194 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --filter "LV: Found an estimated cost of [0-9]+ for VF [0-9]+ For instruction:\s*store i16 %[0,2], ptr %[a-zA-Z0-7]+, align 2"2; RUN: opt -passes=loop-vectorize -enable-interleaved-mem-accesses -prefer-predicate-over-epilogue=predicate-dont-vectorize -S -mcpu=skx --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefix=DISABLED_MASKED_STRIDED3; RUN: opt -passes=loop-vectorize -enable-interleaved-mem-accesses -enable-masked-interleaved-mem-accesses -prefer-predicate-over-epilogue=predicate-dont-vectorize -S -mcpu=skx --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefix=ENABLED_MASKED_STRIDED4; REQUIRES: asserts5 6target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"7target triple = "x86_64-unknown-linux-gnu"8 9; (1) Interleave-group with factor 4, storing only 2 members out of the 4.10; Check that when we allow masked-memops to support interleave-group with gaps,11; the store is vectorized using a wide masked store, with a 1,1,0,0,1,1,0,0,... mask.12; Check that when we don't allow masked-memops to support interleave-group with gaps,13; the store is scalarized.14; The input IR was generated from this source:15; for(i=0;i<1024;i++){16; points[i*4] = x[i];17; points[i*4 + 1] = y[i];18; }19; (relates to the testcase in PR50566)20 21define void @test1(ptr noalias nocapture %points, ptr noalias nocapture readonly %x, ptr noalias nocapture readonly %y) {22; DISABLED_MASKED_STRIDED-LABEL: 'test1'23; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 224; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 225; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 226; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 227; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 6 for VF 2 For instruction: store i16 %0, ptr %arrayidx2, align 228; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 6 for VF 2 For instruction: store i16 %2, ptr %arrayidx7, align 229; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 13 for VF 4 For instruction: store i16 %0, ptr %arrayidx2, align 230; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 13 for VF 4 For instruction: store i16 %2, ptr %arrayidx7, align 231; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 27 for VF 8 For instruction: store i16 %0, ptr %arrayidx2, align 232; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 27 for VF 8 For instruction: store i16 %2, ptr %arrayidx7, align 233; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 55 for VF 16 For instruction: store i16 %0, ptr %arrayidx2, align 234; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 55 for VF 16 For instruction: store i16 %2, ptr %arrayidx7, align 235;36; ENABLED_MASKED_STRIDED-LABEL: 'test1'37; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 238; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 239; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 240; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 241; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 6 for VF 2 For instruction: store i16 %0, ptr %arrayidx2, align 242; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 6 for VF 2 For instruction: store i16 %2, ptr %arrayidx7, align 243; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 4 For instruction: store i16 %0, ptr %arrayidx2, align 244; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 14 for VF 4 For instruction: store i16 %2, ptr %arrayidx7, align 245; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 8 For instruction: store i16 %0, ptr %arrayidx2, align 246; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 14 for VF 8 For instruction: store i16 %2, ptr %arrayidx7, align 247; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 16 For instruction: store i16 %0, ptr %arrayidx2, align 248; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 27 for VF 16 For instruction: store i16 %2, ptr %arrayidx7, align 249;50entry:51 br label %for.body52 53for.body:54 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]55 %arrayidx = getelementptr inbounds i16, ptr %x, i64 %indvars.iv56 %0 = load i16, ptr %arrayidx, align 257 %1 = shl nuw nsw i64 %indvars.iv, 258 %arrayidx2 = getelementptr inbounds i16, ptr %points, i64 %159 store i16 %0, ptr %arrayidx2, align 260 %arrayidx4 = getelementptr inbounds i16, ptr %y, i64 %indvars.iv61 %2 = load i16, ptr %arrayidx4, align 262 %3 = or disjoint i64 %1, 163 %arrayidx7 = getelementptr inbounds i16, ptr %points, i64 %364 store i16 %2, ptr %arrayidx7, align 265 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 166 %exitcond.not = icmp eq i64 %indvars.iv.next, 102467 br i1 %exitcond.not, label %for.end, label %for.body68 69for.end:70 ret void71}72 73; (2) Same as above, but this time the gaps mask of the store is also And-ed with the74; fold-tail mask. If using masked memops to vectorize interleaved-group with gaps is75; not allowed, the store is scalarized and predicated.76; The input IR was generated from this source:77; for(i=0;i<numPoints;i++){78; points[i*4] = x[i];79; points[i*4 + 1] = y[i];80; }81 82define void @test2(ptr noalias nocapture %points, i32 %numPoints, ptr noalias nocapture readonly %x, ptr noalias nocapture readonly %y) {83; DISABLED_MASKED_STRIDED-LABEL: 'test2'84; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 285; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 286; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 287; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 288; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 8 for VF 2 For instruction: store i16 %0, ptr %arrayidx2, align 289; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 3000000 for VF 2 For instruction: store i16 %2, ptr %arrayidx7, align 290; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 17 for VF 4 For instruction: store i16 %0, ptr %arrayidx2, align 291; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 3000000 for VF 4 For instruction: store i16 %2, ptr %arrayidx7, align 292; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 35 for VF 8 For instruction: store i16 %0, ptr %arrayidx2, align 293; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 3000000 for VF 8 For instruction: store i16 %2, ptr %arrayidx7, align 294; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 71 for VF 16 For instruction: store i16 %0, ptr %arrayidx2, align 295; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 3000000 for VF 16 For instruction: store i16 %2, ptr %arrayidx7, align 296;97; ENABLED_MASKED_STRIDED-LABEL: 'test2'98; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 299; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 2100; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx2, align 2101; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %2, ptr %arrayidx7, align 2102; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 2 For instruction: store i16 %0, ptr %arrayidx2, align 2103; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 13 for VF 2 For instruction: store i16 %2, ptr %arrayidx7, align 2104; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 4 For instruction: store i16 %0, ptr %arrayidx2, align 2105; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 14 for VF 4 For instruction: store i16 %2, ptr %arrayidx7, align 2106; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 8 For instruction: store i16 %0, ptr %arrayidx2, align 2107; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 14 for VF 8 For instruction: store i16 %2, ptr %arrayidx7, align 2108; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 0 for VF 16 For instruction: store i16 %0, ptr %arrayidx2, align 2109; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 27 for VF 16 For instruction: store i16 %2, ptr %arrayidx7, align 2110;111entry:112 %cmp15 = icmp sgt i32 %numPoints, 0113 br i1 %cmp15, label %for.body.preheader, label %for.end114 115for.body.preheader:116 %wide.trip.count = zext i32 %numPoints to i64117 br label %for.body118 119for.body:120 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]121 %arrayidx = getelementptr inbounds i16, ptr %x, i64 %indvars.iv122 %0 = load i16, ptr %arrayidx, align 2123 %1 = shl nsw i64 %indvars.iv, 2124 %arrayidx2 = getelementptr inbounds i16, ptr %points, i64 %1125 store i16 %0, ptr %arrayidx2, align 2126 %arrayidx4 = getelementptr inbounds i16, ptr %y, i64 %indvars.iv127 %2 = load i16, ptr %arrayidx4, align 2128 %3 = or disjoint i64 %1, 1129 %arrayidx7 = getelementptr inbounds i16, ptr %points, i64 %3130 store i16 %2, ptr %arrayidx7, align 2131 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1132 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count133 br i1 %exitcond.not, label %for.end.loopexit, label %for.body134 135for.end.loopexit:136 br label %for.end137 138for.end:139 ret void140}141 142; (3) Testing a scenario of a conditional store. The gaps mask of the store is also143; And-ed with the condition mask (x[i] > 0).144; If using masked memops to vectorize interleaved-group with gaps is145; not allowed, the store is scalarized and predicated.146; Here the Interleave-group is with factor 3, storing only 1 member out of the 3.147; The input IR was generated from this source:148; for(i=0;i<1024;i++){149; if (x[i] > 0)150; points[i*3] = x[i];151; }152 153define void @test(ptr noalias nocapture %points, ptr noalias nocapture readonly %x, ptr noalias nocapture readnone %y) {154; DISABLED_MASKED_STRIDED-LABEL: 'test'155; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx6, align 2156; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx6, align 2157; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 2 for VF 2 For instruction: store i16 %0, ptr %arrayidx6, align 2158; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 4 for VF 4 For instruction: store i16 %0, ptr %arrayidx6, align 2159; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 8 for VF 8 For instruction: store i16 %0, ptr %arrayidx6, align 2160; DISABLED_MASKED_STRIDED: LV: Found an estimated cost of 16 for VF 16 For instruction: store i16 %0, ptr %arrayidx6, align 2161;162; ENABLED_MASKED_STRIDED-LABEL: 'test'163; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx6, align 2164; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 1 for VF 1 For instruction: store i16 %0, ptr %arrayidx6, align 2165; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 2 for VF 2 For instruction: store i16 %0, ptr %arrayidx6, align 2166; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 4 for VF 4 For instruction: store i16 %0, ptr %arrayidx6, align 2167; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 8 for VF 8 For instruction: store i16 %0, ptr %arrayidx6, align 2168; ENABLED_MASKED_STRIDED: LV: Found an estimated cost of 16 for VF 16 For instruction: store i16 %0, ptr %arrayidx6, align 2169;170entry:171 br label %for.body172 173for.body:174 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]175 %arrayidx = getelementptr inbounds i16, ptr %x, i64 %indvars.iv176 %0 = load i16, ptr %arrayidx, align 2177 %cmp1 = icmp sgt i16 %0, 0178 br i1 %cmp1, label %if.then, label %for.inc179 180if.then:181 %1 = mul nuw nsw i64 %indvars.iv, 3182 %arrayidx6 = getelementptr inbounds i16, ptr %points, i64 %1183 store i16 %0, ptr %arrayidx6, align 2184 br label %for.inc185 186for.inc:187 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1188 %exitcond.not = icmp eq i64 %indvars.iv.next, 1024189 br i1 %exitcond.not, label %for.end, label %for.body190 191for.end:192 ret void193}194