85 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 52; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S | FileCheck %s -check-prefix=NO-ZVFHMIN3; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFHMIN4; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFHMIN5; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v,+zvfhmin -S | FileCheck %s -check-prefix=ZVFHMIN6 7define void @fadd(ptr noalias %a, ptr noalias %b, i64 %n) {8; NO-ZVFHMIN-LABEL: define void @fadd(9; NO-ZVFHMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {10; NO-ZVFHMIN-NEXT: [[ENTRY:.*]]:11; NO-ZVFHMIN-NEXT: br label %[[LOOP:.*]]12; NO-ZVFHMIN: [[LOOP]]:13; NO-ZVFHMIN-NEXT: [[I:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]14; NO-ZVFHMIN-NEXT: [[A_GEP:%.*]] = getelementptr half, ptr [[A]], i64 [[I]]15; NO-ZVFHMIN-NEXT: [[B_GEP:%.*]] = getelementptr half, ptr [[B]], i64 [[I]]16; NO-ZVFHMIN-NEXT: [[X:%.*]] = load half, ptr [[A_GEP]], align 217; NO-ZVFHMIN-NEXT: [[Y:%.*]] = load half, ptr [[B_GEP]], align 218; NO-ZVFHMIN-NEXT: [[Z:%.*]] = fadd half [[X]], [[Y]]19; NO-ZVFHMIN-NEXT: store half [[Z]], ptr [[A_GEP]], align 220; NO-ZVFHMIN-NEXT: [[I_NEXT]] = add i64 [[I]], 121; NO-ZVFHMIN-NEXT: [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]22; NO-ZVFHMIN-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]]23; NO-ZVFHMIN: [[EXIT]]:24; NO-ZVFHMIN-NEXT: ret void25;26; ZVFHMIN-LABEL: define void @fadd(27; ZVFHMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {28; ZVFHMIN-NEXT: [[ENTRY:.*:]]29; ZVFHMIN-NEXT: br label %[[VECTOR_PH:.*]]30; ZVFHMIN: [[VECTOR_PH]]:31; ZVFHMIN-NEXT: br label %[[VECTOR_BODY:.*]]32; ZVFHMIN: [[VECTOR_BODY]]:33; ZVFHMIN-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ]34; ZVFHMIN-NEXT: [[AVL:%.*]] = phi i64 [ [[N]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]35; ZVFHMIN-NEXT: [[TMP6:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 8, i1 true)36; ZVFHMIN-NEXT: [[TMP1:%.*]] = getelementptr half, ptr [[A]], i64 [[INDEX]]37; ZVFHMIN-NEXT: [[TMP2:%.*]] = getelementptr half, ptr [[B]], i64 [[INDEX]]38; ZVFHMIN-NEXT: [[WIDE_LOAD:%.*]] = call <vscale x 8 x half> @llvm.vp.load.nxv8f16.p0(ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])39; ZVFHMIN-NEXT: [[WIDE_LOAD1:%.*]] = call <vscale x 8 x half> @llvm.vp.load.nxv8f16.p0(ptr align 2 [[TMP2]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])40; ZVFHMIN-NEXT: [[TMP11:%.*]] = fadd <vscale x 8 x half> [[WIDE_LOAD]], [[WIDE_LOAD1]]41; ZVFHMIN-NEXT: call void @llvm.vp.store.nxv8f16.p0(<vscale x 8 x half> [[TMP11]], ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])42; ZVFHMIN-NEXT: [[TMP13:%.*]] = zext i32 [[TMP6]] to i6443; ZVFHMIN-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP13]], [[INDEX]]44; ZVFHMIN-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP13]]45; ZVFHMIN-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 046; ZVFHMIN-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]47; ZVFHMIN: [[MIDDLE_BLOCK]]:48; ZVFHMIN-NEXT: br label %[[EXIT:.*]]49; ZVFHMIN: [[EXIT]]:50; ZVFHMIN-NEXT: ret void51;52; NO-ZVFHMIN-PREDICATED-LABEL: define void @fadd(53; NO-ZVFHMIN-PREDICATED-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {54; NO-ZVFHMIN-PREDICATED-NEXT: [[ENTRY:.*]]:55; NO-ZVFHMIN-PREDICATED-NEXT: br label %[[LOOP:.*]]56; NO-ZVFHMIN-PREDICATED: [[LOOP]]:57; NO-ZVFHMIN-PREDICATED-NEXT: [[I:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]58; NO-ZVFHMIN-PREDICATED-NEXT: [[A_GEP:%.*]] = getelementptr half, ptr [[A]], i64 [[I]]59; NO-ZVFHMIN-PREDICATED-NEXT: [[B_GEP:%.*]] = getelementptr half, ptr [[B]], i64 [[I]]60; NO-ZVFHMIN-PREDICATED-NEXT: [[X:%.*]] = load half, ptr [[A_GEP]], align 261; NO-ZVFHMIN-PREDICATED-NEXT: [[Y:%.*]] = load half, ptr [[B_GEP]], align 262; NO-ZVFHMIN-PREDICATED-NEXT: [[Z:%.*]] = fadd half [[X]], [[Y]]63; NO-ZVFHMIN-PREDICATED-NEXT: store half [[Z]], ptr [[A_GEP]], align 264; NO-ZVFHMIN-PREDICATED-NEXT: [[I_NEXT]] = add i64 [[I]], 165; NO-ZVFHMIN-PREDICATED-NEXT: [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]66; NO-ZVFHMIN-PREDICATED-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]]67; NO-ZVFHMIN-PREDICATED: [[EXIT]]:68; NO-ZVFHMIN-PREDICATED-NEXT: ret void69entry:70 br label %loop71loop:72 %i = phi i64 [0, %entry], [%i.next, %loop]73 %a.gep = getelementptr half, ptr %a, i64 %i74 %b.gep = getelementptr half, ptr %b, i64 %i75 %x = load half, ptr %a.gep76 %y = load half, ptr %b.gep77 %z = fadd half %x, %y78 store half %z, ptr %a.gep79 %i.next = add i64 %i, 180 %done = icmp eq i64 %i.next, %n81 br i1 %done, label %exit, label %loop82exit:83 ret void84}85