62 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize -force-tail-folding-style=data -prefer-predicate-over-epilogue=predicate-dont-vectorize -force-target-supports-scalable-vectors -S < %s | FileCheck %s3 4; vscale is not guaranteed to be a power of two, so this test (which5; deliberately doesn't correspond to an in-tree backend since those6; *do* have vscale as power-of-two) exercises the code required for the7; minimum iteration check in the non-power-of-two case.8define void @foo(i32 %val, ptr dereferenceable(1024) %ptr) {9; CHECK-LABEL: @foo(10; CHECK-NEXT: entry:11; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()12; CHECK-NEXT: [[TMP7:%.*]] = shl nuw i64 [[TMP6]], 213; CHECK-NEXT: [[TMP8:%.*]] = icmp ult i64 -257, [[TMP7]]14; CHECK-NEXT: br i1 [[TMP8]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]15; CHECK: vector.ph:16; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()17; CHECK-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 418; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[TMP1]], 119; CHECK-NEXT: [[N_RND_UP:%.*]] = add i64 256, [[TMP2]]20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]21; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]22; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]23; CHECK: vector.body:24; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VECTOR_BODY]] ]25; CHECK-NEXT: [[INDEX_NEXT2]] = add i64 [[INDEX1]], [[TMP1]]26; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT2]], [[N_VEC]]27; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]28; CHECK: middle.block:29; CHECK-NEXT: br label [[WHILE_END_LOOPEXIT:%.*]]30; CHECK: scalar.ph:31; CHECK-NEXT: br label [[WHILE_BODY:%.*]]32; CHECK: while.body:33; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[WHILE_BODY]] ], [ 0, [[SCALAR_PH]] ]34; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 [[INDEX]]35; CHECK-NEXT: [[LD1:%.*]] = load i32, ptr [[GEP]], align 436; CHECK-NEXT: [[INDEX_NEXT]] = add nsw i64 [[INDEX]], 137; CHECK-NEXT: [[CMP10:%.*]] = icmp ult i64 [[INDEX_NEXT]], 25638; CHECK-NEXT: br i1 [[CMP10]], label [[WHILE_BODY]], label [[WHILE_END_LOOPEXIT]], !llvm.loop [[LOOP3:![0-9]+]]39; CHECK: while.end.loopexit:40; CHECK-NEXT: ret void41;42entry:43 br label %while.body44 45while.body: ; preds = %while.body, %entry46 %index = phi i64 [ %index.next, %while.body ], [ 0, %entry ]47 %gep = getelementptr i32, ptr %ptr, i64 %index48 %ld1 = load i32, ptr %gep, align 449 %index.next = add nsw i64 %index, 150 %cmp10 = icmp ult i64 %index.next, 25651 br i1 %cmp10, label %while.body, label %while.end.loopexit, !llvm.loop !052 53while.end.loopexit: ; preds = %while.body54 ret void55}56 57!0 = distinct !{!0, !1, !2, !3, !4}58!1 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}59!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}60!3 = !{!"llvm.loop.interleave.count", i32 1}61!4 = !{!"llvm.loop.vectorize.width", i32 4}62