1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s -loop-vectorize -dce -instcombine -S | FileCheck -check-prefix=GFX9 %s 3; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=fiji < %s -loop-vectorize -dce -instcombine -S | FileCheck -check-prefix=VI %s 4; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s -loop-vectorize -dce -instcombine -S | FileCheck -check-prefix=CI %s 5 6define half @vectorize_v2f16_loop(half addrspace(1)* noalias %s) { 7; GFX9-LABEL: @vectorize_v2f16_loop( 8; GFX9-NEXT: entry: 9; GFX9-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 10; GFX9: vector.ph: 11; GFX9-NEXT: br label [[VECTOR_BODY:%.*]] 12; GFX9: vector.body: 13; GFX9-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 14; GFX9-NEXT: [[VEC_PHI:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ] 15; GFX9-NEXT: [[VEC_PHI1:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ] 16; GFX9-NEXT: [[VEC_PHI2:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ] 17; GFX9-NEXT: [[VEC_PHI3:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ] 18; GFX9-NEXT: [[VEC_PHI4:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ] 19; GFX9-NEXT: [[VEC_PHI5:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ] 20; GFX9-NEXT: [[VEC_PHI6:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ] 21; GFX9-NEXT: [[VEC_PHI7:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ] 22; GFX9-NEXT: [[TMP0:%.*]] = getelementptr inbounds half, half addrspace(1)* [[S:%.*]], i64 [[INDEX]] 23; GFX9-NEXT: [[TMP1:%.*]] = bitcast half addrspace(1)* [[TMP0]] to <2 x half> addrspace(1)* 24; GFX9-NEXT: [[WIDE_LOAD:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP1]], align 2 25; GFX9-NEXT: [[TMP2:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 2 26; GFX9-NEXT: [[TMP3:%.*]] = bitcast half addrspace(1)* [[TMP2]] to <2 x half> addrspace(1)* 27; GFX9-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP3]], align 2 28; GFX9-NEXT: [[TMP4:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 4 29; GFX9-NEXT: [[TMP5:%.*]] = bitcast half addrspace(1)* [[TMP4]] to <2 x half> addrspace(1)* 30; GFX9-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP5]], align 2 31; GFX9-NEXT: [[TMP6:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 6 32; GFX9-NEXT: [[TMP7:%.*]] = bitcast half addrspace(1)* [[TMP6]] to <2 x half> addrspace(1)* 33; GFX9-NEXT: [[WIDE_LOAD10:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP7]], align 2 34; GFX9-NEXT: [[TMP8:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 8 35; GFX9-NEXT: [[TMP9:%.*]] = bitcast half addrspace(1)* [[TMP8]] to <2 x half> addrspace(1)* 36; GFX9-NEXT: [[WIDE_LOAD11:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP9]], align 2 37; GFX9-NEXT: [[TMP10:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 10 38; GFX9-NEXT: [[TMP11:%.*]] = bitcast half addrspace(1)* [[TMP10]] to <2 x half> addrspace(1)* 39; GFX9-NEXT: [[WIDE_LOAD12:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP11]], align 2 40; GFX9-NEXT: [[TMP12:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 12 41; GFX9-NEXT: [[TMP13:%.*]] = bitcast half addrspace(1)* [[TMP12]] to <2 x half> addrspace(1)* 42; GFX9-NEXT: [[WIDE_LOAD13:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP13]], align 2 43; GFX9-NEXT: [[TMP14:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 14 44; GFX9-NEXT: [[TMP15:%.*]] = bitcast half addrspace(1)* [[TMP14]] to <2 x half> addrspace(1)* 45; GFX9-NEXT: [[WIDE_LOAD14:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP15]], align 2 46; GFX9-NEXT: [[TMP16]] = fadd fast <2 x half> [[VEC_PHI]], [[WIDE_LOAD]] 47; GFX9-NEXT: [[TMP17]] = fadd fast <2 x half> [[VEC_PHI1]], [[WIDE_LOAD8]] 48; GFX9-NEXT: [[TMP18]] = fadd fast <2 x half> [[VEC_PHI2]], [[WIDE_LOAD9]] 49; GFX9-NEXT: [[TMP19]] = fadd fast <2 x half> [[VEC_PHI3]], [[WIDE_LOAD10]] 50; GFX9-NEXT: [[TMP20]] = fadd fast <2 x half> [[VEC_PHI4]], [[WIDE_LOAD11]] 51; GFX9-NEXT: [[TMP21]] = fadd fast <2 x half> [[VEC_PHI5]], [[WIDE_LOAD12]] 52; GFX9-NEXT: [[TMP22]] = fadd fast <2 x half> [[VEC_PHI6]], [[WIDE_LOAD13]] 53; GFX9-NEXT: [[TMP23]] = fadd fast <2 x half> [[VEC_PHI7]], [[WIDE_LOAD14]] 54; GFX9-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16 55; GFX9-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256 56; GFX9-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]] 57; GFX9: middle.block: 58; GFX9-NEXT: [[BIN_RDX:%.*]] = fadd fast <2 x half> [[TMP17]], [[TMP16]] 59; GFX9-NEXT: [[BIN_RDX15:%.*]] = fadd fast <2 x half> [[TMP18]], [[BIN_RDX]] 60; GFX9-NEXT: [[BIN_RDX16:%.*]] = fadd fast <2 x half> [[TMP19]], [[BIN_RDX15]] 61; GFX9-NEXT: [[BIN_RDX17:%.*]] = fadd fast <2 x half> [[TMP20]], [[BIN_RDX16]] 62; GFX9-NEXT: [[BIN_RDX18:%.*]] = fadd fast <2 x half> [[TMP21]], [[BIN_RDX17]] 63; GFX9-NEXT: [[BIN_RDX19:%.*]] = fadd fast <2 x half> [[TMP22]], [[BIN_RDX18]] 64; GFX9-NEXT: [[BIN_RDX20:%.*]] = fadd fast <2 x half> [[TMP23]], [[BIN_RDX19]] 65; GFX9-NEXT: [[RDX_SHUF:%.*]] = shufflevector <2 x half> [[BIN_RDX20]], <2 x half> undef, <2 x i32> <i32 1, i32 undef> 66; GFX9-NEXT: [[BIN_RDX21:%.*]] = fadd fast <2 x half> [[BIN_RDX20]], [[RDX_SHUF]] 67; GFX9-NEXT: [[TMP25:%.*]] = extractelement <2 x half> [[BIN_RDX21]], i32 0 68; GFX9-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 69; GFX9: scalar.ph: 70; GFX9-NEXT: br label [[FOR_BODY:%.*]] 71; GFX9: for.body: 72; GFX9-NEXT: br i1 undef, label [[FOR_END]], label [[FOR_BODY]], [[LOOP2:!llvm.loop !.*]] 73; GFX9: for.end: 74; GFX9-NEXT: [[ADD_LCSSA:%.*]] = phi half [ undef, [[FOR_BODY]] ], [ [[TMP25]], [[MIDDLE_BLOCK]] ] 75; GFX9-NEXT: ret half [[ADD_LCSSA]] 76; 77; VI-LABEL: @vectorize_v2f16_loop( 78; VI-NEXT: entry: 79; VI-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 80; VI: vector.ph: 81; VI-NEXT: br label [[VECTOR_BODY:%.*]] 82; VI: vector.body: 83; VI-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 84; VI-NEXT: [[VEC_PHI:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ] 85; VI-NEXT: [[VEC_PHI1:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ] 86; VI-NEXT: [[VEC_PHI2:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ] 87; VI-NEXT: [[VEC_PHI3:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ] 88; VI-NEXT: [[VEC_PHI4:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ] 89; VI-NEXT: [[VEC_PHI5:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ] 90; VI-NEXT: [[VEC_PHI6:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ] 91; VI-NEXT: [[VEC_PHI7:%.*]] = phi <2 x half> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ] 92; VI-NEXT: [[TMP0:%.*]] = getelementptr inbounds half, half addrspace(1)* [[S:%.*]], i64 [[INDEX]] 93; VI-NEXT: [[TMP1:%.*]] = bitcast half addrspace(1)* [[TMP0]] to <2 x half> addrspace(1)* 94; VI-NEXT: [[WIDE_LOAD:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP1]], align 2 95; VI-NEXT: [[TMP2:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 2 96; VI-NEXT: [[TMP3:%.*]] = bitcast half addrspace(1)* [[TMP2]] to <2 x half> addrspace(1)* 97; VI-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP3]], align 2 98; VI-NEXT: [[TMP4:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 4 99; VI-NEXT: [[TMP5:%.*]] = bitcast half addrspace(1)* [[TMP4]] to <2 x half> addrspace(1)* 100; VI-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP5]], align 2 101; VI-NEXT: [[TMP6:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 6 102; VI-NEXT: [[TMP7:%.*]] = bitcast half addrspace(1)* [[TMP6]] to <2 x half> addrspace(1)* 103; VI-NEXT: [[WIDE_LOAD10:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP7]], align 2 104; VI-NEXT: [[TMP8:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 8 105; VI-NEXT: [[TMP9:%.*]] = bitcast half addrspace(1)* [[TMP8]] to <2 x half> addrspace(1)* 106; VI-NEXT: [[WIDE_LOAD11:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP9]], align 2 107; VI-NEXT: [[TMP10:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 10 108; VI-NEXT: [[TMP11:%.*]] = bitcast half addrspace(1)* [[TMP10]] to <2 x half> addrspace(1)* 109; VI-NEXT: [[WIDE_LOAD12:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP11]], align 2 110; VI-NEXT: [[TMP12:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 12 111; VI-NEXT: [[TMP13:%.*]] = bitcast half addrspace(1)* [[TMP12]] to <2 x half> addrspace(1)* 112; VI-NEXT: [[WIDE_LOAD13:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP13]], align 2 113; VI-NEXT: [[TMP14:%.*]] = getelementptr inbounds half, half addrspace(1)* [[TMP0]], i64 14 114; VI-NEXT: [[TMP15:%.*]] = bitcast half addrspace(1)* [[TMP14]] to <2 x half> addrspace(1)* 115; VI-NEXT: [[WIDE_LOAD14:%.*]] = load <2 x half>, <2 x half> addrspace(1)* [[TMP15]], align 2 116; VI-NEXT: [[TMP16]] = fadd fast <2 x half> [[VEC_PHI]], [[WIDE_LOAD]] 117; VI-NEXT: [[TMP17]] = fadd fast <2 x half> [[VEC_PHI1]], [[WIDE_LOAD8]] 118; VI-NEXT: [[TMP18]] = fadd fast <2 x half> [[VEC_PHI2]], [[WIDE_LOAD9]] 119; VI-NEXT: [[TMP19]] = fadd fast <2 x half> [[VEC_PHI3]], [[WIDE_LOAD10]] 120; VI-NEXT: [[TMP20]] = fadd fast <2 x half> [[VEC_PHI4]], [[WIDE_LOAD11]] 121; VI-NEXT: [[TMP21]] = fadd fast <2 x half> [[VEC_PHI5]], [[WIDE_LOAD12]] 122; VI-NEXT: [[TMP22]] = fadd fast <2 x half> [[VEC_PHI6]], [[WIDE_LOAD13]] 123; VI-NEXT: [[TMP23]] = fadd fast <2 x half> [[VEC_PHI7]], [[WIDE_LOAD14]] 124; VI-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16 125; VI-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256 126; VI-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]] 127; VI: middle.block: 128; VI-NEXT: [[BIN_RDX:%.*]] = fadd fast <2 x half> [[TMP17]], [[TMP16]] 129; VI-NEXT: [[BIN_RDX15:%.*]] = fadd fast <2 x half> [[TMP18]], [[BIN_RDX]] 130; VI-NEXT: [[BIN_RDX16:%.*]] = fadd fast <2 x half> [[TMP19]], [[BIN_RDX15]] 131; VI-NEXT: [[BIN_RDX17:%.*]] = fadd fast <2 x half> [[TMP20]], [[BIN_RDX16]] 132; VI-NEXT: [[BIN_RDX18:%.*]] = fadd fast <2 x half> [[TMP21]], [[BIN_RDX17]] 133; VI-NEXT: [[BIN_RDX19:%.*]] = fadd fast <2 x half> [[TMP22]], [[BIN_RDX18]] 134; VI-NEXT: [[BIN_RDX20:%.*]] = fadd fast <2 x half> [[TMP23]], [[BIN_RDX19]] 135; VI-NEXT: [[RDX_SHUF:%.*]] = shufflevector <2 x half> [[BIN_RDX20]], <2 x half> undef, <2 x i32> <i32 1, i32 undef> 136; VI-NEXT: [[BIN_RDX21:%.*]] = fadd fast <2 x half> [[BIN_RDX20]], [[RDX_SHUF]] 137; VI-NEXT: [[TMP25:%.*]] = extractelement <2 x half> [[BIN_RDX21]], i32 0 138; VI-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 139; VI: scalar.ph: 140; VI-NEXT: br label [[FOR_BODY:%.*]] 141; VI: for.body: 142; VI-NEXT: br i1 undef, label [[FOR_END]], label [[FOR_BODY]], [[LOOP2:!llvm.loop !.*]] 143; VI: for.end: 144; VI-NEXT: [[ADD_LCSSA:%.*]] = phi half [ undef, [[FOR_BODY]] ], [ [[TMP25]], [[MIDDLE_BLOCK]] ] 145; VI-NEXT: ret half [[ADD_LCSSA]] 146; 147; CI-LABEL: @vectorize_v2f16_loop( 148; CI-NEXT: entry: 149; CI-NEXT: br label [[FOR_BODY:%.*]] 150; CI: for.body: 151; CI-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 152; CI-NEXT: [[Q_04:%.*]] = phi half [ 0xH0000, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 153; CI-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds half, half addrspace(1)* [[S:%.*]], i64 [[INDVARS_IV]] 154; CI-NEXT: [[TMP0:%.*]] = load half, half addrspace(1)* [[ARRAYIDX]], align 2 155; CI-NEXT: [[ADD]] = fadd fast half [[Q_04]], [[TMP0]] 156; CI-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 157; CI-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 256 158; CI-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]] 159; CI: for.end: 160; CI-NEXT: ret half [[ADD]] 161; 162entry: 163 br label %for.body 164 165for.body: 166 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 167 %q.04 = phi half [ 0.0, %entry ], [ %add, %for.body ] 168 %arrayidx = getelementptr inbounds half, half addrspace(1)* %s, i64 %indvars.iv 169 %0 = load half, half addrspace(1)* %arrayidx, align 2 170 %add = fadd fast half %q.04, %0 171 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 172 %exitcond = icmp eq i64 %indvars.iv.next, 256 173 br i1 %exitcond, label %for.end, label %for.body 174 175for.end: 176 %add.lcssa = phi half [ %add, %for.body ] 177 ret half %add.lcssa 178} 179