56 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=slp-vectorizer -mattr=+avx2 -S | FileCheck %s3 4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"5target triple = "x86_64-unknown-linux-gnu"6 7; In this test case we start trying to vectorize reduction but end up8; in tryToVectorize() method which then can make several attempts to9; find a pair (roots) for a tree that can be vectorized.10; The order (path) it makes probes for various pairs is predefined by11; the method implementation and it is not guaranteed that the best option12; encountered first (like here).13 14define double @root_selection(double %a, double %b, double %c, double %d) local_unnamed_addr {15; CHECK-LABEL: @root_selection(16; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x double> poison, double [[B:%.*]], i32 017; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double [[A:%.*]], i32 118; CHECK-NEXT: [[TMP3:%.*]] = fdiv fast <2 x double> [[TMP2]], <double 5.000000e+00, double 7.000000e+00>19; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 020; CHECK-NEXT: [[I09:%.*]] = fmul fast double [[TMP4]], undef21; CHECK-NEXT: [[I10:%.*]] = fsub fast double undef, [[I09]]22; CHECK-NEXT: [[TMP5:%.*]] = fmul fast <2 x double> [[TMP3]], <double undef, double 3.000000e+00>23; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x double> <double poison, double undef>, double [[I10]], i32 024; CHECK-NEXT: [[TMP7:%.*]] = fmul fast <2 x double> [[TMP6]], [[TMP5]]25; CHECK-NEXT: [[TMP8:%.*]] = fsub fast <2 x double> [[TMP7]], <double 1.100000e+01, double undef>26; CHECK-NEXT: [[TMP9:%.*]] = fmul fast <2 x double> [[TMP8]], <double 1.200000e+01, double 4.000000e+00>27; CHECK-NEXT: [[TMP10:%.*]] = fdiv fast <2 x double> [[TMP9]], splat (double 1.400000e+00)28; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x double> [[TMP10]], i32 129; CHECK-NEXT: [[I07:%.*]] = fadd fast double undef, [[TMP11]]30; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[TMP10]], i32 031; CHECK-NEXT: [[I16:%.*]] = fadd fast double [[I07]], [[TMP12]]32; CHECK-NEXT: [[I17:%.*]] = fadd fast double [[I16]], [[C:%.*]]33; CHECK-NEXT: [[I18:%.*]] = fadd fast double [[I17]], [[D:%.*]]34; CHECK-NEXT: ret double [[I18]]35;36 %i01 = fdiv fast double %a, 7.037 %i02 = fmul fast double %i01, 3.038 %i03 = fmul fast double undef, %i0239 %i04 = fsub fast double %i03, undef40 %i05 = fmul fast double %i04, 4.041 %i06 = fdiv fast double %i05, 1.442 %i07 = fadd fast double undef, %i0643 %i08 = fdiv fast double %b, 5.044 %i09 = fmul fast double %i08, undef45 %i10 = fsub fast double undef, %i0946 %i11 = fmul fast double %i08, undef47 %i12 = fmul fast double %i10, %i1148 %i13 = fsub fast double %i12, 11.049 %i14 = fmul fast double %i13, 12.050 %i15 = fdiv fast double %i14, 1.451 %i16 = fadd fast double %i07, %i1552 %i17 = fadd fast double %i16, %c53 %i18 = fadd fast double %i17, %d54 ret double %i1855}56