372 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -mtriple aarch64-linux-gnu -mattr=+sve -passes=loop-vectorize,dce,instcombine -S \3; RUN: -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s4 5; Ensure that we can vectorize loops such as:6; int *ptr = c;7; for (long long i = 0; i < n; i++) {8; int X1 = *ptr++;9; int X2 = *ptr++;10; a[i] = X1 + 1;11; b[i] = X2 + 1;12; }13; with scalable vectors, including unrolling. The test below makes sure14; that we can use gather instructions with the correct offsets, taking15; vscale into account.16 17define void @widen_ptr_phi_unrolled(ptr noalias nocapture %a, ptr noalias nocapture %b, ptr nocapture readonly %c, i64 %n) #0 {18; CHECK-LABEL: @widen_ptr_phi_unrolled(19; CHECK-NEXT: entry:20; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()21; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 322; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]23; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]24; CHECK: vector.ph:25; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()26; CHECK-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 327; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP5]]28; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], [[DOTNOT]]29; CHECK-NEXT: [[TMP26:%.*]] = shl i64 [[N_VEC]], 330; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[C:%.*]], i64 [[TMP26]]31; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]32; CHECK: vector.body:33; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]34; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 335; CHECK-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 536; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[C]], i64 [[OFFSET_IDX]]37; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[C]], i64 [[OFFSET_IDX]]38; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[TMP8]], i64 [[TMP7]]39; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[NEXT_GEP]], align 440; CHECK-NEXT: [[STRIDED_VEC:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC]])41; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 042; CHECK-NEXT: [[TMP10:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 143; CHECK-NEXT: [[WIDE_VEC3:%.*]] = load <vscale x 8 x i32>, ptr [[NEXT_GEP2]], align 444; CHECK-NEXT: [[STRIDED_VEC4:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC3]])45; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC4]], 046; CHECK-NEXT: [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC4]], 147; CHECK-NEXT: [[TMP13:%.*]] = add nsw <vscale x 4 x i32> [[TMP9]], splat (i32 1)48; CHECK-NEXT: [[TMP14:%.*]] = add nsw <vscale x 4 x i32> [[TMP11]], splat (i32 1)49; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]50; CHECK-NEXT: [[TMP16:%.*]] = call i64 @llvm.vscale.i64()51; CHECK-NEXT: [[DOTIDX:%.*]] = shl nuw nsw i64 [[TMP16]], 452; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP15]], i64 [[DOTIDX]]53; CHECK-NEXT: store <vscale x 4 x i32> [[TMP13]], ptr [[TMP15]], align 454; CHECK-NEXT: store <vscale x 4 x i32> [[TMP14]], ptr [[TMP17]], align 455; CHECK-NEXT: [[TMP18:%.*]] = add nsw <vscale x 4 x i32> [[TMP10]], splat (i32 1)56; CHECK-NEXT: [[TMP19:%.*]] = add nsw <vscale x 4 x i32> [[TMP12]], splat (i32 1)57; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[INDEX]]58; CHECK-NEXT: [[TMP21:%.*]] = call i64 @llvm.vscale.i64()59; CHECK-NEXT: [[DOTIDX5:%.*]] = shl nuw nsw i64 [[TMP21]], 460; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP20]], i64 [[DOTIDX5]]61; CHECK-NEXT: store <vscale x 4 x i32> [[TMP18]], ptr [[TMP20]], align 462; CHECK-NEXT: store <vscale x 4 x i32> [[TMP19]], ptr [[TMP22]], align 463; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]64; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]65; CHECK-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]66; CHECK: middle.block:67; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]68; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]69; CHECK: scalar.ph:70; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[C]], [[ENTRY:%.*]] ]71; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]72; CHECK-NEXT: br label [[FOR_BODY:%.*]]73; CHECK: for.body:74; CHECK-NEXT: [[PTR_014:%.*]] = phi ptr [ [[INCDEC_PTR1:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]75; CHECK-NEXT: [[I_013:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]76; CHECK-NEXT: [[INCDEC_PTR:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_014]], i64 477; CHECK-NEXT: [[TMP24:%.*]] = load i32, ptr [[PTR_014]], align 478; CHECK-NEXT: [[INCDEC_PTR1]] = getelementptr inbounds nuw i8, ptr [[PTR_014]], i64 879; CHECK-NEXT: [[TMP25:%.*]] = load i32, ptr [[INCDEC_PTR]], align 480; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP24]], 181; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I_013]]82; CHECK-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX]], align 483; CHECK-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP25]], 184; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I_013]]85; CHECK-NEXT: store i32 [[ADD2]], ptr [[ARRAYIDX3]], align 486; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_013]], 187; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]88; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]89; CHECK: for.exit:90; CHECK-NEXT: ret void91;92entry:93 br label %for.body94 95for.body: ; preds = %entry, %for.body96 %ptr.014 = phi ptr [ %incdec.ptr1, %for.body ], [ %c, %entry ]97 %i.013 = phi i64 [ %inc, %for.body ], [ 0, %entry ]98 %incdec.ptr = getelementptr inbounds i32, ptr %ptr.014, i64 199 %0 = load i32, ptr %ptr.014, align 4100 %incdec.ptr1 = getelementptr inbounds i32, ptr %ptr.014, i64 2101 %1 = load i32, ptr %incdec.ptr, align 4102 %add = add nsw i32 %0, 1103 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %i.013104 store i32 %add, ptr %arrayidx, align 4105 %add2 = add nsw i32 %1, 1106 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 %i.013107 store i32 %add2, ptr %arrayidx3, align 4108 %inc = add nuw nsw i64 %i.013, 1109 %exitcond.not = icmp eq i64 %inc, %n110 br i1 %exitcond.not, label %for.exit, label %for.body, !llvm.loop !0111 112for.exit: ; preds = %for.body113 ret void114}115 116 117; Ensure we can vectorise loops without interleaving, e.g.:118; int *D = dst;119; int *S = src;120; for (long long i = 0; i < n; i++) {121; *D = *S * 2;122; D++;123; S++;124; }125; This takes us down a different codepath to the test above, where126; here we treat the PHIs as being uniform.127 128define void @widen_2ptrs_phi_unrolled(ptr noalias nocapture %dst, ptr noalias nocapture readonly %src, i64 %n) #0 {129; CHECK-LABEL: @widen_2ptrs_phi_unrolled(130; CHECK-NEXT: entry:131; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()132; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 3133; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]134; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]135; CHECK: vector.ph:136; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()137; CHECK-NEXT: [[TMP6:%.*]] = shl nuw nsw i64 [[TMP5]], 3138; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP6]]139; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], [[DOTNOT]]140; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[N_VEC]], 2141; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[SRC:%.*]], i64 [[TMP3]]142; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[N_VEC]], 2143; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 [[TMP4]]144; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]145; CHECK: vector.body:146; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]147; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 2148; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[OFFSET_IDX]]149; CHECK-NEXT: [[OFFSET_IDX4:%.*]] = shl i64 [[INDEX]], 2150; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX4]]151; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()152; CHECK-NEXT: [[DOTIDX:%.*]] = shl nuw nsw i64 [[TMP7]], 4153; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 [[DOTIDX]]154; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[NEXT_GEP]], align 4155; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <vscale x 4 x i32>, ptr [[TMP8]], align 4156; CHECK-NEXT: [[TMP9:%.*]] = shl nsw <vscale x 4 x i32> [[WIDE_LOAD]], splat (i32 1)157; CHECK-NEXT: [[TMP10:%.*]] = shl nsw <vscale x 4 x i32> [[WIDE_LOAD6]], splat (i32 1)158; CHECK-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()159; CHECK-NEXT: [[DOTIDX7:%.*]] = shl nuw nsw i64 [[TMP11]], 4160; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[NEXT_GEP5]], i64 [[DOTIDX7]]161; CHECK-NEXT: store <vscale x 4 x i32> [[TMP9]], ptr [[NEXT_GEP5]], align 4162; CHECK-NEXT: store <vscale x 4 x i32> [[TMP10]], ptr [[TMP12]], align 4163; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]164; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]165; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]166; CHECK: middle.block:167; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]168; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]169; CHECK: scalar.ph:170; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]171; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[SRC]], [[ENTRY]] ]172; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi ptr [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[DST]], [[ENTRY]] ]173; CHECK-NEXT: br label [[FOR_BODY:%.*]]174; CHECK: for.body:175; CHECK-NEXT: [[I_011:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]176; CHECK-NEXT: [[S_010:%.*]] = phi ptr [ [[INCDEC_PTR1:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]177; CHECK-NEXT: [[D_09:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]178; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr [[S_010]], align 4179; CHECK-NEXT: [[MUL:%.*]] = shl nsw i32 [[TMP14]], 1180; CHECK-NEXT: store i32 [[MUL]], ptr [[D_09]], align 4181; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds nuw i8, ptr [[D_09]], i64 4182; CHECK-NEXT: [[INCDEC_PTR1]] = getelementptr inbounds nuw i8, ptr [[S_010]], i64 4183; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_011]], 1184; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]185; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]186; CHECK: for.cond.cleanup:187; CHECK-NEXT: ret void188;189entry:190 br label %for.body191 192for.body: ; preds = %entry, %for.body193 %i.011 = phi i64 [ %inc, %for.body ], [ 0, %entry ]194 %S.010 = phi ptr [ %incdec.ptr1, %for.body ], [ %src, %entry ]195 %D.09 = phi ptr [ %incdec.ptr, %for.body ], [ %dst, %entry ]196 %0 = load i32, ptr %S.010, align 4197 %mul = shl nsw i32 %0, 1198 store i32 %mul, ptr %D.09, align 4199 %incdec.ptr = getelementptr inbounds i32, ptr %D.09, i64 1200 %incdec.ptr1 = getelementptr inbounds i32, ptr %S.010, i64 1201 %inc = add nuw nsw i64 %i.011, 1202 %exitcond.not = icmp eq i64 %inc, %n203 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !llvm.loop !0204 205for.cond.cleanup: ; preds = %for.body206 ret void207}208 209 210;211; Check multiple pointer induction variables where only one is recognized as212; uniform and remains uniform after vectorization. The other pointer induction213; variable is not recognized as uniform and is not uniform after vectorization214; because it is stored to memory.215;216 217define i32 @pointer_iv_mixed(ptr noalias %a, ptr noalias %b, i64 %n) #0 {218; CHECK-LABEL: @pointer_iv_mixed(219; CHECK-NEXT: entry:220; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)221; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()222; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 1223; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp samesign ult i64 [[SMAX]], [[TMP1]]224; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]225; CHECK: vector.ph:226; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()227; CHECK-NEXT: [[TMP6:%.*]] = shl nuw nsw i64 [[TMP5]], 1228; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP6]]229; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], [[DOTNOT]]230; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[N_VEC]], 2231; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP3]]232; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[N_VEC]], 3233; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 [[TMP4]]234; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]235; CHECK: vector.body:236; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]237; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[A]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]238; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[VECTOR_BODY]] ]239; CHECK-NEXT: [[TMP9:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()240; CHECK-NEXT: [[TMP10:%.*]] = shl <vscale x 2 x i64> [[TMP9]], splat (i64 2)241; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[TMP10]]242; CHECK-NEXT: [[TMP11:%.*]] = extractelement <vscale x 2 x ptr> [[VECTOR_GEP]], i64 0243; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3244; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX]]245; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i32>, ptr [[TMP11]], align 8246; CHECK-NEXT: [[TMP12]] = add <vscale x 2 x i32> [[WIDE_LOAD]], [[VEC_PHI]]247; CHECK-NEXT: store <vscale x 2 x ptr> [[VECTOR_GEP]], ptr [[NEXT_GEP]], align 8248; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]249; CHECK-NEXT: [[TMP8:%.*]] = shl nuw nsw i64 [[TMP5]], 3250; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP8]]251; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]252; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]253; CHECK: middle.block:254; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[TMP12]])255; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]256; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]257; CHECK: scalar.ph:258; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]259; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[A]], [[ENTRY]] ]260; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi ptr [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[B]], [[ENTRY]] ]261; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP14]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]262; CHECK-NEXT: br label [[FOR_BODY:%.*]]263; CHECK: for.body:264; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]265; CHECK-NEXT: [[P:%.*]] = phi ptr [ [[VAR3:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]266; CHECK-NEXT: [[Q:%.*]] = phi ptr [ [[VAR4:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]267; CHECK-NEXT: [[VAR0:%.*]] = phi i32 [ [[VAR2:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]268; CHECK-NEXT: [[VAR1:%.*]] = load i32, ptr [[P]], align 8269; CHECK-NEXT: [[VAR2]] = add i32 [[VAR1]], [[VAR0]]270; CHECK-NEXT: store ptr [[P]], ptr [[Q]], align 8271; CHECK-NEXT: [[VAR3]] = getelementptr inbounds nuw i8, ptr [[P]], i64 4272; CHECK-NEXT: [[VAR4]] = getelementptr inbounds nuw i8, ptr [[Q]], i64 8273; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1274; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]275; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP8:![0-9]+]]276; CHECK: for.end:277; CHECK-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR2]], [[FOR_BODY]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]278; CHECK-NEXT: ret i32 [[VAR5]]279;280entry:281 br label %for.body282 283for.body:284 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]285 %p = phi ptr [ %var3, %for.body ], [ %a, %entry ]286 %q = phi ptr [ %var4, %for.body ], [ %b, %entry ]287 %var0 = phi i32 [ %var2, %for.body ], [ 0, %entry ]288 %var1 = load i32, ptr %p, align 8289 %var2 = add i32 %var1, %var0290 store ptr %p, ptr %q, align 8291 %var3 = getelementptr inbounds i32, ptr %p, i32 1292 %var4 = getelementptr inbounds ptr, ptr %q, i32 1293 %i.next = add nuw nsw i64 %i, 1294 %cond = icmp slt i64 %i.next, %n295 br i1 %cond, label %for.body, label %for.end, !llvm.loop !6296 297for.end:298 %var5 = phi i32 [ %var2, %for.body ]299 ret i32 %var5300}301 302define void @phi_used_in_vector_compare_and_scalar_indvar_update_and_store(ptr %ptr) #0 {303; CHECK-LABEL: @phi_used_in_vector_compare_and_scalar_indvar_update_and_store(304; CHECK-NEXT: entry:305; CHECK-NEXT: br label [[VECTOR_PH:%.*]]306; CHECK: vector.ph:307; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()308; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 1309; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]310; CHECK: vector.body:311; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]312; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[PTR:%.*]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]313; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()314; CHECK-NEXT: [[TMP5:%.*]] = shl <vscale x 2 x i64> [[TMP4]], splat (i64 1)315; CHECK-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[TMP5]]316; CHECK-NEXT: [[TMP7:%.*]] = extractelement <vscale x 2 x ptr> [[VECTOR_GEP]], i64 0317; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <vscale x 2 x ptr> [[VECTOR_GEP]], zeroinitializer318; CHECK-NEXT: call void @llvm.masked.store.nxv2i16.p0(<vscale x 2 x i16> zeroinitializer, ptr align 2 [[TMP7]], <vscale x 2 x i1> [[TMP6]])319; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]320; CHECK-NEXT: [[TMP3:%.*]] = shl nuw nsw i64 [[TMP0]], 2321; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP3]]322; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024323; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]324; CHECK: middle.block:325; CHECK-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH:%.*]]326; CHECK: scalar.ph:327; CHECK-NEXT: br label [[FOR_BODY:%.*]]328; CHECK: for.body:329; CHECK-NEXT: br i1 poison, label [[IF_END_SINK_SPLIT:%.*]], label [[IF_END:%.*]]330; CHECK: if.end.sink.split:331; CHECK-NEXT: br label [[IF_END]]332; CHECK: if.end:333; CHECK-NEXT: br i1 poison, label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP10:![0-9]+]]334; CHECK: for.end:335; CHECK-NEXT: ret void336;337entry:338 br label %for.body339 340for.body: ; preds = %if.end, %entry341 %iv = phi i64 [ %inc, %if.end ], [ 0, %entry ]342 %iv.ptr = phi ptr [ %incdec.iv.ptr, %if.end ], [ %ptr, %entry ]343 %cmp.i = icmp ne ptr %iv.ptr, null344 br i1 %cmp.i, label %if.end.sink.split, label %if.end345 346if.end.sink.split: ; preds = %for.body347 store i16 0, ptr %iv.ptr, align 2348 br label %if.end349 350if.end: ; preds = %if.end.sink.split, %for.body351 %incdec.iv.ptr = getelementptr inbounds i16, ptr %iv.ptr, i64 1352 %inc = add nuw nsw i64 %iv, 1353 %exitcond.not = icmp ult i64 %inc, 1024354 br i1 %exitcond.not, label %for.body, label %for.end, !llvm.loop !6355 356for.end: ; preds = %if.end, %for.end357 %iv.ptr.1.lcssa = phi ptr [ %incdec.iv.ptr, %if.end ]358 ret void359}360 361attributes #0 = { vscale_range(1, 16) }362 363!0 = distinct !{!0, !1, !2, !3, !4, !5}364!1 = !{!"llvm.loop.mustprogress"}365!2 = !{!"llvm.loop.vectorize.width", i32 4}366!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}367!4 = !{!"llvm.loop.vectorize.enable", i1 true}368!5 = !{!"llvm.loop.interleave.count", i32 2}369!6 = distinct !{!6, !1, !7, !3, !4, !8}370!7 = !{!"llvm.loop.vectorize.width", i32 2}371!8 = !{!"llvm.loop.interleave.count", i32 1}372