
|
|
; replicate would incorrectly extend the scope across multiple original
|
|
; iterations packed into one vector iteration. Bail out of vectorization for
|
|
; outer loops containing noalias.scope.decl.
|
|
; FIXME: We could still vectorize by dropping the noalias.scope.decl and
|
|
; stripping !alias.scope/!noalias metadata from the loop body.
|
|
|
|
define void @test_noalias_scope_decl(ptr %a, ptr %b, i64 %n) {
|
|
; CHECK-LABEL: define void @test_noalias_scope_decl(
|
|
; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) {
|
|
; CHECK-NEXT: [[SCALAR_PH:.*]]:
|
|
; CHECK-NEXT: br label %[[OUTER_HEADER:.*]]
|
|
; CHECK: [[OUTER_HEADER]]:
|
|
; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[J_NEXT:%.*]], %[[OUTER_LATCH:.*]] ]
|
|
; CHECK-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META0:![0-9]+]])
|
|
; CHECK-NEXT: br label %[[INNER:.*]]
|
|
; CHECK: [[INNER]]:
|
|
; CHECK-NEXT: [[K:%.*]] = phi i64 [ 0, %[[OUTER_HEADER]] ], [ [[K_NEXT:%.*]], %[[INNER]] ]
|
|
; CHECK-NEXT: [[IDX:%.*]] = add i64 [[J]], [[K]]
|
|
; CHECK-NEXT: [[PTR_A:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDX]]
|
|
; CHECK-NEXT: [[VAL:%.*]] = load i32, ptr [[PTR_A]], align 4, !alias.scope [[META0]]
|
|
; CHECK-NEXT: [[PTR_B:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IDX]]
|
|
; CHECK-NEXT: store i32 [[VAL]], ptr [[PTR_B]], align 4, !noalias [[META0]]
|
|
; CHECK-NEXT: [[K_NEXT]] = add i64 [[K]], 1
|
|
; CHECK-NEXT: [[INNER_COND:%.*]] = icmp eq i64 [[K_NEXT]], [[N]]
|
|
; CHECK-NEXT: br i1 [[INNER_COND]], label %[[OUTER_LATCH]], label %[[INNER]]
|
|
; CHECK: [[OUTER_LATCH]]:
|
|
; CHECK-NEXT: [[J_NEXT]] = add i64 [[J]], 1
|
|
; CHECK-NEXT: [[OUTER_COND:%.*]] = icmp eq i64 [[J_NEXT]], [[N]]
|
|
; CHECK-NEXT: br i1 [[OUTER_COND]], label %[[EXIT:.*]], label %[[OUTER_HEADER]], !llvm.loop [[LOOP3:![0-9]+]]
|
|
; CHECK: [[EXIT]]:
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
entry:
|
|
br label %outer.header
|
|
|
|
outer.header:
|
|
%j = phi i64 [ 0, %entry ], [ %j.next, %outer.latch ]
|
|
call void @llvm.experimental.noalias.scope.decl(metadata !3)
|
|
br label %inner
|
|
|
|
inner:
|
|
%k = phi i64 [ 0, %outer.header ], [ %k.next, %inner ]
|
|
%idx = add i64 %j, %k
|
|
%ptr.a = getelementptr inbounds i32, ptr %a, i64 %idx
|
|
%val = load i32, ptr %ptr.a, align 4, !alias.scope !3
|
|
%ptr.b = getelementptr inbounds i32, ptr %b, i64 %idx
|
|
store i32 %val, ptr %ptr.b, align 4, !noalias !3
|
|
%k.next = add i64 %k, 1
|
|
%inner.cond = icmp eq i64 %k.next, %n
|
|
br i1 %inner.cond, label %outer.latch, label %inner
|
|
|
|
outer.latch:
|
|
%j.next = add i64 %j, 1
|
|
%outer.cond = icmp eq i64 %j.next, %n
|
|
br i1 %outer.cond, label %exit, label %outer.header, !llvm.loop !0
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
declare void @llvm.experimental.noalias.scope.decl(metadata)
|
|
|
|
!0 = distinct !{!0, !1, !2}
|
|
!1 = !{!"llvm.loop.mustprogress"}
|
|
!2 = !{!"llvm.loop.vectorize.enable", i1 true}
|
|
!3 = !{!4}
|
|
!4 = distinct !{!4, !5}
|
|
!5 = distinct !{!5}
|