88 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize \3; RUN: -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue \4; RUN: -mtriple=riscv64 -mattr=+v -S < %s | FileCheck --check-prefix=IF-EVL %s5 6; RUN: opt -passes=loop-vectorize \7; RUN: -prefer-predicate-over-epilogue=scalar-epilogue \8; RUN: -mtriple=riscv64 -mattr=+v -S < %s | FileCheck --check-prefix=NO-VP %s9 10define void @iv32(ptr noalias %a, ptr noalias %b, i32 %N) {11; IF-EVL-LABEL: @iv32(12; IF-EVL-NEXT: entry:13; IF-EVL-NEXT: br label [[ENTRY:%.*]]14; IF-EVL: vector.ph:15; IF-EVL-NEXT: br label [[FOR_BODY:%.*]]16; IF-EVL: vector.body:17; IF-EVL-NEXT: [[EVL_BASED_IV:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INDEX_EVL_NEXT:%.*]], [[FOR_BODY]] ]18; IF-EVL-NEXT: [[TMP11:%.*]] = phi i32 [ [[N:%.*]], [[ENTRY]] ], [ [[AVL_NEXT:%.*]], [[FOR_BODY]] ]19; IF-EVL-NEXT: [[TMP12:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[TMP11]], i32 4, i1 true)20; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 [[EVL_BASED_IV]]21; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP12]])22; IF-EVL-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[EVL_BASED_IV]]23; IF-EVL-NEXT: call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_OP_LOAD]], ptr align 4 [[TMP16]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP12]])24; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i32 [[TMP12]], [[EVL_BASED_IV]]25; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i32 [[TMP11]], [[TMP12]]26; IF-EVL-NEXT: [[TMP13:%.*]] = icmp eq i32 [[AVL_NEXT]], 027; IF-EVL-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]28; IF-EVL: middle.block:29; IF-EVL-NEXT: br label [[FOR_BODY1:%.*]]30; IF-EVL: for.cond.cleanup:31; IF-EVL-NEXT: ret void32;33; NO-VP-LABEL: @iv32(34; NO-VP-NEXT: entry:35; NO-VP-NEXT: [[TMP0:%.*]] = call i32 @llvm.vscale.i32()36; NO-VP-NEXT: [[TMP10:%.*]] = shl nuw i32 [[TMP0]], 237; NO-VP-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N:%.*]], [[TMP10]]38; NO-VP-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]39; NO-VP: vector.ph:40; NO-VP-NEXT: [[TMP1:%.*]] = call i32 @llvm.vscale.i32()41; NO-VP-NEXT: [[TMP11:%.*]] = mul nuw i32 [[TMP1]], 442; NO-VP-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], [[TMP11]]43; NO-VP-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]]44; NO-VP-NEXT: br label [[VECTOR_BODY:%.*]]45; NO-VP: vector.body:46; NO-VP-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]47; NO-VP-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 [[INDEX]]48; NO-VP-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP4]], align 449; NO-VP-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[INDEX]]50; NO-VP-NEXT: store <vscale x 4 x i32> [[WIDE_LOAD]], ptr [[TMP6]], align 451; NO-VP-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP11]]52; NO-VP-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]53; NO-VP-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]54; NO-VP: middle.block:55; NO-VP-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]]56; NO-VP-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]57; NO-VP: scalar.ph:58; NO-VP-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]59; NO-VP-NEXT: br label [[FOR_BODY:%.*]]60; NO-VP: for.body:61; NO-VP-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]62; NO-VP-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[IV]]63; NO-VP-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX]], align 464; NO-VP-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[IV]]65; NO-VP-NEXT: store i32 [[TMP9]], ptr [[ARRAYIDX4]], align 466; NO-VP-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 167; NO-VP-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]]68; NO-VP-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]69; NO-VP: for.cond.cleanup:70; NO-VP-NEXT: ret void71;72entry:73 br label %for.body74 75for.body:76 %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]77 %arrayidx = getelementptr inbounds i32, ptr %b, i32 %iv78 %0 = load i32, ptr %arrayidx, align 479 %arrayidx4 = getelementptr inbounds i32, ptr %a, i32 %iv80 store i32 %0, ptr %arrayidx4, align 481 %iv.next = add nuw nsw i32 %iv, 182 %exitcond.not = icmp eq i32 %iv.next, %N83 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body84 85for.cond.cleanup:86 ret void87}88