71 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 52; RUN: opt -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck %s3 4define i8 @pr87407(i8 %x, i64 %y, i64 %n) {5; CHECK-LABEL: define i8 @pr87407(6; CHECK-SAME: i8 [[X:%.*]], i64 [[Y:%.*]], i64 [[N:%.*]]) {7; CHECK-NEXT: [[ENTRY:.*]]:8; CHECK-NEXT: [[ZEXT_X:%.*]] = zext i8 [[X]] to i649; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 410; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]11; CHECK: [[VECTOR_PH]]:12; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 413; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]14; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[Y]], i64 015; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer16; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i64> poison, i64 [[ZEXT_X]], i64 017; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT1]], <4 x i64> poison, <4 x i32> zeroinitializer18; CHECK-NEXT: [[TMP0:%.*]] = call <4 x i64> @llvm.umax.v4i64(<4 x i64> [[BROADCAST_SPLAT2]], <4 x i64> [[BROADCAST_SPLAT]])19; CHECK-NEXT: [[TMP1:%.*]] = trunc <4 x i64> [[TMP0]] to <4 x i1>20; CHECK-NEXT: [[TMP2:%.*]] = icmp ne <4 x i1> [[TMP1]], zeroinitializer21; CHECK-NEXT: [[TMP3:%.*]] = zext <4 x i1> [[TMP2]] to <4 x i32>22; CHECK-NEXT: [[TMP4:%.*]] = shl <4 x i32> [[TMP3]], splat (i32 8)23; CHECK-NEXT: [[TMP5:%.*]] = trunc <4 x i32> [[TMP4]] to <4 x i8>24; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]25; CHECK: [[VECTOR_BODY]]:26; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]27; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 428; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]29; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]30; CHECK: [[MIDDLE_BLOCK]]:31; CHECK-NEXT: [[TMP7:%.*]] = extractelement <4 x i8> [[TMP5]], i32 332; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]33; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]34; CHECK: [[SCALAR_PH]]:35; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]36; CHECK-NEXT: br label %[[LOOP:.*]]37; CHECK: [[LOOP]]:38; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]39; CHECK-NEXT: [[MAX:%.*]] = tail call i64 @llvm.umax.i64(i64 [[ZEXT_X]], i64 [[Y]])40; CHECK-NEXT: [[CMP_MAX_0:%.*]] = icmp ne i64 [[MAX]], 041; CHECK-NEXT: [[ZEXT_CMP:%.*]] = zext i1 [[CMP_MAX_0]] to i6442; CHECK-NEXT: [[TRUNC:%.*]] = trunc i64 [[ZEXT_CMP]] to i3243; CHECK-NEXT: [[SHL:%.*]] = shl i32 [[TRUNC]], 844; CHECK-NEXT: [[RES:%.*]] = trunc i32 [[SHL]] to i845; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 146; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp ne i64 [[IV_NEXT]], [[N]]47; CHECK-NEXT: br i1 [[EXIT_COND]], label %[[LOOP]], label %[[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]48; CHECK: [[EXIT]]:49; CHECK-NEXT: [[RES_LCSSA:%.*]] = phi i8 [ [[RES]], %[[LOOP]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ]50; CHECK-NEXT: ret i8 [[RES_LCSSA]]51;52entry:53 %zext.x = zext i8 %x to i6454 br label %loop55 56loop:57 %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]58 %max = tail call i64 @llvm.umax.i64(i64 %zext.x, i64 %y)59 %cmp.max.0 = icmp ne i64 %max, 060 %zext.cmp = zext i1 %cmp.max.0 to i6461 %trunc = trunc i64 %zext.cmp to i3262 %shl = shl i32 %trunc, 863 %res = trunc i32 %shl to i864 %iv.next = add i64 %iv, 165 %exit.cond = icmp ne i64 %iv.next, %n66 br i1 %exit.cond, label %loop, label %exit67 68exit:69 ret i8 %res70}71