| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 |
| ; RUN: opt -S -passes=loop-vectorize -force-vector-width=2 \ |
| ; RUN: -force-vector-interleave=2 < %s | FileCheck %s |
| |
| ; When interleaving, the reduction parts are combined with a synthesized |
| ; fcmp+select built by createMinMaxOp. That select is not derived from any |
| ; existing branch, so in a profiled function its branch weights must be marked |
| ; explicitly unknown to keep profile verification happy. |
| define float @fmax_reduction_profiled(ptr %a, i64 %n) !prof !0 { |
| ; CHECK-LABEL: define float @fmax_reduction_profiled( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) !prof [[PROF0:![0-9]+]] { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x float> [ splat (float -1.000000e+00), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <2 x float> [ splat (float -1.000000e+00), %[[VECTOR_PH]] ], [ [[TMP5:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[TMP0]], i64 2 |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP0]], align 4 |
| ; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x float>, ptr [[TMP1]], align 4 |
| ; CHECK-NEXT: [[TMP2:%.*]] = fcmp fast ogt <2 x float> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = fcmp fast ogt <2 x float> [[WIDE_LOAD2]], [[VEC_PHI1]] |
| ; CHECK-NEXT: [[TMP4]] = select nnan nsz <2 x i1> [[TMP2]], <2 x float> [[WIDE_LOAD]], <2 x float> [[VEC_PHI]] |
| ; CHECK-NEXT: [[TMP5]] = select nnan nsz <2 x i1> [[TMP3]], <2 x float> [[WIDE_LOAD2]], <2 x float> [[VEC_PHI1]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP1:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <2 x float> [[TMP4]], [[TMP5]] |
| ; CHECK-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select fast <2 x i1> [[RDX_MINMAX_CMP]], <2 x float> [[TMP4]], <2 x float> [[TMP5]], !prof [[PROF4:![0-9]+]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = call fast float @llvm.vector.reduce.fmax.v2f32(<2 x float> [[RDX_MINMAX_SELECT]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP7]], %[[MIDDLE_BLOCK]] ], [ -1.000000e+00, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[FOR_BODY:.*]] |
| ; CHECK: [[FOR_BODY]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ] |
| ; CHECK-NEXT: [[MAX:%.*]] = phi float [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[MAX_NEXT:%.*]], %[[FOR_BODY]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LD:%.*]] = load float, ptr [[GEP]], align 4 |
| ; CHECK-NEXT: [[CMP:%.*]] = fcmp fast ogt float [[LD]], [[MAX]] |
| ; CHECK-NEXT: [[MAX_NEXT]] = select nnan nsz i1 [[CMP]], float [[LD]], float [[MAX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] |
| ; CHECK: [[EXIT]]: |
| ; CHECK-NEXT: [[MAX_NEXT_LCSSA:%.*]] = phi float [ [[MAX_NEXT]], %[[FOR_BODY]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret float [[MAX_NEXT_LCSSA]] |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] |
| %max = phi float [ -1.000000e+00, %entry ], [ %max.next, %for.body ] |
| %gep = getelementptr inbounds float, ptr %a, i64 %iv |
| %ld = load float, ptr %gep, align 4 |
| %cmp = fcmp fast ogt float %ld, %max |
| %max.next = select nnan nsz i1 %cmp, float %ld, float %max |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %ec = icmp eq i64 %iv.next, %n |
| br i1 %ec, label %exit, label %for.body |
| |
| exit: |
| ret float %max.next |
| } |
| |
| ; Without an entry count the function is not profiled, so the synthesized |
| ; select must not be annotated. |
| define float @fmax_reduction_not_profiled(ptr %a, i64 %n) { |
| ; CHECK-LABEL: define float @fmax_reduction_not_profiled( |
| ; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { |
| ; CHECK-NEXT: [[ENTRY:.*]]: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| ; CHECK: [[VECTOR_PH]]: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| ; CHECK: [[VECTOR_BODY]]: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x float> [ splat (float -1.000000e+00), %[[VECTOR_PH]] ], [ [[TMP4:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <2 x float> [ splat (float -1.000000e+00), %[[VECTOR_PH]] ], [ [[TMP5:%.*]], %[[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[TMP0]], i64 2 |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP0]], align 4 |
| ; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x float>, ptr [[TMP1]], align 4 |
| ; CHECK-NEXT: [[TMP2:%.*]] = fcmp fast ogt <2 x float> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = fcmp fast ogt <2 x float> [[WIDE_LOAD2]], [[VEC_PHI1]] |
| ; CHECK-NEXT: [[TMP4]] = select nnan nsz <2 x i1> [[TMP2]], <2 x float> [[WIDE_LOAD]], <2 x float> [[VEC_PHI]] |
| ; CHECK-NEXT: [[TMP5]] = select nnan nsz <2 x i1> [[TMP3]], <2 x float> [[WIDE_LOAD2]], <2 x float> [[VEC_PHI1]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] |
| ; CHECK: [[MIDDLE_BLOCK]]: |
| ; CHECK-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <2 x float> [[TMP4]], [[TMP5]] |
| ; CHECK-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select fast <2 x i1> [[RDX_MINMAX_CMP]], <2 x float> [[TMP4]], <2 x float> [[TMP5]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = call fast float @llvm.vector.reduce.fmax.v2f32(<2 x float> [[RDX_MINMAX_SELECT]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] |
| ; CHECK: [[SCALAR_PH]]: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP7]], %[[MIDDLE_BLOCK]] ], [ -1.000000e+00, %[[ENTRY]] ] |
| ; CHECK-NEXT: br label %[[FOR_BODY:.*]] |
| ; CHECK: [[FOR_BODY]]: |
| ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ] |
| ; CHECK-NEXT: [[MAX:%.*]] = phi float [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[MAX_NEXT:%.*]], %[[FOR_BODY]] ] |
| ; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] |
| ; CHECK-NEXT: [[LD:%.*]] = load float, ptr [[GEP]], align 4 |
| ; CHECK-NEXT: [[CMP:%.*]] = fcmp fast ogt float [[LD]], [[MAX]] |
| ; CHECK-NEXT: [[MAX_NEXT]] = select nnan nsz i1 [[CMP]], float [[LD]], float [[MAX]] |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] |
| ; CHECK: [[EXIT]]: |
| ; CHECK-NEXT: [[MAX_NEXT_LCSSA:%.*]] = phi float [ [[MAX_NEXT]], %[[FOR_BODY]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret float [[MAX_NEXT_LCSSA]] |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] |
| %max = phi float [ -1.000000e+00, %entry ], [ %max.next, %for.body ] |
| %gep = getelementptr inbounds float, ptr %a, i64 %iv |
| %ld = load float, ptr %gep, align 4 |
| %cmp = fcmp fast ogt float %ld, %max |
| %max.next = select nnan nsz i1 %cmp, float %ld, float %max |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %ec = icmp eq i64 %iv.next, %n |
| br i1 %ec, label %exit, label %for.body |
| |
| exit: |
| ret float %max.next |
| } |
| |
| |
| !0 = !{!"function_entry_count", i64 1000} |
| ;. |
| ; CHECK: [[PROF0]] = !{!"function_entry_count", i64 1000} |
| ; CHECK: [[LOOP1]] = distinct !{[[LOOP1]], [[META2:![0-9]+]], [[META3:![0-9]+]]} |
| ; CHECK: [[META2]] = !{!"llvm.loop.isvectorized", i32 1} |
| ; CHECK: [[META3]] = !{!"llvm.loop.unroll.runtime.disable"} |
| ; CHECK: [[PROF4]] = !{!"unknown", !"loop-utils"} |
| ; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META3]], [[META2]]} |
| ; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META2]], [[META3]]} |
| ; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META3]], [[META2]]} |
| ;. |
