93 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "scalar.ph\:" --version 52; RUN: opt -S -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -force-target-supports-scalable-vectors=true -scalable-vectorization=on < %s | FileCheck %s --check-prefix=CHECKUF13; RUN: opt -S -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -force-target-supports-scalable-vectors=true -scalable-vectorization=on < %s | FileCheck %s --check-prefix=CHECKUF24 5; For an interleave factor of 2, vscale is scaled by 8 instead of 4 (and thus shifted left by 3 instead of 2).6; There is also the increment for the next iteration, e.g. instead of indexing IDXB, it indexes at IDXB + vscale * 4.7define void @loop(i64 %N, ptr noalias %a, ptr noalias %b) {8; CHECKUF1-LABEL: define void @loop(9; CHECKUF1-SAME: i64 [[N:%.*]], ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) {10; CHECKUF1-NEXT: [[ENTRY:.*:]]11; CHECKUF1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()12; CHECKUF1-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 213; CHECKUF1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]14; CHECKUF1-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]15; CHECKUF1: [[VECTOR_PH]]:16; CHECKUF1-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()17; CHECKUF1-NEXT: [[TMP6:%.*]] = mul nuw i64 [[TMP5]], 418; CHECKUF1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP6]]19; CHECKUF1-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]20; CHECKUF1-NEXT: br label %[[VECTOR_BODY:.*]]21; CHECKUF1: [[VECTOR_BODY]]:22; CHECKUF1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]23; CHECKUF1-NEXT: [[TMP7:%.*]] = getelementptr inbounds double, ptr [[B]], i64 [[INDEX]]24; CHECKUF1-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x double>, ptr [[TMP7]], align 825; CHECKUF1-NEXT: [[TMP8:%.*]] = fadd <vscale x 4 x double> [[WIDE_LOAD]], splat (double 1.000000e+00)26; CHECKUF1-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, ptr [[A]], i64 [[INDEX]]27; CHECKUF1-NEXT: store <vscale x 4 x double> [[TMP8]], ptr [[TMP9]], align 828; CHECKUF1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]29; CHECKUF1-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]30; CHECKUF1-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]31; CHECKUF1: [[MIDDLE_BLOCK]]:32; CHECKUF1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]33; CHECKUF1-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]34; CHECKUF1: [[SCALAR_PH]]:35;36; CHECKUF2-LABEL: define void @loop(37; CHECKUF2-SAME: i64 [[N:%.*]], ptr noalias [[A:%.*]], ptr noalias [[B:%.*]]) {38; CHECKUF2-NEXT: [[ENTRY:.*:]]39; CHECKUF2-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()40; CHECKUF2-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 341; CHECKUF2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]42; CHECKUF2-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]43; CHECKUF2: [[VECTOR_PH]]:44; CHECKUF2-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()45; CHECKUF2-NEXT: [[TMP6:%.*]] = mul nuw i64 [[TMP5]], 846; CHECKUF2-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP6]]47; CHECKUF2-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]48; CHECKUF2-NEXT: br label %[[VECTOR_BODY:.*]]49; CHECKUF2: [[VECTOR_BODY]]:50; CHECKUF2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]51; CHECKUF2-NEXT: [[TMP7:%.*]] = getelementptr inbounds double, ptr [[B]], i64 [[INDEX]]52; CHECKUF2-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()53; CHECKUF2-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP8]], 254; CHECKUF2-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, ptr [[TMP7]], i64 [[TMP16]]55; CHECKUF2-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x double>, ptr [[TMP7]], align 856; CHECKUF2-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x double>, ptr [[TMP9]], align 857; CHECKUF2-NEXT: [[TMP10:%.*]] = fadd <vscale x 4 x double> [[WIDE_LOAD]], splat (double 1.000000e+00)58; CHECKUF2-NEXT: [[TMP11:%.*]] = fadd <vscale x 4 x double> [[WIDE_LOAD3]], splat (double 1.000000e+00)59; CHECKUF2-NEXT: [[TMP12:%.*]] = getelementptr inbounds double, ptr [[A]], i64 [[INDEX]]60; CHECKUF2-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()61; CHECKUF2-NEXT: [[TMP17:%.*]] = shl nuw i64 [[TMP13]], 262; CHECKUF2-NEXT: [[TMP14:%.*]] = getelementptr inbounds double, ptr [[TMP12]], i64 [[TMP17]]63; CHECKUF2-NEXT: store <vscale x 4 x double> [[TMP10]], ptr [[TMP12]], align 864; CHECKUF2-NEXT: store <vscale x 4 x double> [[TMP11]], ptr [[TMP14]], align 865; CHECKUF2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]66; CHECKUF2-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]67; CHECKUF2-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]68; CHECKUF2: [[MIDDLE_BLOCK]]:69; CHECKUF2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]70; CHECKUF2-NEXT: br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]71; CHECKUF2: [[SCALAR_PH]]:72;73entry:74 br label %loop75 76loop:77 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]78 %arrayidx = getelementptr inbounds double, ptr %b, i64 %iv79 %0 = load double, ptr %arrayidx, align 880 %add = fadd double %0, 1.000000e+0081 %arrayidx2 = getelementptr inbounds double, ptr %a, i64 %iv82 store double %add, ptr %arrayidx2, align 883 %iv.next = add nuw nsw i64 %iv, 184 %ec = icmp eq i64 %iv.next, %N85 br i1 %ec, label %exit, label %loop, !llvm.loop !186 87exit:88 ret void89}90 91!1 = distinct !{!1, !2}92!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}93