1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2 2; RUN: opt -S -pass-remarks=loop-vectorize -passes=loop-vectorize -enable-vplan-native-path -force-target-supports-scalable-vectors < %s 2>&1 | FileCheck %s 3; RUN: opt -S -pass-remarks=loop-vectorize -passes=loop-vectorize -enable-vplan-native-path < %s 2>&1 | FileCheck %s -check-prefix=NO_SCALABLE_VECS 4 5; Test if the vplan-native-path successfully vectorizes a loop using scalable 6; vectors if the target supports scalable vectors and rejects vectorization 7; if a scalable VF is requested but not supported by the target. 8 9; CHECK: remark: <unknown>:0:0: vectorized outer loop (vectorization width: vscale x 4, interleaved count: 1) 10; NO_SCALABLE_VECS: remark: <unknown>:0:0: loop not vectorized: the scalable user-specified vectorization width for outer-loop vectorization cannot be used because the target does not support scalable vectors. 11 12@A = external local_unnamed_addr global [1024 x float], align 4 13@B = external local_unnamed_addr global [512 x float], align 4 14 15define void @foo() { 16; CHECK-LABEL: define void @foo() { 17; CHECK-NEXT: entry: 18; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 19; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 4 20; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]] 21; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 22; CHECK: vector.ph: 23; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 24; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 4 25; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] 26; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] 27; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 4 x i64> @llvm.experimental.stepvector.nxv4i64() 28; CHECK-NEXT: [[TMP5:%.*]] = add <vscale x 4 x i64> [[TMP4]], zeroinitializer 29; CHECK-NEXT: [[TMP6:%.*]] = mul <vscale x 4 x i64> [[TMP5]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> poison, i64 1, i64 0), <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer) 30; CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 4 x i64> zeroinitializer, [[TMP6]] 31; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64() 32; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 4 33; CHECK-NEXT: [[TMP9:%.*]] = mul i64 1, [[TMP8]] 34; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x i64> poison, i64 [[TMP9]], i64 0 35; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x i64> [[DOTSPLATINSERT]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer 36; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 37; CHECK: vector.body: 38; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[OUTER_LOOP_LATCH4:%.*]] ] 39; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[OUTER_LOOP_LATCH4]] ] 40; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x float], ptr @A, i64 0, <vscale x 4 x i64> [[VEC_IND]] 41; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x float> @llvm.masked.gather.nxv4f32.nxv4p0(<vscale x 4 x ptr> [[TMP10]], i32 4, <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), <vscale x 4 x float> poison) 42; CHECK-NEXT: br label [[INNER_LOOP1:%.*]] 43; CHECK: inner_loop1: 44; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i64> [ zeroinitializer, [[VECTOR_BODY]] ], [ [[TMP13:%.*]], [[INNER_LOOP1]] ] 45; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <vscale x 4 x float> [ [[WIDE_MASKED_GATHER]], [[VECTOR_BODY]] ], [ [[TMP12:%.*]], [[INNER_LOOP1]] ] 46; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds [512 x float], ptr @B, i64 0, <vscale x 4 x i64> [[VEC_PHI]] 47; CHECK-NEXT: [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 4 x float> @llvm.masked.gather.nxv4f32.nxv4p0(<vscale x 4 x ptr> [[TMP11]], i32 4, <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), <vscale x 4 x float> poison) 48; CHECK-NEXT: [[TMP12]] = fmul <vscale x 4 x float> [[VEC_PHI2]], [[WIDE_MASKED_GATHER3]] 49; CHECK-NEXT: [[TMP13]] = add nuw nsw <vscale x 4 x i64> [[VEC_PHI]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> poison, i64 1, i64 0), <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer) 50; CHECK-NEXT: [[TMP14:%.*]] = icmp eq <vscale x 4 x i64> [[TMP13]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> poison, i64 512, i64 0), <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer) 51; CHECK-NEXT: [[TMP15:%.*]] = extractelement <vscale x 4 x i1> [[TMP14]], i32 0 52; CHECK-NEXT: br i1 [[TMP15]], label [[OUTER_LOOP_LATCH4]], label [[INNER_LOOP1]] 53; CHECK: outer_loop_latch4: 54; CHECK-NEXT: [[VEC_PHI5:%.*]] = phi <vscale x 4 x float> [ [[TMP12]], [[INNER_LOOP1]] ] 55; CHECK-NEXT: call void @llvm.masked.scatter.nxv4f32.nxv4p0(<vscale x 4 x float> [[VEC_PHI5]], <vscale x 4 x ptr> [[TMP10]], i32 4, <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer)) 56; CHECK-NEXT: [[TMP16:%.*]] = add nuw nsw <vscale x 4 x i64> [[VEC_IND]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> poison, i64 1, i64 0), <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer) 57; CHECK-NEXT: [[TMP17:%.*]] = icmp eq <vscale x 4 x i64> [[TMP16]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> poison, i64 1024, i64 0), <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer) 58; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64() 59; CHECK-NEXT: [[TMP19:%.*]] = mul i64 [[TMP18]], 4 60; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP19]] 61; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]] 62; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 63; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 64; CHECK: middle.block: 65; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] 66; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 67; CHECK: scalar.ph: 68; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 69; CHECK-NEXT: br label [[OUTER_LOOP:%.*]] 70; CHECK: outer_loop: 71; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[OUTER_LOOP_LATCH:%.*]] ] 72; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds [1024 x float], ptr @A, i64 0, i64 [[I]] 73; CHECK-NEXT: [[X_START:%.*]] = load float, ptr [[ARRAYIDX1]], align 4 74; CHECK-NEXT: br label [[INNER_LOOP:%.*]] 75; CHECK: inner_loop: 76; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[OUTER_LOOP]] ], [ [[J_NEXT:%.*]], [[INNER_LOOP]] ] 77; CHECK-NEXT: [[X:%.*]] = phi float [ [[X_START]], [[OUTER_LOOP]] ], [ [[X_NEXT:%.*]], [[INNER_LOOP]] ] 78; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [512 x float], ptr @B, i64 0, i64 [[J]] 79; CHECK-NEXT: [[B:%.*]] = load float, ptr [[ARRAYIDX2]], align 4 80; CHECK-NEXT: [[X_NEXT]] = fmul float [[X]], [[B]] 81; CHECK-NEXT: [[J_NEXT]] = add nuw nsw i64 [[J]], 1 82; CHECK-NEXT: [[INNER_EXITCOND:%.*]] = icmp eq i64 [[J_NEXT]], 512 83; CHECK-NEXT: br i1 [[INNER_EXITCOND]], label [[OUTER_LOOP_LATCH]], label [[INNER_LOOP]] 84; CHECK: outer_loop_latch: 85; CHECK-NEXT: [[X_NEXT_LCSSA:%.*]] = phi float [ [[X_NEXT]], [[INNER_LOOP]] ] 86; CHECK-NEXT: store float [[X_NEXT_LCSSA]], ptr [[ARRAYIDX1]], align 4 87; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 88; CHECK-NEXT: [[OUTER_EXITCOND:%.*]] = icmp eq i64 [[I_NEXT]], 1024 89; CHECK-NEXT: br i1 [[OUTER_EXITCOND]], label [[EXIT]], label [[OUTER_LOOP]], !llvm.loop [[LOOP3:![0-9]+]] 90; CHECK: exit: 91; CHECK-NEXT: ret void 92; 93; NO_SCALABLE_VECS-LABEL: define void @foo() { 94; NO_SCALABLE_VECS-NEXT: entry: 95; NO_SCALABLE_VECS-NEXT: br label [[OUTER_LOOP:%.*]] 96; NO_SCALABLE_VECS: outer_loop: 97; NO_SCALABLE_VECS-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[OUTER_LOOP_LATCH:%.*]] ] 98; NO_SCALABLE_VECS-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds [1024 x float], ptr @A, i64 0, i64 [[I]] 99; NO_SCALABLE_VECS-NEXT: [[X_START:%.*]] = load float, ptr [[ARRAYIDX1]], align 4 100; NO_SCALABLE_VECS-NEXT: br label [[INNER_LOOP:%.*]] 101; NO_SCALABLE_VECS: inner_loop: 102; NO_SCALABLE_VECS-NEXT: [[J:%.*]] = phi i64 [ 0, [[OUTER_LOOP]] ], [ [[J_NEXT:%.*]], [[INNER_LOOP]] ] 103; NO_SCALABLE_VECS-NEXT: [[X:%.*]] = phi float [ [[X_START]], [[OUTER_LOOP]] ], [ [[X_NEXT:%.*]], [[INNER_LOOP]] ] 104; NO_SCALABLE_VECS-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [512 x float], ptr @B, i64 0, i64 [[J]] 105; NO_SCALABLE_VECS-NEXT: [[B:%.*]] = load float, ptr [[ARRAYIDX2]], align 4 106; NO_SCALABLE_VECS-NEXT: [[X_NEXT]] = fmul float [[X]], [[B]] 107; NO_SCALABLE_VECS-NEXT: [[J_NEXT]] = add nuw nsw i64 [[J]], 1 108; NO_SCALABLE_VECS-NEXT: [[INNER_EXITCOND:%.*]] = icmp eq i64 [[J_NEXT]], 512 109; NO_SCALABLE_VECS-NEXT: br i1 [[INNER_EXITCOND]], label [[OUTER_LOOP_LATCH]], label [[INNER_LOOP]] 110; NO_SCALABLE_VECS: outer_loop_latch: 111; NO_SCALABLE_VECS-NEXT: [[X_NEXT_LCSSA:%.*]] = phi float [ [[X_NEXT]], [[INNER_LOOP]] ] 112; NO_SCALABLE_VECS-NEXT: store float [[X_NEXT_LCSSA]], ptr [[ARRAYIDX1]], align 4 113; NO_SCALABLE_VECS-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 114; NO_SCALABLE_VECS-NEXT: [[OUTER_EXITCOND:%.*]] = icmp eq i64 [[I_NEXT]], 1024 115; NO_SCALABLE_VECS-NEXT: br i1 [[OUTER_EXITCOND]], label [[EXIT:%.*]], label [[OUTER_LOOP]], !llvm.loop [[LOOP0:![0-9]+]] 116; NO_SCALABLE_VECS: exit: 117; NO_SCALABLE_VECS-NEXT: ret void 118entry: 119 br label %outer_loop 120 121outer_loop: 122 %i = phi i64 [ 0, %entry ], [ %i.next, %outer_loop_latch ] 123 %arrayidx1 = getelementptr inbounds [1024 x float], ptr @A, i64 0, i64 %i 124 %x.start = load float, ptr %arrayidx1, align 4 125 br label %inner_loop 126 127inner_loop: 128 %j = phi i64 [ 0, %outer_loop ], [ %j.next, %inner_loop ] 129 %x = phi float [ %x.start, %outer_loop ], [ %x.next, %inner_loop ] 130 %arrayidx2 = getelementptr inbounds [512 x float], ptr @B, i64 0, i64 %j 131 %b = load float, ptr %arrayidx2, align 4 132 %x.next = fmul float %x, %b 133 %j.next = add nuw nsw i64 %j, 1 134 %inner_exitcond = icmp eq i64 %j.next, 512 135 br i1 %inner_exitcond, label %outer_loop_latch, label %inner_loop 136 137outer_loop_latch: 138 store float %x.next, ptr %arrayidx1, align 4 139 %i.next = add nuw nsw i64 %i, 1 140 %outer_exitcond = icmp eq i64 %i.next, 1024 141 br i1 %outer_exitcond, label %exit, label %outer_loop, !llvm.loop !1 142 143exit: 144 ret void 145} 146 147!1 = distinct !{!1, !2, !3, !4} 148!2 = !{!"llvm.loop.vectorize.enable", i1 true} 149!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} 150!4 = !{!"llvm.loop.vectorize.width", i32 4} 151