223 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -mtriple=aarch64--linux-gnu -passes=slp-vectorizer -slp-threshold=-12 -pass-remarks-output=%t < %s | FileCheck %s3; RUN: cat %t | FileCheck -check-prefix=YAML %s4; RUN: opt -S -mtriple=aarch64--linux-gnu -passes=slp-vectorizer -slp-threshold=-12 -pass-remarks-output=%t < %s | FileCheck %s5; RUN: cat %t | FileCheck -check-prefix=YAML %s6 7; These tests check that we remove from consideration pairs of seed8; getelementptrs when they are known to have a constant difference. Such pairs9; are likely not good candidates for vectorization since one can be computed10; from the other. We use an unprofitable threshold to force vectorization.11;12; int getelementptr(int *g, int n, int w, int x, int y, int z) {13; int sum = 0;14; for (int i = 0; i < n ; ++i) {15; sum += g[2*i + w]; sum += g[2*i + x];16; sum += g[2*i + y]; sum += g[2*i + z];17; }18; return sum;19; }20;21 22; YAML-LABEL: Function: getelementptr_4x3223; YAML: --- !Passed24; YAML-NEXT: Pass: slp-vectorizer25; YAML-NEXT: Name: VectorizedList26; YAML-NEXT: Function: getelementptr_4x3227; YAML-NEXT: Args:28; YAML-NEXT: - String: 'SLP vectorized with cost '29; YAML-NEXT: - Cost: '6'30; YAML-NEXT: - String: ' and with tree size '31; YAML-NEXT: - TreeSize: '3'32 33; YAML: --- !Passed34; YAML-NEXT: Pass: slp-vectorizer35; YAML-NEXT: Name: VectorizedList36; YAML-NEXT: Function: getelementptr_4x3237; YAML-NEXT: Args:38; YAML-NEXT: - String: 'SLP vectorized with cost '39; YAML-NEXT: - Cost: '8'40; YAML-NEXT: - String: ' and with tree size '41; YAML-NEXT: - TreeSize: '3'42 43define i32 @getelementptr_4x32(ptr nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {44; CHECK-LABEL: @getelementptr_4x32(45; CHECK-NEXT: entry:46; CHECK-NEXT: [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 047; CHECK-NEXT: br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]48; CHECK: for.body.preheader:49; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> <i32 0, i32 poison>, i32 [[X:%.*]], i32 150; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i32 051; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[Z:%.*]], i32 152; CHECK-NEXT: br label [[FOR_BODY:%.*]]53; CHECK: for.cond.cleanup.loopexit:54; CHECK-NEXT: [[ADD16:%.*]] = extractelement <2 x i32> [[TMP17:%.*]], i32 055; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]56; CHECK: for.cond.cleanup:57; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD16]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]58; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]59; CHECK: for.body:60; CHECK-NEXT: [[TMP6:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP17]], [[FOR_BODY]] ]61; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x i32> [[TMP6]], i32 162; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP15]], 163; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x i32> [[TMP6]], i32 064; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i32> poison, i32 [[T4]], i32 065; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> poison, <2 x i32> zeroinitializer66; CHECK-NEXT: [[TMP5:%.*]] = add nsw <2 x i32> [[TMP4]], [[TMP0]]67; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i32> [[TMP5]], i32 068; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[G:%.*]], i32 [[TMP12]]69; CHECK-NEXT: [[T6:%.*]] = load i32, ptr [[ARRAYIDX]], align 470; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP16]]71; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[TMP5]], i32 172; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[TMP11]]73; CHECK-NEXT: [[T8:%.*]] = load i32, ptr [[ARRAYIDX5]], align 474; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]75; CHECK-NEXT: [[TMP18:%.*]] = add nsw <2 x i32> [[TMP4]], [[TMP2]]76; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i32> [[TMP18]], i32 077; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[TMP13]]78; CHECK-NEXT: [[T10:%.*]] = load i32, ptr [[ARRAYIDX10]], align 479; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]80; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i32> [[TMP18]], i32 181; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[TMP14]]82; CHECK-NEXT: [[T12:%.*]] = load i32, ptr [[ARRAYIDX15]], align 483; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x i32> <i32 poison, i32 1>, i32 [[ADD11]], i32 084; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[T12]], i32 085; CHECK-NEXT: [[TMP17]] = add nsw <2 x i32> [[TMP19]], [[TMP20]]86; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = extractelement <2 x i32> [[TMP17]], i32 187; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[N]]88; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]89;90entry:91 %cmp31 = icmp sgt i32 %n, 092 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup93 94for.body.preheader:95 br label %for.body96 97for.cond.cleanup.loopexit:98 br label %for.cond.cleanup99 100for.cond.cleanup:101 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]102 ret i32 %sum.0.lcssa103 104for.body:105 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]106 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]107 %t4 = shl nsw i32 %indvars.iv, 1108 %t5 = add nsw i32 %t4, 0109 %arrayidx = getelementptr inbounds i32, ptr %g, i32 %t5110 %t6 = load i32, ptr %arrayidx, align 4111 %add1 = add nsw i32 %t6, %sum.032112 %t7 = add nsw i32 %t4, %x113 %arrayidx5 = getelementptr inbounds i32, ptr %g, i32 %t7114 %t8 = load i32, ptr %arrayidx5, align 4115 %add6 = add nsw i32 %add1, %t8116 %t9 = add nsw i32 %t4, %y117 %arrayidx10 = getelementptr inbounds i32, ptr %g, i32 %t9118 %t10 = load i32, ptr %arrayidx10, align 4119 %add11 = add nsw i32 %add6, %t10120 %t11 = add nsw i32 %t4, %z121 %arrayidx15 = getelementptr inbounds i32, ptr %g, i32 %t11122 %t12 = load i32, ptr %arrayidx15, align 4123 %add16 = add nsw i32 %add11, %t12124 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1125 %exitcond = icmp eq i32 %indvars.iv.next , %n126 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body127}128 129; YAML-LABEL: Function: getelementptr_2x32130; YAML: --- !Passed131; YAML: Pass: slp-vectorizer132; YAML: Name: VectorizedList133; YAML: Function: getelementptr_2x32134; YAML: Args:135; YAML: - String: 'SLP vectorized with cost '136; YAML: - Cost: '8'137; YAML-NEXT: - String: ' and with tree size '138; YAML-NEXT: - TreeSize: '3'139 140define i32 @getelementptr_2x32(ptr nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {141; CHECK-LABEL: @getelementptr_2x32(142; CHECK-NEXT: entry:143; CHECK-NEXT: [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0144; CHECK-NEXT: br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]145; CHECK: for.body.preheader:146; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i32 0147; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[Z:%.*]], i32 1148; CHECK-NEXT: br label [[FOR_BODY:%.*]]149; CHECK: for.cond.cleanup.loopexit:150; CHECK-NEXT: [[OP_RDX:%.*]] = extractelement <2 x i32> [[TMP13:%.*]], i32 0151; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]152; CHECK: for.cond.cleanup:153; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_RDX]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]154; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]155; CHECK: for.body:156; CHECK-NEXT: [[TMP6:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP13]], [[FOR_BODY]] ]157; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i32> [[TMP6]], i32 1158; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP12]], 1159; CHECK-NEXT: [[T5:%.*]] = add nsw i32 [[T4]], 0160; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[G:%.*]], i32 [[T5]]161; CHECK-NEXT: [[T6:%.*]] = load i32, ptr [[ARRAYIDX]], align 4162; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i32> [[TMP6]], i32 0163; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP5]]164; CHECK-NEXT: [[T7:%.*]] = add nsw i32 [[T4]], 1165; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[T7]]166; CHECK-NEXT: [[T8:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4167; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]168; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[T4]], i32 0169; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer170; CHECK-NEXT: [[TMP3:%.*]] = add nsw <2 x i32> [[TMP2]], [[TMP0]]171; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[TMP3]], i32 0172; CHECK-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[TMP9]]173; CHECK-NEXT: [[T10:%.*]] = load i32, ptr [[ARRAYIDX11]], align 4174; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]175; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1176; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, ptr [[G]], i32 [[TMP10]]177; CHECK-NEXT: [[T12:%.*]] = load i32, ptr [[ARRAYIDX15]], align 4178; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x i32> <i32 poison, i32 1>, i32 [[ADD11]], i32 0179; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[T12]], i32 0180; CHECK-NEXT: [[TMP13]] = add nsw <2 x i32> [[TMP11]], [[TMP14]]181; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = extractelement <2 x i32> [[TMP13]], i32 1182; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[N]]183; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]184;185entry:186 %cmp31 = icmp sgt i32 %n, 0187 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup188 189for.body.preheader:190 br label %for.body191 192for.cond.cleanup.loopexit:193 br label %for.cond.cleanup194 195for.cond.cleanup:196 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]197 ret i32 %sum.0.lcssa198 199for.body:200 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]201 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]202 %t4 = shl nsw i32 %indvars.iv, 1203 %t5 = add nsw i32 %t4, 0204 %arrayidx = getelementptr inbounds i32, ptr %g, i32 %t5205 %t6 = load i32, ptr %arrayidx, align 4206 %add1 = add nsw i32 %t6, %sum.032207 %t7 = add nsw i32 %t4, 1208 %arrayidx5 = getelementptr inbounds i32, ptr %g, i32 %t7209 %t8 = load i32, ptr %arrayidx5, align 4210 %add6 = add nsw i32 %add1, %t8211 %t9 = add nsw i32 %t4, %y212 %arrayidx10 = getelementptr inbounds i32, ptr %g, i32 %t9213 %t10 = load i32, ptr %arrayidx10, align 4214 %add11 = add nsw i32 %add6, %t10215 %t11 = add nsw i32 %t4, %z216 %arrayidx15 = getelementptr inbounds i32, ptr %g, i32 %t11217 %t12 = load i32, ptr %arrayidx15, align 4218 %add16 = add nsw i32 %add11, %t12219 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1220 %exitcond = icmp eq i32 %indvars.iv.next , %n221 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body222}223