| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| ; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -force-widen-divrem-via-safe-divisor=0 -S | FileCheck %s |
| ; RUN: opt < %s -passes=loop-vectorize,instcombine -force-vector-interleave=1 -force-vector-width=2 -force-widen-divrem-via-safe-divisor=0 -S | FileCheck %s --check-prefix=IND |
| ; RUN: opt < %s -passes=loop-vectorize,instcombine -force-vector-interleave=2 -force-vector-width=2 -force-widen-divrem-via-safe-divisor=0 -S | FileCheck %s --check-prefix=UNROLL |
| ; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=2 -force-widen-divrem-via-safe-divisor=0 -S | FileCheck %s --check-prefix=UNROLL-NO-IC |
| ; RUN: opt < %s -passes=loop-vectorize,instcombine -force-vector-interleave=2 -force-vector-width=4 -force-widen-divrem-via-safe-divisor=0 -enable-interleaved-mem-accesses -S | FileCheck %s --check-prefix=INTERLEAVE |
| |
| target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" |
| |
| ; Make sure that we can handle multiple integer induction variables. |
| ; |
| define void @multi_int_induction(ptr %A, i32 %N) { |
| ; CHECK-LABEL: @multi_int_induction( |
| ; CHECK-NEXT: for.body.lr.ph: |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 |
| ; CHECK-NEXT: [[IND_END:%.*]] = add i32 190, [[DOTCAST]] |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 0 |
| ; CHECK-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP5]], align 4 |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ] |
| ; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; CHECK-NEXT: [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] |
| ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]] |
| ; CHECK-NEXT: store i32 [[COUNT_09]], ptr [[ARRAYIDX2]], align 4 |
| ; CHECK-NEXT: [[INC]] = add nsw i32 [[COUNT_09]], 1 |
| ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 |
| ; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 |
| ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @multi_int_induction( |
| ; IND-NEXT: for.body.lr.ph: |
| ; IND-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; IND-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; IND-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590 |
| ; IND-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 |
| ; IND-NEXT: [[IND_END:%.*]] = add i32 [[DOTCAST]], 190 |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] |
| ; IND-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP3]], align 4 |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; IND-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ] |
| ; IND-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]] |
| ; IND-NEXT: store i32 [[COUNT_09]], ptr [[ARRAYIDX2]], align 4 |
| ; IND-NEXT: [[INC]] = add nsw i32 [[COUNT_09]], 1 |
| ; IND-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 |
| ; IND-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 |
| ; IND-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[N]], [[LFTR_WIDEIV]] |
| ; IND-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @multi_int_induction( |
| ; UNROLL-NEXT: for.body.lr.ph: |
| ; UNROLL-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588 |
| ; UNROLL-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 |
| ; UNROLL-NEXT: [[IND_END:%.*]] = add i32 [[DOTCAST]], 190 |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; UNROLL-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] |
| ; UNROLL-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP3]], i64 8 |
| ; UNROLL-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP3]], align 4 |
| ; UNROLL-NEXT: store <2 x i32> [[STEP_ADD]], ptr [[TMP4]], align 4 |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 4) |
| ; UNROLL-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ] |
| ; UNROLL-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]] |
| ; UNROLL-NEXT: store i32 [[COUNT_09]], ptr [[ARRAYIDX2]], align 4 |
| ; UNROLL-NEXT: [[INC]] = add nsw i32 [[COUNT_09]], 1 |
| ; UNROLL-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 |
| ; UNROLL-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 |
| ; UNROLL-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[N]], [[LFTR_WIDEIV]] |
| ; UNROLL-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @multi_int_induction( |
| ; UNROLL-NO-IC-NEXT: for.body.lr.ph: |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[IND_END:%.*]] = add i32 190, [[DOTCAST]] |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP3]] |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 2 |
| ; UNROLL-NO-IC-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP5]], align 4 |
| ; UNROLL-NO-IC-NEXT: store <2 x i32> [[STEP_ADD]], ptr [[TMP6]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], splat (i32 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]] |
| ; UNROLL-NO-IC-NEXT: store i32 [[COUNT_09]], ptr [[ARRAYIDX2]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[INC]] = add nsw i32 [[COUNT_09]], 1 |
| ; UNROLL-NO-IC-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 |
| ; UNROLL-NO-IC-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: ret void |
| ; |
| ; INTERLEAVE-LABEL: @multi_int_induction( |
| ; INTERLEAVE-NEXT: for.body.lr.ph: |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934584 |
| ; INTERLEAVE-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32 |
| ; INTERLEAVE-NEXT: [[IND_END:%.*]] = add i32 [[DOTCAST]], 190 |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 190, i32 191, i32 192, i32 193>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], splat (i32 4) |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]] |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP3]], i64 16 |
| ; INTERLEAVE-NEXT: store <4 x i32> [[VEC_IND]], ptr [[TMP3]], align 4 |
| ; INTERLEAVE-NEXT: store <4 x i32> [[STEP_ADD]], ptr [[TMP4]], align 4 |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 8) |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ] |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]] |
| ; INTERLEAVE-NEXT: store i32 [[COUNT_09]], ptr [[ARRAYIDX2]], align 4 |
| ; INTERLEAVE-NEXT: [[INC]] = add nsw i32 [[COUNT_09]], 1 |
| ; INTERLEAVE-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 |
| ; INTERLEAVE-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 |
| ; INTERLEAVE-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[N]], [[LFTR_WIDEIV]] |
| ; INTERLEAVE-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: ret void |
| ; |
| for.body.lr.ph: |
| br label %for.body |
| |
| for.body: |
| %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] |
| %count.09 = phi i32 [ 190, %for.body.lr.ph ], [ %inc, %for.body ] |
| %arrayidx2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv |
| store i32 %count.09, ptr %arrayidx2, align 4 |
| %inc = add nsw i32 %count.09, 1 |
| %indvars.iv.next = add i64 %indvars.iv, 1 |
| %lftr.wideiv = trunc i64 %indvars.iv.next to i32 |
| %exitcond = icmp ne i32 %lftr.wideiv, %N |
| br i1 %exitcond, label %for.body, label %for.end |
| |
| for.end: |
| ret void |
| } |
| |
| ; Make sure we remove unneeded vectorization of induction variables. |
| ; In order for instcombine to cleanup the vectorized induction variables that we |
| ; create in the loop vectorizer we need to perform some form of redundancy |
| ; elimination to get rid of multiple uses. |
| |
| |
| ; Vectorized induction variable. |
| |
| define void @scalar_use(ptr %a, float %b, i64 %offset, i64 %offset2, i64 %n) { |
| ; CHECK-LABEL: @scalar_use( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; CHECK: vector.memcheck: |
| ; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2 |
| ; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]] |
| ; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[TMP0]] |
| ; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP2]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2 |
| ; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[TMP1]], [[TMP3]] |
| ; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP4]] |
| ; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]] |
| ; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]] |
| ; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[TMP5]], [[OFFSET]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP6]] |
| ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i32 0 |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP8]], align 4, !alias.scope [[META4:![0-9]+]], !noalias [[META7:![0-9]+]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP5]], [[OFFSET2]] |
| ; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP9]] |
| ; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, ptr [[TMP10]], i32 0 |
| ; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x float>, ptr [[TMP11]], align 4, !alias.scope [[META7]] |
| ; CHECK-NEXT: [[TMP12:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD4]] |
| ; CHECK-NEXT: [[TMP13:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP12]] |
| ; CHECK-NEXT: store <2 x float> [[TMP13]], ptr [[TMP8]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[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: [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]] |
| ; CHECK-NEXT: [[ARR_IDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM]] |
| ; CHECK-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4 |
| ; CHECK-NEXT: [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]] |
| ; CHECK-NEXT: [[ARR_IDX2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM2]] |
| ; CHECK-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4 |
| ; CHECK-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]] |
| ; CHECK-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]] |
| ; CHECK-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4 |
| ; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; CHECK: loopexit: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @scalar_use( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; IND: vector.memcheck: |
| ; IND-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2 |
| ; IND-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]] |
| ; IND-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; IND-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; IND-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP0]] |
| ; IND-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2 |
| ; IND-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]] |
| ; IND-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; IND-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]] |
| ; IND-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]] |
| ; IND-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]] |
| ; IND-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; IND-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[N]], -2 |
| ; IND-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0 |
| ; IND-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP5:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; IND-NEXT: [[TMP6:%.*]] = getelementptr float, ptr [[TMP5]], i64 [[OFFSET]] |
| ; IND-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP6]], align 4, !alias.scope [[META4:![0-9]+]], !noalias [[META7:![0-9]+]] |
| ; IND-NEXT: [[TMP7:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; IND-NEXT: [[TMP8:%.*]] = getelementptr float, ptr [[TMP7]], i64 [[OFFSET2]] |
| ; IND-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x float>, ptr [[TMP8]], align 4, !alias.scope [[META7]] |
| ; IND-NEXT: [[TMP9:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD4]] |
| ; IND-NEXT: [[TMP10:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP9]] |
| ; IND-NEXT: store <2 x float> [[TMP10]], ptr [[TMP6]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[TMP12:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; IND-NEXT: [[ARR_IDX:%.*]] = getelementptr float, ptr [[TMP12]], i64 [[OFFSET]] |
| ; IND-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4 |
| ; IND-NEXT: [[TMP13:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; IND-NEXT: [[ARR_IDX2:%.*]] = getelementptr float, ptr [[TMP13]], i64 [[OFFSET2]] |
| ; IND-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4 |
| ; IND-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]] |
| ; IND-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]] |
| ; IND-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4 |
| ; IND-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; IND-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; IND-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; IND: loopexit: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @scalar_use( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; UNROLL: vector.memcheck: |
| ; UNROLL-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2 |
| ; UNROLL-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]] |
| ; UNROLL-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; UNROLL-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP0]] |
| ; UNROLL-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2 |
| ; UNROLL-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]] |
| ; UNROLL-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; UNROLL-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]] |
| ; UNROLL-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]] |
| ; UNROLL-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]] |
| ; UNROLL-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; UNROLL-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[N]], -4 |
| ; UNROLL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0 |
| ; UNROLL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP5:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; UNROLL-NEXT: [[TMP6:%.*]] = getelementptr float, ptr [[TMP5]], i64 [[OFFSET]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP6]], i64 8 |
| ; UNROLL-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP6]], align 4, !alias.scope [[META4:![0-9]+]], !noalias [[META7:![0-9]+]] |
| ; UNROLL-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x float>, ptr [[TMP7]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NEXT: [[TMP8:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; UNROLL-NEXT: [[TMP9:%.*]] = getelementptr float, ptr [[TMP8]], i64 [[OFFSET2]] |
| ; UNROLL-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP9]], i64 8 |
| ; UNROLL-NEXT: [[WIDE_LOAD5:%.*]] = load <2 x float>, ptr [[TMP9]], align 4, !alias.scope [[META7]] |
| ; UNROLL-NEXT: [[WIDE_LOAD6:%.*]] = load <2 x float>, ptr [[TMP10]], align 4, !alias.scope [[META7]] |
| ; UNROLL-NEXT: [[TMP11:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD5]] |
| ; UNROLL-NEXT: [[TMP12:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD6]] |
| ; UNROLL-NEXT: [[TMP13:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP11]] |
| ; UNROLL-NEXT: [[TMP14:%.*]] = fadd fast <2 x float> [[WIDE_LOAD4]], [[TMP12]] |
| ; UNROLL-NEXT: store <2 x float> [[TMP13]], ptr [[TMP6]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NEXT: store <2 x float> [[TMP14]], ptr [[TMP7]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP16:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; UNROLL-NEXT: [[ARR_IDX:%.*]] = getelementptr float, ptr [[TMP16]], i64 [[OFFSET]] |
| ; UNROLL-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4 |
| ; UNROLL-NEXT: [[TMP17:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; UNROLL-NEXT: [[ARR_IDX2:%.*]] = getelementptr float, ptr [[TMP17]], i64 [[OFFSET2]] |
| ; UNROLL-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4 |
| ; UNROLL-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]] |
| ; UNROLL-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]] |
| ; UNROLL-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4 |
| ; UNROLL-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; UNROLL-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; UNROLL-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; UNROLL: loopexit: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalar_use( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; UNROLL-NO-IC: vector.memcheck: |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2 |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP2]] |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2 |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]] |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = add i64 [[TMP1]], [[TMP3]] |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP4]] |
| ; UNROLL-NO-IC-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]] |
| ; UNROLL-NO-IC-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]] |
| ; UNROLL-NO-IC-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add i64 [[TMP5]], [[OFFSET]] |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP6]] |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i32 2 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP8]], align 4, !alias.scope [[META4:![0-9]+]], !noalias [[META7:![0-9]+]] |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x float>, ptr [[TMP9]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = add i64 [[TMP5]], [[OFFSET2]] |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP10]] |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, ptr [[TMP11]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, ptr [[TMP11]], i32 2 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD5:%.*]] = load <2 x float>, ptr [[TMP12]], align 4, !alias.scope [[META7]] |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD6:%.*]] = load <2 x float>, ptr [[TMP13]], align 4, !alias.scope [[META7]] |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD5]] |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD6]] |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP14]] |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = fadd fast <2 x float> [[WIDE_LOAD4]], [[TMP15]] |
| ; UNROLL-NO-IC-NEXT: store <2 x float> [[TMP16]], ptr [[TMP8]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NO-IC-NEXT: store <2 x float> [[TMP17]], ptr [[TMP9]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]] |
| ; UNROLL-NO-IC-NEXT: [[ARR_IDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM]] |
| ; UNROLL-NO-IC-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]] |
| ; UNROLL-NO-IC-NEXT: [[ARR_IDX2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IND_SUM2]] |
| ; UNROLL-NO-IC-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]] |
| ; UNROLL-NO-IC-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]] |
| ; UNROLL-NO-IC-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; UNROLL-NO-IC-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; UNROLL-NO-IC: loopexit: |
| ; UNROLL-NO-IC-NEXT: ret void |
| ; |
| ; INTERLEAVE-LABEL: @scalar_use( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; INTERLEAVE: vector.memcheck: |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = shl i64 [[OFFSET:%.*]], 2 |
| ; INTERLEAVE-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP0]] |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = shl i64 [[N]], 2 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; INTERLEAVE-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP2]], i64 [[TMP0]] |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = shl i64 [[OFFSET2:%.*]], 2 |
| ; INTERLEAVE-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]] |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]] |
| ; INTERLEAVE-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP4]], i64 [[TMP3]] |
| ; INTERLEAVE-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP3]] |
| ; INTERLEAVE-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP1]] |
| ; INTERLEAVE-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; INTERLEAVE-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i64 [[N]], -8 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[B:%.*]], i64 0 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = getelementptr float, ptr [[TMP5]], i64 [[OFFSET]] |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP6]], i64 16 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP6]], align 4, !alias.scope [[META4:![0-9]+]], !noalias [[META7:![0-9]+]] |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x float>, ptr [[TMP7]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = getelementptr float, ptr [[A]], i64 [[INDEX]] |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr float, ptr [[TMP8]], i64 [[OFFSET2]] |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP9]], i64 16 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD5:%.*]] = load <4 x float>, ptr [[TMP9]], align 4, !alias.scope [[META7]] |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x float>, ptr [[TMP10]], align 4, !alias.scope [[META7]] |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = fmul fast <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD5]] |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = fmul fast <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD6]] |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = fadd fast <4 x float> [[WIDE_LOAD]], [[TMP11]] |
| ; INTERLEAVE-NEXT: [[TMP14:%.*]] = fadd fast <4 x float> [[WIDE_LOAD4]], [[TMP12]] |
| ; INTERLEAVE-NEXT: store <4 x float> [[TMP13]], ptr [[TMP6]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; INTERLEAVE-NEXT: store <4 x float> [[TMP14]], ptr [[TMP7]], align 4, !alias.scope [[META4]], !noalias [[META7]] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; INTERLEAVE-NEXT: [[ARR_IDX:%.*]] = getelementptr float, ptr [[TMP16]], i64 [[OFFSET]] |
| ; INTERLEAVE-NEXT: [[L1:%.*]] = load float, ptr [[ARR_IDX]], align 4 |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = getelementptr float, ptr [[A]], i64 [[IV]] |
| ; INTERLEAVE-NEXT: [[ARR_IDX2:%.*]] = getelementptr float, ptr [[TMP17]], i64 [[OFFSET2]] |
| ; INTERLEAVE-NEXT: [[L2:%.*]] = load float, ptr [[ARR_IDX2]], align 4 |
| ; INTERLEAVE-NEXT: [[M:%.*]] = fmul fast float [[B]], [[L2]] |
| ; INTERLEAVE-NEXT: [[AD:%.*]] = fadd fast float [[L1]], [[M]] |
| ; INTERLEAVE-NEXT: store float [[AD]], ptr [[ARR_IDX]], align 4 |
| ; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 |
| ; INTERLEAVE-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] |
| ; INTERLEAVE-NEXT: br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] |
| ; INTERLEAVE: loopexit: |
| ; INTERLEAVE-NEXT: ret void |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] |
| %ind.sum = add i64 %iv, %offset |
| %arr.idx = getelementptr inbounds float, ptr %a, i64 %ind.sum |
| %l1 = load float, ptr %arr.idx, align 4 |
| %ind.sum2 = add i64 %iv, %offset2 |
| %arr.idx2 = getelementptr inbounds float, ptr %a, i64 %ind.sum2 |
| %l2 = load float, ptr %arr.idx2, align 4 |
| %m = fmul fast float %b, %l2 |
| %ad = fadd fast float %l1, %m |
| store float %ad, ptr %arr.idx, align 4 |
| %iv.next = add nuw nsw i64 %iv, 1 |
| %exitcond = icmp eq i64 %iv.next, %n |
| br i1 %exitcond, label %loopexit, label %for.body |
| |
| loopexit: |
| ret void |
| } |
| |
| ; Make sure we don't create a vector induction phi node that is unused. |
| ; Scalarize the step vectors instead. |
| ; |
| ; for (int i = 0; i < n; ++i) |
| ; sum += a[i]; |
| ; |
| ; |
| ; |
| ; |
| |
| define i64 @scalarize_induction_variable_01(ptr %a, i64 %n) { |
| ; CHECK-LABEL: @scalarize_induction_variable_01( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[SMAX]], [[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 i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] |
| ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 0 |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8 |
| ; CHECK-NEXT: [[TMP3]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP3]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], 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 i64 [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[SUM:%.*]] = phi i64 [ [[TMP8:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[I]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = load i64, ptr [[TMP6]], align 8 |
| ; CHECK-NEXT: [[TMP8]] = add i64 [[TMP7]], [[SUM]] |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: [[TMP9:%.*]] = phi i64 [ [[TMP8]], [[FOR_BODY]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i64 [[TMP9]] |
| ; |
| ; IND-LABEL: @scalarize_induction_variable_01( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 2 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775806 |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP0:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]] |
| ; IND-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP0]], align 8 |
| ; IND-NEXT: [[TMP1]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[TMP3:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP1]]) |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[SUM:%.*]] = phi i64 [ [[TMP6:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw i64, ptr [[A]], i64 [[I]] |
| ; IND-NEXT: [[TMP5:%.*]] = load i64, ptr [[TMP4]], align 8 |
| ; IND-NEXT: [[TMP6]] = add i64 [[TMP5]], [[SUM]] |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; IND-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: [[TMP7:%.*]] = phi i64 [ [[TMP6]], [[FOR_BODY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] |
| ; IND-NEXT: ret i64 [[TMP7]] |
| ; |
| ; UNROLL-LABEL: @scalarize_induction_variable_01( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775804 |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP2:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP0:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]] |
| ; UNROLL-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 16 |
| ; UNROLL-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP0]], align 8 |
| ; UNROLL-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i64>, ptr [[TMP1]], align 8 |
| ; UNROLL-NEXT: [[TMP2]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; UNROLL-NEXT: [[TMP3]] = add <2 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[TMP3]], [[TMP2]] |
| ; UNROLL-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]]) |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[SUM:%.*]] = phi i64 [ [[TMP8:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw i64, ptr [[A]], i64 [[I]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = load i64, ptr [[TMP6]], align 8 |
| ; UNROLL-NEXT: [[TMP8]] = add i64 [[TMP7]], [[SUM]] |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: [[TMP9:%.*]] = phi i64 [ [[TMP8]], [[FOR_BODY]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NEXT: ret i64 [[TMP9]] |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_01( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 2 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP2]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i64>, ptr [[TMP3]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP4]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; UNROLL-NO-IC-NEXT: [[TMP5]] = add <2 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[TMP5]], [[TMP4]] |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: [[SUM:%.*]] = phi i64 [ [[TMP10:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[I]] |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP8]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP10]] = add i64 [[TMP9]], [[SUM]] |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[FOR_BODY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret i64 [[TMP11]] |
| ; |
| ; INTERLEAVE-LABEL: @scalarize_induction_variable_01( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 8 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775800 |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP2:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]] |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 32 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP0]], align 8 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP2]] = add <4 x i64> [[WIDE_LOAD]], [[VEC_PHI]] |
| ; INTERLEAVE-NEXT: [[TMP3]] = add <4 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[BIN_RDX:%.*]] = add <4 x i64> [[TMP3]], [[TMP2]] |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[BIN_RDX]]) |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; INTERLEAVE-NEXT: [[SUM:%.*]] = phi i64 [ [[TMP8:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw i64, ptr [[A]], i64 [[I]] |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = load i64, ptr [[TMP6]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP8]] = add i64 [[TMP7]], [[SUM]] |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = phi i64 [ [[TMP8]], [[FOR_BODY]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ] |
| ; INTERLEAVE-NEXT: ret i64 [[TMP9]] |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] |
| %sum = phi i64 [ %2, %for.body ], [ 0, %entry ] |
| %0 = getelementptr inbounds i64, ptr %a, i64 %i |
| %1 = load i64, ptr %0, align 8 |
| %2 = add i64 %1, %sum |
| %i.next = add nuw nsw i64 %i, 1 |
| %cond = icmp slt i64 %i.next, %n |
| br i1 %cond, label %for.body, label %for.end |
| |
| for.end: |
| %3 = phi i64 [ %2, %for.body ] |
| ret i64 %3 |
| } |
| |
| ; Make sure we scalarize the step vectors used for the pointer arithmetic. We |
| ; can't easily simplify vectorized step vectors. |
| ; |
| ; float s = 0; |
| ; for (int i ; 0; i < n; i += 8) |
| ; s += (a[i] + b[i] + 1.0f); |
| ; |
| ; |
| ; |
| ; |
| |
| define float @scalarize_induction_variable_02(ptr %a, ptr %b, i64 %n) { |
| ; CHECK-LABEL: @scalarize_induction_variable_02( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8) |
| ; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1 |
| ; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 3 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[IND_END:%.*]] = mul i64 [[N_VEC]], 8 |
| ; 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> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8 |
| ; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0 |
| ; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 8 |
| ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = load float, ptr [[TMP5]], align 4 |
| ; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[TMP6]], align 4 |
| ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0 |
| ; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP8]], i32 1 |
| ; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[TMP3]] |
| ; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP4]] |
| ; CHECK-NEXT: [[TMP13:%.*]] = load float, ptr [[TMP11]], align 4 |
| ; CHECK-NEXT: [[TMP14:%.*]] = load float, ptr [[TMP12]], align 4 |
| ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i32 0 |
| ; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x float> [[TMP15]], float [[TMP14]], i32 1 |
| ; CHECK-NEXT: [[TMP17:%.*]] = fadd fast <2 x float> [[VEC_PHI]], splat (float 1.000000e+00) |
| ; CHECK-NEXT: [[TMP18:%.*]] = fadd fast <2 x float> [[TMP17]], [[TMP10]] |
| ; CHECK-NEXT: [[TMP19]] = fadd fast <2 x float> [[TMP18]], [[TMP16]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP21:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float 0.000000e+00, <2 x float> [[TMP19]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP21]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ] |
| ; CHECK-NEXT: [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP28:%.*]], [[FOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I]] |
| ; CHECK-NEXT: [[TMP23:%.*]] = load float, ptr [[TMP22]], align 4 |
| ; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[I]] |
| ; CHECK-NEXT: [[TMP25:%.*]] = load float, ptr [[TMP24]], align 4 |
| ; CHECK-NEXT: [[TMP26:%.*]] = fadd fast float [[S]], 1.000000e+00 |
| ; CHECK-NEXT: [[TMP27:%.*]] = fadd fast float [[TMP26]], [[TMP23]] |
| ; CHECK-NEXT: [[TMP28]] = fadd fast float [[TMP27]], [[TMP25]] |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 8 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: [[S_LCSSA:%.*]] = phi float [ [[TMP28]], [[FOR_BODY]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret float [[S_LCSSA]] |
| ; |
| ; IND-LABEL: @scalarize_induction_variable_02( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8) |
| ; IND-NEXT: [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1 |
| ; IND-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 3 |
| ; IND-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 9 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 4611686018427387902 |
| ; IND-NEXT: [[IND_END:%.*]] = shl i64 [[N_VEC]], 3 |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3 |
| ; IND-NEXT: [[TMP3:%.*]] = or disjoint i64 [[OFFSET_IDX]], 8 |
| ; IND-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[OFFSET_IDX]] |
| ; IND-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP3]] |
| ; IND-NEXT: [[TMP6:%.*]] = load float, ptr [[TMP4]], align 4 |
| ; IND-NEXT: [[TMP7:%.*]] = load float, ptr [[TMP5]], align 4 |
| ; IND-NEXT: [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i64 0 |
| ; IND-NEXT: [[TMP9:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP7]], i64 1 |
| ; IND-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[OFFSET_IDX]] |
| ; IND-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP3]] |
| ; IND-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP10]], align 4 |
| ; IND-NEXT: [[TMP13:%.*]] = load float, ptr [[TMP11]], align 4 |
| ; IND-NEXT: [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i64 0 |
| ; IND-NEXT: [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP13]], i64 1 |
| ; IND-NEXT: [[TMP16:%.*]] = fadd fast <2 x float> [[VEC_PHI]], splat (float 1.000000e+00) |
| ; IND-NEXT: [[TMP17:%.*]] = fadd fast <2 x float> [[TMP16]], [[TMP9]] |
| ; IND-NEXT: [[TMP18]] = fadd fast <2 x float> [[TMP17]], [[TMP15]] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[TMP20:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float 0.000000e+00, <2 x float> [[TMP18]]) |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP20]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP27:%.*]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I]] |
| ; IND-NEXT: [[TMP22:%.*]] = load float, ptr [[TMP21]], align 4 |
| ; IND-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[I]] |
| ; IND-NEXT: [[TMP24:%.*]] = load float, ptr [[TMP23]], align 4 |
| ; IND-NEXT: [[TMP25:%.*]] = fadd fast float [[S]], 1.000000e+00 |
| ; IND-NEXT: [[TMP26:%.*]] = fadd fast float [[TMP25]], [[TMP22]] |
| ; IND-NEXT: [[TMP27]] = fadd fast float [[TMP26]], [[TMP24]] |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 8 |
| ; IND-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: [[S_LCSSA:%.*]] = phi float [ [[TMP27]], [[FOR_BODY]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ] |
| ; IND-NEXT: ret float [[S_LCSSA]] |
| ; |
| ; UNROLL-LABEL: @scalarize_induction_variable_02( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8) |
| ; UNROLL-NEXT: [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 3 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 25 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 4611686018427387900 |
| ; UNROLL-NEXT: [[IND_END:%.*]] = shl i64 [[N_VEC]], 3 |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP34:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_PHI1:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP35:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3 |
| ; UNROLL-NEXT: [[TMP3:%.*]] = or disjoint i64 [[OFFSET_IDX]], 8 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = or disjoint i64 [[OFFSET_IDX]], 16 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = or disjoint i64 [[OFFSET_IDX]], 24 |
| ; UNROLL-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[OFFSET_IDX]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP3]] |
| ; UNROLL-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] |
| ; UNROLL-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]] |
| ; UNROLL-NEXT: [[TMP10:%.*]] = load float, ptr [[TMP6]], align 4 |
| ; UNROLL-NEXT: [[TMP11:%.*]] = load float, ptr [[TMP7]], align 4 |
| ; UNROLL-NEXT: [[TMP12:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0 |
| ; UNROLL-NEXT: [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[TMP11]], i64 1 |
| ; UNROLL-NEXT: [[TMP14:%.*]] = load float, ptr [[TMP8]], align 4 |
| ; UNROLL-NEXT: [[TMP15:%.*]] = load float, ptr [[TMP9]], align 4 |
| ; UNROLL-NEXT: [[TMP16:%.*]] = insertelement <2 x float> poison, float [[TMP14]], i64 0 |
| ; UNROLL-NEXT: [[TMP17:%.*]] = insertelement <2 x float> [[TMP16]], float [[TMP15]], i64 1 |
| ; UNROLL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[OFFSET_IDX]] |
| ; UNROLL-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP3]] |
| ; UNROLL-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP4]] |
| ; UNROLL-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP5]] |
| ; UNROLL-NEXT: [[TMP22:%.*]] = load float, ptr [[TMP18]], align 4 |
| ; UNROLL-NEXT: [[TMP23:%.*]] = load float, ptr [[TMP19]], align 4 |
| ; UNROLL-NEXT: [[TMP24:%.*]] = insertelement <2 x float> poison, float [[TMP22]], i64 0 |
| ; UNROLL-NEXT: [[TMP25:%.*]] = insertelement <2 x float> [[TMP24]], float [[TMP23]], i64 1 |
| ; UNROLL-NEXT: [[TMP26:%.*]] = load float, ptr [[TMP20]], align 4 |
| ; UNROLL-NEXT: [[TMP27:%.*]] = load float, ptr [[TMP21]], align 4 |
| ; UNROLL-NEXT: [[TMP28:%.*]] = insertelement <2 x float> poison, float [[TMP26]], i64 0 |
| ; UNROLL-NEXT: [[TMP29:%.*]] = insertelement <2 x float> [[TMP28]], float [[TMP27]], i64 1 |
| ; UNROLL-NEXT: [[TMP30:%.*]] = fadd fast <2 x float> [[VEC_PHI]], splat (float 1.000000e+00) |
| ; UNROLL-NEXT: [[TMP31:%.*]] = fadd fast <2 x float> [[VEC_PHI1]], splat (float 1.000000e+00) |
| ; UNROLL-NEXT: [[TMP32:%.*]] = fadd fast <2 x float> [[TMP30]], [[TMP13]] |
| ; UNROLL-NEXT: [[TMP33:%.*]] = fadd fast <2 x float> [[TMP31]], [[TMP17]] |
| ; UNROLL-NEXT: [[TMP34]] = fadd fast <2 x float> [[TMP32]], [[TMP25]] |
| ; UNROLL-NEXT: [[TMP35]] = fadd fast <2 x float> [[TMP33]], [[TMP29]] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[BIN_RDX:%.*]] = fadd fast <2 x float> [[TMP35]], [[TMP34]] |
| ; UNROLL-NEXT: [[TMP37:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float 0.000000e+00, <2 x float> [[BIN_RDX]]) |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP37]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP44:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP38:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I]] |
| ; UNROLL-NEXT: [[TMP39:%.*]] = load float, ptr [[TMP38]], align 4 |
| ; UNROLL-NEXT: [[TMP40:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[I]] |
| ; UNROLL-NEXT: [[TMP41:%.*]] = load float, ptr [[TMP40]], align 4 |
| ; UNROLL-NEXT: [[TMP42:%.*]] = fadd fast float [[S]], 1.000000e+00 |
| ; UNROLL-NEXT: [[TMP43:%.*]] = fadd fast float [[TMP42]], [[TMP39]] |
| ; UNROLL-NEXT: [[TMP44]] = fadd fast float [[TMP43]], [[TMP41]] |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 8 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: [[S_LCSSA:%.*]] = phi float [ [[TMP44]], [[FOR_BODY]] ], [ [[TMP37]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NEXT: ret float [[S_LCSSA]] |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_02( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8) |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[IND_END:%.*]] = mul i64 [[N_VEC]], 8 |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP35:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP36:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 16 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 24 |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP3]] |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]] |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP6]] |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = load float, ptr [[TMP7]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP8]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP12]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = load float, ptr [[TMP9]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = load float, ptr [[TMP10]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP16]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[TMP3]] |
| ; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP4]] |
| ; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP5]] |
| ; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP6]] |
| ; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = load float, ptr [[TMP19]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = load float, ptr [[TMP20]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP23]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP26:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP24]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP27:%.*]] = load float, ptr [[TMP21]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP28:%.*]] = load float, ptr [[TMP22]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP29:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP30:%.*]] = insertelement <2 x float> [[TMP29]], float [[TMP28]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP31:%.*]] = fadd fast <2 x float> [[VEC_PHI]], splat (float 1.000000e+00) |
| ; UNROLL-NO-IC-NEXT: [[TMP32:%.*]] = fadd fast <2 x float> [[VEC_PHI1]], splat (float 1.000000e+00) |
| ; UNROLL-NO-IC-NEXT: [[TMP33:%.*]] = fadd fast <2 x float> [[TMP31]], [[TMP14]] |
| ; UNROLL-NO-IC-NEXT: [[TMP34:%.*]] = fadd fast <2 x float> [[TMP32]], [[TMP18]] |
| ; UNROLL-NO-IC-NEXT: [[TMP35]] = fadd fast <2 x float> [[TMP33]], [[TMP26]] |
| ; UNROLL-NO-IC-NEXT: [[TMP36]] = fadd fast <2 x float> [[TMP34]], [[TMP30]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP37:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP37]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = fadd fast <2 x float> [[TMP36]], [[TMP35]] |
| ; UNROLL-NO-IC-NEXT: [[TMP38:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float 0.000000e+00, <2 x float> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP38]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP45:%.*]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I]] |
| ; UNROLL-NO-IC-NEXT: [[TMP40:%.*]] = load float, ptr [[TMP39]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[I]] |
| ; UNROLL-NO-IC-NEXT: [[TMP42:%.*]] = load float, ptr [[TMP41]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP43:%.*]] = fadd fast float [[S]], 1.000000e+00 |
| ; UNROLL-NO-IC-NEXT: [[TMP44:%.*]] = fadd fast float [[TMP43]], [[TMP40]] |
| ; UNROLL-NO-IC-NEXT: [[TMP45]] = fadd fast float [[TMP44]], [[TMP42]] |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 8 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: [[S_LCSSA:%.*]] = phi float [ [[TMP45]], [[FOR_BODY]] ], [ [[TMP38]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret float [[S_LCSSA]] |
| ; |
| ; INTERLEAVE-LABEL: @scalarize_induction_variable_02( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 65 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = add nsw i64 [[N]], -1 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 3 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = and i64 [[TMP2]], 7 |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = select i1 [[TMP3]], i64 8, i64 [[N_MOD_VF]] |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub nsw i64 [[TMP2]], [[TMP4]] |
| ; INTERLEAVE-NEXT: [[IND_END:%.*]] = shl i64 [[N_VEC]], 3 |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI1:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3 |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = or disjoint i64 [[OFFSET_IDX]], 32 |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[OFFSET_IDX]] |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]] |
| ; INTERLEAVE-NEXT: [[WIDE_VEC:%.*]] = load <32 x float>, ptr [[TMP6]], align 4 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <32 x float> [[WIDE_VEC]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24> |
| ; INTERLEAVE-NEXT: [[WIDE_VEC2:%.*]] = load <32 x float>, ptr [[TMP7]], align 4 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC3:%.*]] = shufflevector <32 x float> [[WIDE_VEC2]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24> |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[OFFSET_IDX]] |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP5]] |
| ; INTERLEAVE-NEXT: [[WIDE_VEC4:%.*]] = load <32 x float>, ptr [[TMP8]], align 4 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC5:%.*]] = shufflevector <32 x float> [[WIDE_VEC4]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24> |
| ; INTERLEAVE-NEXT: [[WIDE_VEC6:%.*]] = load <32 x float>, ptr [[TMP9]], align 4 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC7:%.*]] = shufflevector <32 x float> [[WIDE_VEC6]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24> |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = fadd fast <4 x float> [[VEC_PHI]], splat (float 1.000000e+00) |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = fadd fast <4 x float> [[VEC_PHI1]], splat (float 1.000000e+00) |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = fadd fast <4 x float> [[TMP10]], [[STRIDED_VEC]] |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = fadd fast <4 x float> [[TMP11]], [[STRIDED_VEC3]] |
| ; INTERLEAVE-NEXT: [[TMP14]] = fadd fast <4 x float> [[TMP12]], [[STRIDED_VEC5]] |
| ; INTERLEAVE-NEXT: [[TMP15]] = fadd fast <4 x float> [[TMP13]], [[STRIDED_VEC7]] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[BIN_RDX:%.*]] = fadd fast <4 x float> [[TMP15]], [[TMP14]] |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[BIN_RDX]]) |
| ; INTERLEAVE-NEXT: br label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP17]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP24:%.*]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[I]] |
| ; INTERLEAVE-NEXT: [[TMP19:%.*]] = load float, ptr [[TMP18]], align 4 |
| ; INTERLEAVE-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[I]] |
| ; INTERLEAVE-NEXT: [[TMP21:%.*]] = load float, ptr [[TMP20]], align 4 |
| ; INTERLEAVE-NEXT: [[TMP22:%.*]] = fadd fast float [[S]], 1.000000e+00 |
| ; INTERLEAVE-NEXT: [[TMP23:%.*]] = fadd fast float [[TMP22]], [[TMP19]] |
| ; INTERLEAVE-NEXT: [[TMP24]] = fadd fast float [[TMP23]], [[TMP21]] |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 8 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP14:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: ret float [[TMP24]] |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] |
| %s = phi float [ 0.0, %entry ], [ %6, %for.body ] |
| %0 = getelementptr inbounds float, ptr %a, i64 %i |
| %1 = load float, ptr %0, align 4 |
| %2 = getelementptr inbounds float, ptr %b, i64 %i |
| %3 = load float, ptr %2, align 4 |
| %4 = fadd fast float %s, 1.0 |
| %5 = fadd fast float %4, %1 |
| %6 = fadd fast float %5, %3 |
| %i.next = add nuw nsw i64 %i, 8 |
| %cond = icmp slt i64 %i.next, %n |
| br i1 %cond, label %for.body, label %for.end |
| |
| for.end: |
| %s.lcssa = phi float [ %6, %for.body ] |
| ret float %s.lcssa |
| } |
| |
| ; Make sure we scalarize the step vectors used for the pointer arithmetic. We |
| ; can't easily simplify vectorized step vectors. (Interleaved accesses.) |
| ; |
| ; for (int i = 0; i < n; ++i) |
| ; a[i].f ^= y; |
| ; |
| |
| %pair.i32 = type { i32, i32 } |
| define void @scalarize_induction_variable_03(ptr %p, i32 %y, i64 %n) { |
| ; CHECK-LABEL: @scalarize_induction_variable_03( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 |
| ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[TMP0]], i32 1 |
| ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP1]], i32 1 |
| ; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP2]], align 8 |
| ; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 8 |
| ; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP4]], i32 0 |
| ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[TMP5]], i32 1 |
| ; CHECK-NEXT: [[TMP8:%.*]] = xor <2 x i32> [[TMP7]], [[BROADCAST_SPLAT]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[TMP8]], i32 0 |
| ; CHECK-NEXT: store i32 [[TMP9]], ptr [[TMP2]], align 8 |
| ; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i32> [[TMP8]], i32 1 |
| ; CHECK-NEXT: store i32 [[TMP10]], ptr [[TMP3]], align 8 |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; CHECK-NEXT: [[TMP12:%.*]] = load i32, ptr [[F]], align 8 |
| ; CHECK-NEXT: [[TMP13:%.*]] = xor i32 [[TMP12]], [[Y]] |
| ; CHECK-NEXT: store i32 [[TMP13]], ptr [[F]], align 8 |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @scalarize_induction_variable_03( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 2 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775806 |
| ; IND-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 |
| ; IND-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP0:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; IND-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; IND-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP0]], i32 1 |
| ; IND-NEXT: [[TMP3:%.*]] = load i32, ptr [[TMP1]], align 8 |
| ; IND-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP2]], align 8 |
| ; IND-NEXT: [[TMP5:%.*]] = insertelement <2 x i32> poison, i32 [[TMP3]], i64 0 |
| ; IND-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[TMP4]], i64 1 |
| ; IND-NEXT: [[TMP7:%.*]] = xor <2 x i32> [[TMP6]], [[BROADCAST_SPLAT]] |
| ; IND-NEXT: [[TMP8:%.*]] = extractelement <2 x i32> [[TMP7]], i64 0 |
| ; IND-NEXT: store i32 [[TMP8]], ptr [[TMP1]], align 8 |
| ; IND-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[TMP7]], i64 1 |
| ; IND-NEXT: store i32 [[TMP9]], ptr [[TMP2]], align 8 |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[F:%.*]] = getelementptr inbounds nuw [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; IND-NEXT: [[TMP11:%.*]] = load i32, ptr [[F]], align 8 |
| ; IND-NEXT: [[TMP12:%.*]] = xor i32 [[TMP11]], [[Y]] |
| ; IND-NEXT: store i32 [[TMP12]], ptr [[F]], align 8 |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; IND-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @scalarize_induction_variable_03( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775804 |
| ; UNROLL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 |
| ; UNROLL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP0:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; UNROLL-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP0]], i32 1 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP1]], i32 1 |
| ; UNROLL-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1 |
| ; UNROLL-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP3]], align 8 |
| ; UNROLL-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 8 |
| ; UNROLL-NEXT: [[TMP9:%.*]] = insertelement <2 x i32> poison, i32 [[TMP7]], i64 0 |
| ; UNROLL-NEXT: [[TMP10:%.*]] = insertelement <2 x i32> [[TMP9]], i32 [[TMP8]], i64 1 |
| ; UNROLL-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP5]], align 8 |
| ; UNROLL-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP6]], align 8 |
| ; UNROLL-NEXT: [[TMP13:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0 |
| ; UNROLL-NEXT: [[TMP14:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP12]], i64 1 |
| ; UNROLL-NEXT: [[TMP15:%.*]] = xor <2 x i32> [[TMP10]], [[BROADCAST_SPLAT]] |
| ; UNROLL-NEXT: [[TMP16:%.*]] = xor <2 x i32> [[TMP14]], [[BROADCAST_SPLAT]] |
| ; UNROLL-NEXT: [[TMP17:%.*]] = extractelement <2 x i32> [[TMP15]], i64 0 |
| ; UNROLL-NEXT: store i32 [[TMP17]], ptr [[TMP3]], align 8 |
| ; UNROLL-NEXT: [[TMP18:%.*]] = extractelement <2 x i32> [[TMP15]], i64 1 |
| ; UNROLL-NEXT: store i32 [[TMP18]], ptr [[TMP4]], align 8 |
| ; UNROLL-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP16]], i64 0 |
| ; UNROLL-NEXT: store i32 [[TMP19]], ptr [[TMP5]], align 8 |
| ; UNROLL-NEXT: [[TMP20:%.*]] = extractelement <2 x i32> [[TMP16]], i64 1 |
| ; UNROLL-NEXT: store i32 [[TMP20]], ptr [[TMP6]], align 8 |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[F:%.*]] = getelementptr inbounds nuw [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NEXT: [[TMP22:%.*]] = load i32, ptr [[F]], align 8 |
| ; UNROLL-NEXT: [[TMP23:%.*]] = xor i32 [[TMP22]], [[Y]] |
| ; UNROLL-NEXT: store i32 [[TMP23]], ptr [[F]], align 8 |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_03( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1) |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[TMP0]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP1]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP4]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = insertelement <2 x i32> poison, i32 [[TMP8]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = insertelement <2 x i32> [[TMP10]], i32 [[TMP9]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP6]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP7]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = insertelement <2 x i32> poison, i32 [[TMP12]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = insertelement <2 x i32> [[TMP14]], i32 [[TMP13]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = xor <2 x i32> [[TMP11]], [[BROADCAST_SPLAT]] |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = xor <2 x i32> [[TMP15]], [[BROADCAST_SPLAT]] |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = extractelement <2 x i32> [[TMP16]], i32 0 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP18]], ptr [[TMP4]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP16]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP19]], ptr [[TMP5]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = extractelement <2 x i32> [[TMP17]], i32 0 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP20]], ptr [[TMP6]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = extractelement <2 x i32> [[TMP17]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP21]], ptr [[TMP7]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = load i32, ptr [[F]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = xor i32 [[TMP23]], [[Y]] |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP24]], ptr [[F]], align 8 |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: ret void |
| ; |
| ; INTERLEAVE-LABEL: @scalarize_induction_variable_03( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 9 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = and i64 [[N]], 7 |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = select i1 [[TMP0]], i64 8, i64 [[N_MOD_VF]] |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub nsw i64 [[N]], [[TMP1]] |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[Y:%.*]], i64 0 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = or disjoint i64 [[INDEX]], 4 |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = or disjoint i64 [[INDEX]], 5 |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = or disjoint i64 [[INDEX]], 6 |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = or disjoint i64 [[INDEX]], 7 |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP5]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP6]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP7]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP8]], i32 1 |
| ; INTERLEAVE-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP9]], align 8 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> |
| ; INTERLEAVE-NEXT: [[WIDE_VEC1:%.*]] = load <8 x i32>, ptr [[TMP13]], align 8 |
| ; INTERLEAVE-NEXT: [[STRIDED_VEC2:%.*]] = shufflevector <8 x i32> [[WIDE_VEC1]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6> |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = xor <4 x i32> [[STRIDED_VEC]], [[BROADCAST_SPLAT]] |
| ; INTERLEAVE-NEXT: [[TMP18:%.*]] = xor <4 x i32> [[STRIDED_VEC2]], [[BROADCAST_SPLAT]] |
| ; INTERLEAVE-NEXT: [[TMP19:%.*]] = extractelement <4 x i32> [[TMP17]], i64 0 |
| ; INTERLEAVE-NEXT: store i32 [[TMP19]], ptr [[TMP9]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP20:%.*]] = extractelement <4 x i32> [[TMP17]], i64 1 |
| ; INTERLEAVE-NEXT: store i32 [[TMP20]], ptr [[TMP10]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP21:%.*]] = extractelement <4 x i32> [[TMP17]], i64 2 |
| ; INTERLEAVE-NEXT: store i32 [[TMP21]], ptr [[TMP11]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP22:%.*]] = extractelement <4 x i32> [[TMP17]], i64 3 |
| ; INTERLEAVE-NEXT: store i32 [[TMP22]], ptr [[TMP12]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP23:%.*]] = extractelement <4 x i32> [[TMP18]], i64 0 |
| ; INTERLEAVE-NEXT: store i32 [[TMP23]], ptr [[TMP13]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP24:%.*]] = extractelement <4 x i32> [[TMP18]], i64 1 |
| ; INTERLEAVE-NEXT: store i32 [[TMP24]], ptr [[TMP14]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP25:%.*]] = extractelement <4 x i32> [[TMP18]], i64 2 |
| ; INTERLEAVE-NEXT: store i32 [[TMP25]], ptr [[TMP15]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP26:%.*]] = extractelement <4 x i32> [[TMP18]], i64 3 |
| ; INTERLEAVE-NEXT: store i32 [[TMP26]], ptr [[TMP16]], align 8 |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP27:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP27]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: br label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; INTERLEAVE-NEXT: [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP28:%.*]] = load i32, ptr [[F]], align 8 |
| ; INTERLEAVE-NEXT: [[TMP29:%.*]] = xor i32 [[TMP28]], [[Y]] |
| ; INTERLEAVE-NEXT: store i32 [[TMP29]], ptr [[F]], align 8 |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP16:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: ret void |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] |
| %f = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1 |
| %0 = load i32, ptr %f, align 8 |
| %1 = xor i32 %0, %y |
| store i32 %1, ptr %f, align 8 |
| %i.next = add nuw nsw i64 %i, 1 |
| %cond = icmp slt i64 %i.next, %n |
| br i1 %cond, label %for.body, label %for.end |
| |
| for.end: |
| ret void |
| } |
| |
| ; Make sure we scalarize the step vectors used for the pointer arithmetic. We |
| ; can't easily simplify vectorized step vectors. (Interleaved accesses.) |
| ; |
| ; for (int i = 0; i < n; ++i) |
| ; p[i].f = a[i * 4] |
| ; |
| |
| define void @scalarize_induction_variable_04(ptr %a, ptr %p, i32 %n) { |
| ; CHECK-LABEL: @scalarize_induction_variable_04( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; CHECK: vector.memcheck: |
| ; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P:%.*]], i64 4 |
| ; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1 |
| ; CHECK-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64 |
| ; CHECK-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 3 |
| ; CHECK-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP5]], 8 |
| ; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP6]] |
| ; CHECK-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 4 |
| ; CHECK-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[TMP7]], 4 |
| ; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP8]] |
| ; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]] |
| ; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] |
| ; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[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_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 1 |
| ; CHECK-NEXT: [[TMP11:%.*]] = shl nsw <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i64> [[TMP11]], i32 0 |
| ; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP12]] |
| ; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i32 1 |
| ; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP14]] |
| ; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP13]], align 1, !alias.scope [[META17:![0-9]+]] |
| ; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[TMP15]], align 1, !alias.scope [[META17]] |
| ; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P]], i64 [[TMP9]], i32 1 |
| ; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP10]], i32 1 |
| ; CHECK-NEXT: store i32 [[TMP16]], ptr [[TMP18]], align 1, !alias.scope [[META20:![0-9]+]], !noalias [[META17]] |
| ; CHECK-NEXT: store i32 [[TMP17]], ptr [[TMP19]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[TMP21:%.*]] = shl nsw i64 [[I]], 2 |
| ; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP21]] |
| ; CHECK-NEXT: [[TMP23:%.*]] = load i32, ptr [[TMP22]], align 1 |
| ; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; CHECK-NEXT: store i32 [[TMP23]], ptr [[TMP24]], align 1 |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; CHECK-NEXT: [[TMP25:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[TMP25]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @scalarize_induction_variable_04( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; IND-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; IND-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; IND: vector.memcheck: |
| ; IND-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P:%.*]], i64 4 |
| ; IND-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1 |
| ; IND-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64 |
| ; IND-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 3 |
| ; IND-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP5]] |
| ; IND-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP6]], i64 8 |
| ; IND-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 4 |
| ; IND-NEXT: [[TMP8:%.*]] = or disjoint i64 [[TMP7]], 4 |
| ; IND-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP8]] |
| ; IND-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]] |
| ; IND-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] |
| ; IND-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; IND-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590 |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; IND-NEXT: [[TMP10:%.*]] = shl nsw <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; IND-NEXT: [[TMP11:%.*]] = extractelement <2 x i64> [[TMP10]], i64 0 |
| ; IND-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP11]] |
| ; IND-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP10]], i64 1 |
| ; IND-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP13]] |
| ; IND-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP12]], align 1, !alias.scope [[META17:![0-9]+]] |
| ; IND-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP14]], align 1, !alias.scope [[META17]] |
| ; IND-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P]], i64 [[INDEX]], i32 1 |
| ; IND-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP9]], i32 1 |
| ; IND-NEXT: store i32 [[TMP15]], ptr [[TMP17]], align 1, !alias.scope [[META20:![0-9]+]], !noalias [[META17]] |
| ; IND-NEXT: store i32 [[TMP16]], ptr [[TMP18]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; IND-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[I]], 4 |
| ; IND-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[DOTIDX]] |
| ; IND-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 1 |
| ; IND-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; IND-NEXT: store i32 [[TMP21]], ptr [[TMP22]], align 1 |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; IND-NEXT: [[TMP23:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; IND-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP23]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @scalarize_induction_variable_04( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; UNROLL: vector.memcheck: |
| ; UNROLL-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P:%.*]], i64 4 |
| ; UNROLL-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 3 |
| ; UNROLL-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP5]] |
| ; UNROLL-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP6]], i64 8 |
| ; UNROLL-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 4 |
| ; UNROLL-NEXT: [[TMP8:%.*]] = or disjoint i64 [[TMP7]], 4 |
| ; UNROLL-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP8]] |
| ; UNROLL-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]] |
| ; UNROLL-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] |
| ; UNROLL-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; UNROLL-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588 |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; UNROLL-NEXT: [[TMP10:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; UNROLL-NEXT: [[TMP11:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; UNROLL-NEXT: [[TMP12:%.*]] = shl nsw <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; UNROLL-NEXT: [[STEP_ADD:%.*]] = shl <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; UNROLL-NEXT: [[TMP13:%.*]] = add <2 x i64> [[STEP_ADD]], splat (i64 8) |
| ; UNROLL-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP12]], i64 0 |
| ; UNROLL-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP14]] |
| ; UNROLL-NEXT: [[TMP16:%.*]] = extractelement <2 x i64> [[TMP12]], i64 1 |
| ; UNROLL-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP16]] |
| ; UNROLL-NEXT: [[TMP18:%.*]] = extractelement <2 x i64> [[TMP13]], i64 0 |
| ; UNROLL-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP18]] |
| ; UNROLL-NEXT: [[TMP20:%.*]] = extractelement <2 x i64> [[TMP13]], i64 1 |
| ; UNROLL-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP20]] |
| ; UNROLL-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP15]], align 1, !alias.scope [[META17:![0-9]+]] |
| ; UNROLL-NEXT: [[TMP23:%.*]] = load i32, ptr [[TMP17]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP19]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NEXT: [[TMP25:%.*]] = load i32, ptr [[TMP21]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P]], i64 [[INDEX]], i32 1 |
| ; UNROLL-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP9]], i32 1 |
| ; UNROLL-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP10]], i32 1 |
| ; UNROLL-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP11]], i32 1 |
| ; UNROLL-NEXT: store i32 [[TMP22]], ptr [[TMP26]], align 1, !alias.scope [[META20:![0-9]+]], !noalias [[META17]] |
| ; UNROLL-NEXT: store i32 [[TMP23]], ptr [[TMP27]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NEXT: store i32 [[TMP24]], ptr [[TMP28]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NEXT: store i32 [[TMP25]], ptr [[TMP29]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 4) |
| ; UNROLL-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[I]], 4 |
| ; UNROLL-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[DOTIDX]] |
| ; UNROLL-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP31]], align 1 |
| ; UNROLL-NEXT: [[TMP33:%.*]] = getelementptr inbounds nuw [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NEXT: store i32 [[TMP32]], ptr [[TMP33]], align 1 |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NEXT: [[TMP34:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP34]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_04( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; UNROLL-NO-IC: vector.memcheck: |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P:%.*]], i64 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64 |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP5]], 8 |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP6]] |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[TMP7]], 4 |
| ; UNROLL-NO-IC-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP8]] |
| ; UNROLL-NO-IC-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]] |
| ; UNROLL-NO-IC-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] |
| ; UNROLL-NO-IC-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = shl nsw <2 x i64> [[VEC_IND]], splat (i64 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = shl nsw <2 x i64> [[STEP_ADD]], splat (i64 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = extractelement <2 x i64> [[TMP13]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP15]] |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = extractelement <2 x i64> [[TMP13]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP17]] |
| ; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = extractelement <2 x i64> [[TMP14]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP19]] |
| ; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = extractelement <2 x i64> [[TMP14]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP21]] |
| ; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = load i32, ptr [[TMP16]], align 1, !alias.scope [[META17:![0-9]+]] |
| ; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP18]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NO-IC-NEXT: [[TMP25:%.*]] = load i32, ptr [[TMP20]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NO-IC-NEXT: [[TMP26:%.*]] = load i32, ptr [[TMP22]], align 1, !alias.scope [[META17]] |
| ; UNROLL-NO-IC-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P]], i64 [[TMP9]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP10]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP11]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP12]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP23]], ptr [[TMP27]], align 1, !alias.scope [[META20:![0-9]+]], !noalias [[META17]] |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP24]], ptr [[TMP28]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP25]], ptr [[TMP29]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP26]], ptr [[TMP30]], align 1, !alias.scope [[META20]], !noalias [[META17]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD]], splat (i64 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP31:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP31]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP32:%.*]] = shl nsw i64 [[I]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP33:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP32]] |
| ; UNROLL-NO-IC-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i32 [[TMP34]], ptr [[TMP35]], align 1 |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP36:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp eq i32 [[TMP36]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: ret void |
| ; |
| ; INTERLEAVE-LABEL: @scalarize_induction_variable_04( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] |
| ; INTERLEAVE: vector.memcheck: |
| ; INTERLEAVE-NEXT: [[SCEVGEP:%.*]] = getelementptr nuw i8, ptr [[P:%.*]], i64 4 |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64 |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[TMP4]], 3 |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP5]] |
| ; INTERLEAVE-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP6]], i64 8 |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP4]], 4 |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = or disjoint i64 [[TMP7]], 4 |
| ; INTERLEAVE-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[TMP8]] |
| ; INTERLEAVE-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]] |
| ; INTERLEAVE-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] |
| ; INTERLEAVE-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] |
| ; INTERLEAVE-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = and i64 [[TMP2]], 7 |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = select i1 [[TMP9]], i64 8, i64 [[N_MOD_VF]] |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub nsw i64 [[TMP2]], [[TMP10]] |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; INTERLEAVE-NEXT: [[TMP14:%.*]] = or disjoint i64 [[INDEX]], 4 |
| ; INTERLEAVE-NEXT: [[TMP15:%.*]] = or disjoint i64 [[INDEX]], 5 |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = or disjoint i64 [[INDEX]], 6 |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = or disjoint i64 [[INDEX]], 7 |
| ; INTERLEAVE-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 4 |
| ; INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[DOTIDX]] |
| ; INTERLEAVE-NEXT: [[DOTIDX5:%.*]] = shl nsw i64 [[TMP14]], 4 |
| ; INTERLEAVE-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[DOTIDX5]] |
| ; INTERLEAVE-NEXT: [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP18]], align 1 |
| ; INTERLEAVE-NEXT: [[WIDE_VEC3:%.*]] = load <16 x i32>, ptr [[TMP19]], align 1 |
| ; INTERLEAVE-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P]], i64 [[INDEX]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP11]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP12]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP13]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP14]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP15]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP16]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP17]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP28:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 0 |
| ; INTERLEAVE-NEXT: store i32 [[TMP28]], ptr [[TMP20]], align 1, !alias.scope [[META17:![0-9]+]], !noalias [[META20:![0-9]+]] |
| ; INTERLEAVE-NEXT: [[TMP29:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 4 |
| ; INTERLEAVE-NEXT: store i32 [[TMP29]], ptr [[TMP21]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP30:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 8 |
| ; INTERLEAVE-NEXT: store i32 [[TMP30]], ptr [[TMP22]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP31:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 12 |
| ; INTERLEAVE-NEXT: store i32 [[TMP31]], ptr [[TMP23]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP32:%.*]] = extractelement <16 x i32> [[WIDE_VEC3]], i64 0 |
| ; INTERLEAVE-NEXT: store i32 [[TMP32]], ptr [[TMP24]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP33:%.*]] = extractelement <16 x i32> [[WIDE_VEC3]], i64 4 |
| ; INTERLEAVE-NEXT: store i32 [[TMP33]], ptr [[TMP25]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP34:%.*]] = extractelement <16 x i32> [[WIDE_VEC3]], i64 8 |
| ; INTERLEAVE-NEXT: store i32 [[TMP34]], ptr [[TMP26]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[TMP35:%.*]] = extractelement <16 x i32> [[WIDE_VEC3]], i64 12 |
| ; INTERLEAVE-NEXT: store i32 [[TMP35]], ptr [[TMP27]], align 1, !alias.scope [[META17]], !noalias [[META20]] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: br label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; INTERLEAVE-NEXT: [[DOTIDX6:%.*]] = shl nsw i64 [[I]], 4 |
| ; INTERLEAVE-NEXT: [[TMP37:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[DOTIDX6]] |
| ; INTERLEAVE-NEXT: [[TMP38:%.*]] = load i32, ptr [[TMP37]], align 1 |
| ; INTERLEAVE-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1 |
| ; INTERLEAVE-NEXT: store i32 [[TMP38]], ptr [[TMP39]], align 1 |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; INTERLEAVE-NEXT: [[TMP40:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP40]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_END:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: ret void |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i64 [ %i.next, %for.body ], [ 0, %entry] |
| %0 = shl nsw i64 %i, 2 |
| %1 = getelementptr inbounds i32, ptr %a, i64 %0 |
| %2 = load i32, ptr %1, align 1 |
| %3 = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1 |
| store i32 %2, ptr %3, align 1 |
| %i.next = add nuw nsw i64 %i, 1 |
| %4 = trunc i64 %i.next to i32 |
| %cond = icmp eq i32 %4, %n |
| br i1 %cond, label %for.end, label %for.body |
| |
| for.end: |
| ret void |
| } |
| |
| ; PR30542. Ensure we generate all the scalar steps for the induction variable. |
| ; The scalar induction variable is used by a getelementptr instruction |
| ; (uniform), and a udiv (non-uniform). |
| ; |
| ; int sum = 0; |
| ; for (int i = 0; i < n; ++i) { |
| ; int x = a[i]; |
| ; if (c) |
| ; x /= i; |
| ; sum += x; |
| ; } |
| ; |
| ; |
| ; |
| ; |
| |
| define i32 @scalarize_induction_variable_05(ptr %a, i32 %x, i1 %c, i32 %n) { |
| ; CHECK-LABEL: @scalarize_induction_variable_05( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1) |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[SMAX]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[SMAX]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: [[TMP14:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], splat (i1 true) |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[PRED_UDIV_CONTINUE2]] ] |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[TMP0]] |
| ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0 |
| ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP2]], align 4 |
| ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0 |
| ; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]] |
| ; CHECK: pred.udiv.if: |
| ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 0 |
| ; CHECK-NEXT: [[TMP5:%.*]] = udiv i32 [[TMP4]], [[TMP0]] |
| ; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i32 0 |
| ; CHECK-NEXT: br label [[PRED_UDIV_CONTINUE]] |
| ; CHECK: pred.udiv.continue: |
| ; CHECK-NEXT: [[TMP7:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_UDIV_IF]] ] |
| ; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1 |
| ; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]] |
| ; CHECK: pred.udiv.if1: |
| ; CHECK-NEXT: [[TMP9:%.*]] = add i32 [[INDEX]], 1 |
| ; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 1 |
| ; CHECK-NEXT: [[TMP11:%.*]] = udiv i32 [[TMP10]], [[TMP9]] |
| ; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP11]], i32 1 |
| ; CHECK-NEXT: br label [[PRED_UDIV_CONTINUE2]] |
| ; CHECK: pred.udiv.continue2: |
| ; CHECK-NEXT: [[TMP13:%.*]] = phi <2 x i32> [ [[TMP7]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP12]], [[PRED_UDIV_IF1]] ] |
| ; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP14]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP13]] |
| ; CHECK-NEXT: [[TMP15]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP17:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP15]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP17]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ] |
| ; CHECK-NEXT: [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ] |
| ; CHECK-NEXT: [[VAR0:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[I]] |
| ; CHECK-NEXT: [[VAR1:%.*]] = load i32, ptr [[VAR0]], align 4 |
| ; CHECK-NEXT: br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]] |
| ; CHECK: if.then: |
| ; CHECK-NEXT: [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]] |
| ; CHECK-NEXT: br label [[IF_END]] |
| ; CHECK: if.end: |
| ; CHECK-NEXT: [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ] |
| ; CHECK-NEXT: [[VAR4]] = add i32 [[VAR3]], [[SUM]] |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[VAR5]] |
| ; |
| ; IND-LABEL: @scalarize_induction_variable_05( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1) |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i32 [[N]], 2 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483646 |
| ; IND-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0 |
| ; IND-NEXT: [[TMP11:%.*]] = xor <2 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison> |
| ; IND-NEXT: [[TMP12:%.*]] = shufflevector <2 x i1> [[TMP11]], <2 x i1> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ] |
| ; IND-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[PRED_UDIV_CONTINUE2]] ] |
| ; IND-NEXT: [[TMP0:%.*]] = sext i32 [[INDEX]] to i64 |
| ; IND-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP0]] |
| ; IND-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4 |
| ; IND-NEXT: br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]] |
| ; IND: pred.udiv.if: |
| ; IND-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 0 |
| ; IND-NEXT: [[TMP3:%.*]] = udiv i32 [[TMP2]], [[INDEX]] |
| ; IND-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> poison, i32 [[TMP3]], i64 0 |
| ; IND-NEXT: br label [[PRED_UDIV_CONTINUE]] |
| ; IND: pred.udiv.continue: |
| ; IND-NEXT: [[TMP5:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP4]], [[PRED_UDIV_IF]] ] |
| ; IND-NEXT: br i1 [[C]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]] |
| ; IND: pred.udiv.if1: |
| ; IND-NEXT: [[TMP6:%.*]] = or disjoint i32 [[INDEX]], 1 |
| ; IND-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 1 |
| ; IND-NEXT: [[TMP8:%.*]] = udiv i32 [[TMP7]], [[TMP6]] |
| ; IND-NEXT: [[TMP9:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[TMP8]], i64 1 |
| ; IND-NEXT: br label [[PRED_UDIV_CONTINUE2]] |
| ; IND: pred.udiv.continue2: |
| ; IND-NEXT: [[TMP10:%.*]] = phi <2 x i32> [ [[TMP5]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP9]], [[PRED_UDIV_IF1]] ] |
| ; IND-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP12]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP10]] |
| ; IND-NEXT: [[TMP13]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP14:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP13]]) |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP15]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ] |
| ; IND-NEXT: [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ] |
| ; IND-NEXT: [[TMP16:%.*]] = zext nneg i32 [[I]] to i64 |
| ; IND-NEXT: [[VAR0:%.*]] = getelementptr inbounds nuw i32, ptr [[A]], i64 [[TMP16]] |
| ; IND-NEXT: [[VAR1:%.*]] = load i32, ptr [[VAR0]], align 4 |
| ; IND-NEXT: br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]] |
| ; IND: if.then: |
| ; IND-NEXT: [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]] |
| ; IND-NEXT: br label [[IF_END]] |
| ; IND: if.end: |
| ; IND-NEXT: [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ] |
| ; IND-NEXT: [[VAR4]] = add i32 [[VAR3]], [[SUM]] |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1 |
| ; IND-NEXT: [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP15]], [[MIDDLE_BLOCK]] ] |
| ; IND-NEXT: ret i32 [[VAR5]] |
| ; |
| ; UNROLL-LABEL: @scalarize_induction_variable_05( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1) |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i32 [[N]], 4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483644 |
| ; UNROLL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0 |
| ; UNROLL-NEXT: [[TMP27:%.*]] = xor <2 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison> |
| ; UNROLL-NEXT: [[TMP28:%.*]] = shufflevector <2 x i1> [[TMP27]], <2 x i1> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE8:%.*]] ] |
| ; UNROLL-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[PRED_UDIV_CONTINUE8]] ] |
| ; UNROLL-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[PRED_UDIV_CONTINUE8]] ] |
| ; UNROLL-NEXT: [[TMP0:%.*]] = sext i32 [[INDEX]] to i64 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP0]] |
| ; UNROLL-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 8 |
| ; UNROLL-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4 |
| ; UNROLL-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, ptr [[TMP2]], align 4 |
| ; UNROLL-NEXT: br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]] |
| ; UNROLL: pred.udiv.if: |
| ; UNROLL-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 0 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = udiv i32 [[TMP3]], [[INDEX]] |
| ; UNROLL-NEXT: [[TMP5:%.*]] = insertelement <2 x i32> poison, i32 [[TMP4]], i64 0 |
| ; UNROLL-NEXT: br label [[PRED_UDIV_CONTINUE]] |
| ; UNROLL: pred.udiv.continue: |
| ; UNROLL-NEXT: [[TMP6:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_UDIV_IF]] ] |
| ; UNROLL-NEXT: br i1 [[C]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]] |
| ; UNROLL: pred.udiv.if3: |
| ; UNROLL-NEXT: [[TMP7:%.*]] = or disjoint i32 [[INDEX]], 1 |
| ; UNROLL-NEXT: [[TMP8:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 1 |
| ; UNROLL-NEXT: [[TMP9:%.*]] = udiv i32 [[TMP8]], [[TMP7]] |
| ; UNROLL-NEXT: [[TMP10:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[TMP9]], i64 1 |
| ; UNROLL-NEXT: br label [[PRED_UDIV_CONTINUE4]] |
| ; UNROLL: pred.udiv.continue4: |
| ; UNROLL-NEXT: [[TMP11:%.*]] = phi <2 x i32> [ [[TMP6]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP10]], [[PRED_UDIV_IF3]] ] |
| ; UNROLL-NEXT: br i1 [[C]], label [[PRED_UDIV_IF5:%.*]], label [[PRED_UDIV_CONTINUE6:%.*]] |
| ; UNROLL: pred.udiv.if5: |
| ; UNROLL-NEXT: [[TMP12:%.*]] = or disjoint i32 [[INDEX]], 2 |
| ; UNROLL-NEXT: [[TMP13:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i64 0 |
| ; UNROLL-NEXT: [[TMP14:%.*]] = udiv i32 [[TMP13]], [[TMP12]] |
| ; UNROLL-NEXT: [[TMP15:%.*]] = insertelement <2 x i32> poison, i32 [[TMP14]], i64 0 |
| ; UNROLL-NEXT: br label [[PRED_UDIV_CONTINUE6]] |
| ; UNROLL: pred.udiv.continue6: |
| ; UNROLL-NEXT: [[TMP16:%.*]] = phi <2 x i32> [ poison, [[PRED_UDIV_CONTINUE4]] ], [ [[TMP15]], [[PRED_UDIV_IF5]] ] |
| ; UNROLL-NEXT: br i1 [[C]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8]] |
| ; UNROLL: pred.udiv.if7: |
| ; UNROLL-NEXT: [[TMP17:%.*]] = or disjoint i32 [[INDEX]], 3 |
| ; UNROLL-NEXT: [[TMP18:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i64 1 |
| ; UNROLL-NEXT: [[TMP19:%.*]] = udiv i32 [[TMP18]], [[TMP17]] |
| ; UNROLL-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> [[TMP16]], i32 [[TMP19]], i64 1 |
| ; UNROLL-NEXT: br label [[PRED_UDIV_CONTINUE8]] |
| ; UNROLL: pred.udiv.continue8: |
| ; UNROLL-NEXT: [[TMP21:%.*]] = phi <2 x i32> [ [[TMP16]], [[PRED_UDIV_CONTINUE6]] ], [ [[TMP20]], [[PRED_UDIV_IF7]] ] |
| ; UNROLL-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP28]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP11]] |
| ; UNROLL-NEXT: [[PREDPHI9:%.*]] = select <2 x i1> [[TMP28]], <2 x i32> [[WIDE_LOAD2]], <2 x i32> [[TMP21]] |
| ; UNROLL-NEXT: [[TMP22]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]] |
| ; UNROLL-NEXT: [[TMP23]] = add <2 x i32> [[PREDPHI9]], [[VEC_PHI1]] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[BIN_RDX:%.*]] = add <2 x i32> [[TMP23]], [[TMP22]] |
| ; UNROLL-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]]) |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP25]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ] |
| ; UNROLL-NEXT: [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ] |
| ; UNROLL-NEXT: [[TMP26:%.*]] = zext nneg i32 [[I]] to i64 |
| ; UNROLL-NEXT: [[VAR0:%.*]] = getelementptr inbounds nuw i32, ptr [[A]], i64 [[TMP26]] |
| ; UNROLL-NEXT: [[VAR1:%.*]] = load i32, ptr [[VAR0]], align 4 |
| ; UNROLL-NEXT: br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]] |
| ; UNROLL: if.then: |
| ; UNROLL-NEXT: [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]] |
| ; UNROLL-NEXT: br label [[IF_END]] |
| ; UNROLL: if.end: |
| ; UNROLL-NEXT: [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ] |
| ; UNROLL-NEXT: [[VAR4]] = add i32 [[VAR3]], [[SUM]] |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP25]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NEXT: ret i32 [[VAR5]] |
| ; |
| ; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_05( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1) |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[SMAX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i32 [[SMAX]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NO-IC-NEXT: [[TMP27:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], splat (i1 true) |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE8:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP29:%.*]], [[PRED_UDIV_CONTINUE8]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP30:%.*]], [[PRED_UDIV_CONTINUE8]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 2 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP2]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, ptr [[TMP3]], align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP4]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]] |
| ; UNROLL-NO-IC: pred.udiv.if: |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = udiv i32 [[TMP5]], [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = insertelement <2 x i32> poison, i32 [[TMP6]], i32 0 |
| ; UNROLL-NO-IC-NEXT: br label [[PRED_UDIV_CONTINUE]] |
| ; UNROLL-NO-IC: pred.udiv.continue: |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_UDIV_IF]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP9]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]] |
| ; UNROLL-NO-IC: pred.udiv.if3: |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = add i32 [[INDEX]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = udiv i32 [[TMP11]], [[TMP10]] |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = insertelement <2 x i32> [[TMP8]], i32 [[TMP12]], i32 1 |
| ; UNROLL-NO-IC-NEXT: br label [[PRED_UDIV_CONTINUE4]] |
| ; UNROLL-NO-IC: pred.udiv.continue4: |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = phi <2 x i32> [ [[TMP8]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP13]], [[PRED_UDIV_IF3]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP15]], label [[PRED_UDIV_IF5:%.*]], label [[PRED_UDIV_CONTINUE6:%.*]] |
| ; UNROLL-NO-IC: pred.udiv.if5: |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = add i32 [[INDEX]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = udiv i32 [[TMP17]], [[TMP16]] |
| ; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0 |
| ; UNROLL-NO-IC-NEXT: br label [[PRED_UDIV_CONTINUE6]] |
| ; UNROLL-NO-IC: pred.udiv.continue6: |
| ; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = phi <2 x i32> [ poison, [[PRED_UDIV_CONTINUE4]] ], [ [[TMP19]], [[PRED_UDIV_IF5]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP21]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8]] |
| ; UNROLL-NO-IC: pred.udiv.if7: |
| ; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = add i32 [[INDEX]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = udiv i32 [[TMP23]], [[TMP22]] |
| ; UNROLL-NO-IC-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP20]], i32 [[TMP24]], i32 1 |
| ; UNROLL-NO-IC-NEXT: br label [[PRED_UDIV_CONTINUE8]] |
| ; UNROLL-NO-IC: pred.udiv.continue8: |
| ; UNROLL-NO-IC-NEXT: [[TMP26:%.*]] = phi <2 x i32> [ [[TMP20]], [[PRED_UDIV_CONTINUE6]] ], [ [[TMP25]], [[PRED_UDIV_IF7]] ] |
| ; UNROLL-NO-IC-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP27]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP14]] |
| ; UNROLL-NO-IC-NEXT: [[PREDPHI9:%.*]] = select <2 x i1> [[TMP27]], <2 x i32> [[WIDE_LOAD2]], <2 x i32> [[TMP26]] |
| ; UNROLL-NO-IC-NEXT: [[TMP29]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]] |
| ; UNROLL-NO-IC-NEXT: [[TMP30]] = add <2 x i32> [[PREDPHI9]], [[VEC_PHI1]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP31:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP31]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = add <2 x i32> [[TMP30]], [[TMP29]] |
| ; UNROLL-NO-IC-NEXT: [[TMP32:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP32]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ] |
| ; UNROLL-NO-IC-NEXT: [[VAR0:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[I]] |
| ; UNROLL-NO-IC-NEXT: [[VAR1:%.*]] = load i32, ptr [[VAR0]], align 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]] |
| ; UNROLL-NO-IC: if.then: |
| ; UNROLL-NO-IC-NEXT: [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]] |
| ; UNROLL-NO-IC-NEXT: br label [[IF_END]] |
| ; UNROLL-NO-IC: if.end: |
| ; UNROLL-NO-IC-NEXT: [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VAR4]] = add i32 [[VAR3]], [[SUM]] |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP32]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret i32 [[VAR5]] |
| ; |
| ; INTERLEAVE-LABEL: @scalarize_induction_variable_05( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1) |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i32 [[N]], 8 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483640 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C:%.*]], i64 0 |
| ; INTERLEAVE-NEXT: [[TMP47:%.*]] = xor <4 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison, i1 poison, i1 poison> |
| ; INTERLEAVE-NEXT: [[TMP48:%.*]] = shufflevector <4 x i1> [[TMP47]], <4 x i1> poison, <4 x i32> zeroinitializer |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE16:%.*]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP42:%.*]], [[PRED_UDIV_CONTINUE16]] ] |
| ; INTERLEAVE-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP43:%.*]], [[PRED_UDIV_CONTINUE16]] ] |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = sext i32 [[INDEX]] to i64 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP0]] |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 16 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4 |
| ; INTERLEAVE-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i32>, ptr [[TMP2]], align 4 |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]] |
| ; INTERLEAVE: pred.udiv.if: |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 0 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = udiv i32 [[TMP3]], [[INDEX]] |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> poison, i32 [[TMP4]], i64 0 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE]] |
| ; INTERLEAVE: pred.udiv.continue: |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = phi <4 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_UDIV_IF]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]] |
| ; INTERLEAVE: pred.udiv.if3: |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = or disjoint i32 [[INDEX]], 1 |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 1 |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = udiv i32 [[TMP8]], [[TMP7]] |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = insertelement <4 x i32> [[TMP6]], i32 [[TMP9]], i64 1 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE4]] |
| ; INTERLEAVE: pred.udiv.continue4: |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = phi <4 x i32> [ [[TMP6]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP10]], [[PRED_UDIV_IF3]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF5:%.*]], label [[PRED_UDIV_CONTINUE6:%.*]] |
| ; INTERLEAVE: pred.udiv.if5: |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = or disjoint i32 [[INDEX]], 2 |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 2 |
| ; INTERLEAVE-NEXT: [[TMP14:%.*]] = udiv i32 [[TMP13]], [[TMP12]] |
| ; INTERLEAVE-NEXT: [[TMP15:%.*]] = insertelement <4 x i32> [[TMP11]], i32 [[TMP14]], i64 2 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE6]] |
| ; INTERLEAVE: pred.udiv.continue6: |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = phi <4 x i32> [ [[TMP11]], [[PRED_UDIV_CONTINUE4]] ], [ [[TMP15]], [[PRED_UDIV_IF5]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8:%.*]] |
| ; INTERLEAVE: pred.udiv.if7: |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = or disjoint i32 [[INDEX]], 3 |
| ; INTERLEAVE-NEXT: [[TMP18:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 3 |
| ; INTERLEAVE-NEXT: [[TMP19:%.*]] = udiv i32 [[TMP18]], [[TMP17]] |
| ; INTERLEAVE-NEXT: [[TMP20:%.*]] = insertelement <4 x i32> [[TMP16]], i32 [[TMP19]], i64 3 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE8]] |
| ; INTERLEAVE: pred.udiv.continue8: |
| ; INTERLEAVE-NEXT: [[TMP21:%.*]] = phi <4 x i32> [ [[TMP16]], [[PRED_UDIV_CONTINUE6]] ], [ [[TMP20]], [[PRED_UDIV_IF7]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF9:%.*]], label [[PRED_UDIV_CONTINUE10:%.*]] |
| ; INTERLEAVE: pred.udiv.if9: |
| ; INTERLEAVE-NEXT: [[TMP22:%.*]] = or disjoint i32 [[INDEX]], 4 |
| ; INTERLEAVE-NEXT: [[TMP23:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 0 |
| ; INTERLEAVE-NEXT: [[TMP24:%.*]] = udiv i32 [[TMP23]], [[TMP22]] |
| ; INTERLEAVE-NEXT: [[TMP25:%.*]] = insertelement <4 x i32> poison, i32 [[TMP24]], i64 0 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE10]] |
| ; INTERLEAVE: pred.udiv.continue10: |
| ; INTERLEAVE-NEXT: [[TMP26:%.*]] = phi <4 x i32> [ poison, [[PRED_UDIV_CONTINUE8]] ], [ [[TMP25]], [[PRED_UDIV_IF9]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF11:%.*]], label [[PRED_UDIV_CONTINUE12:%.*]] |
| ; INTERLEAVE: pred.udiv.if11: |
| ; INTERLEAVE-NEXT: [[TMP27:%.*]] = or disjoint i32 [[INDEX]], 5 |
| ; INTERLEAVE-NEXT: [[TMP28:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 1 |
| ; INTERLEAVE-NEXT: [[TMP29:%.*]] = udiv i32 [[TMP28]], [[TMP27]] |
| ; INTERLEAVE-NEXT: [[TMP30:%.*]] = insertelement <4 x i32> [[TMP26]], i32 [[TMP29]], i64 1 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE12]] |
| ; INTERLEAVE: pred.udiv.continue12: |
| ; INTERLEAVE-NEXT: [[TMP31:%.*]] = phi <4 x i32> [ [[TMP26]], [[PRED_UDIV_CONTINUE10]] ], [ [[TMP30]], [[PRED_UDIV_IF11]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF13:%.*]], label [[PRED_UDIV_CONTINUE14:%.*]] |
| ; INTERLEAVE: pred.udiv.if13: |
| ; INTERLEAVE-NEXT: [[TMP32:%.*]] = or disjoint i32 [[INDEX]], 6 |
| ; INTERLEAVE-NEXT: [[TMP33:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 2 |
| ; INTERLEAVE-NEXT: [[TMP34:%.*]] = udiv i32 [[TMP33]], [[TMP32]] |
| ; INTERLEAVE-NEXT: [[TMP35:%.*]] = insertelement <4 x i32> [[TMP31]], i32 [[TMP34]], i64 2 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE14]] |
| ; INTERLEAVE: pred.udiv.continue14: |
| ; INTERLEAVE-NEXT: [[TMP36:%.*]] = phi <4 x i32> [ [[TMP31]], [[PRED_UDIV_CONTINUE12]] ], [ [[TMP35]], [[PRED_UDIV_IF13]] ] |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[PRED_UDIV_IF15:%.*]], label [[PRED_UDIV_CONTINUE16]] |
| ; INTERLEAVE: pred.udiv.if15: |
| ; INTERLEAVE-NEXT: [[TMP37:%.*]] = or disjoint i32 [[INDEX]], 7 |
| ; INTERLEAVE-NEXT: [[TMP38:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 3 |
| ; INTERLEAVE-NEXT: [[TMP39:%.*]] = udiv i32 [[TMP38]], [[TMP37]] |
| ; INTERLEAVE-NEXT: [[TMP40:%.*]] = insertelement <4 x i32> [[TMP36]], i32 [[TMP39]], i64 3 |
| ; INTERLEAVE-NEXT: br label [[PRED_UDIV_CONTINUE16]] |
| ; INTERLEAVE: pred.udiv.continue16: |
| ; INTERLEAVE-NEXT: [[TMP41:%.*]] = phi <4 x i32> [ [[TMP36]], [[PRED_UDIV_CONTINUE14]] ], [ [[TMP40]], [[PRED_UDIV_IF15]] ] |
| ; INTERLEAVE-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP48]], <4 x i32> [[WIDE_LOAD]], <4 x i32> [[TMP21]] |
| ; INTERLEAVE-NEXT: [[PREDPHI17:%.*]] = select <4 x i1> [[TMP48]], <4 x i32> [[WIDE_LOAD2]], <4 x i32> [[TMP41]] |
| ; INTERLEAVE-NEXT: [[TMP42]] = add <4 x i32> [[PREDPHI]], [[VEC_PHI]] |
| ; INTERLEAVE-NEXT: [[TMP43]] = add <4 x i32> [[PREDPHI17]], [[VEC_PHI1]] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP44:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP44]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[TMP43]], [[TMP42]] |
| ; INTERLEAVE-NEXT: [[TMP45:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]]) |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP45]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ] |
| ; INTERLEAVE-NEXT: [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ] |
| ; INTERLEAVE-NEXT: [[TMP46:%.*]] = zext nneg i32 [[I]] to i64 |
| ; INTERLEAVE-NEXT: [[VAR0:%.*]] = getelementptr inbounds nuw i32, ptr [[A]], i64 [[TMP46]] |
| ; INTERLEAVE-NEXT: [[VAR1:%.*]] = load i32, ptr [[VAR0]], align 4 |
| ; INTERLEAVE-NEXT: br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]] |
| ; INTERLEAVE: if.then: |
| ; INTERLEAVE-NEXT: [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]] |
| ; INTERLEAVE-NEXT: br label [[IF_END]] |
| ; INTERLEAVE: if.end: |
| ; INTERLEAVE-NEXT: [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VAR4]] = add i32 [[VAR3]], [[SUM]] |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP45]], [[MIDDLE_BLOCK]] ] |
| ; INTERLEAVE-NEXT: ret i32 [[VAR5]] |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i32 [ 0, %entry ], [ %i.next, %if.end ] |
| %sum = phi i32 [ 0, %entry ], [ %var4, %if.end ] |
| %var0 = getelementptr inbounds i32, ptr %a, i32 %i |
| %var1 = load i32, ptr %var0, align 4 |
| br i1 %c, label %if.then, label %if.end |
| |
| if.then: |
| %var2 = udiv i32 %var1, %i |
| br label %if.end |
| |
| if.end: |
| %var3 = phi i32 [ %var2, %if.then ], [ %var1, %for.body ] |
| %var4 = add i32 %var3, %sum |
| %i.next = add nuw nsw i32 %i, 1 |
| %cond = icmp slt i32 %i.next, %n |
| br i1 %cond, label %for.body, label %for.end |
| |
| for.end: |
| %var5 = phi i32 [ %var4, %if.end ] |
| ret i32 %var5 |
| } |
| |
| ; Ensure we generate both a vector and a scalar induction variable. In this |
| ; test, the induction variable is used by an instruction that will be |
| ; vectorized (trunc) as well as an instruction that will remain in scalar form |
| ; (gepelementptr). |
| ; |
| ; |
| ; |
| |
| %pair.i16 = type { i16, i16 } |
| define void @iv_vector_and_scalar_users(ptr %p, i32 %a, i32 %n) { |
| ; CHECK-LABEL: @iv_vector_and_scalar_users( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 |
| ; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 1 |
| ; CHECK-NEXT: [[TMP5:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]] |
| ; CHECK-NEXT: [[TMP6:%.*]] = trunc <2 x i32> [[TMP5]] to <2 x i16> |
| ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], ptr [[P:%.*]], i64 [[TMP3]], i32 1 |
| ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP4]], i32 1 |
| ; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i16> [[TMP6]], i32 0 |
| ; CHECK-NEXT: store i16 [[TMP9]], ptr [[TMP7]], align 2 |
| ; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i16> [[TMP6]], i32 1 |
| ; CHECK-NEXT: store i16 [[TMP10]], ptr [[TMP8]], align 2 |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] |
| ; CHECK: for.body: |
| ; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[TMP12:%.*]] = trunc i64 [[I]] to i32 |
| ; CHECK-NEXT: [[TMP13:%.*]] = add i32 [[A]], [[TMP12]] |
| ; CHECK-NEXT: [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16 |
| ; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[I]], i32 1 |
| ; CHECK-NEXT: store i16 [[TMP14]], ptr [[TMP15]], align 2 |
| ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; CHECK-NEXT: [[TMP16:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[TMP16]], [[N]] |
| ; CHECK-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] |
| ; CHECK: for.end: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @iv_vector_and_scalar_users( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; IND-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; IND-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590 |
| ; IND-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0 |
| ; IND-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; IND-NEXT: [[TMP4:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]] |
| ; IND-NEXT: [[TMP5:%.*]] = trunc <2 x i32> [[TMP4]] to <2 x i16> |
| ; IND-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; IND-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP3]], i32 1 |
| ; IND-NEXT: [[TMP8:%.*]] = extractelement <2 x i16> [[TMP5]], i64 0 |
| ; IND-NEXT: store i16 [[TMP8]], ptr [[TMP6]], align 2 |
| ; IND-NEXT: [[TMP9:%.*]] = extractelement <2 x i16> [[TMP5]], i64 1 |
| ; IND-NEXT: store i16 [[TMP9]], ptr [[TMP7]], align 2 |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 |
| ; IND-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; IND-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; IND-NEXT: br label [[FOR_BODY:%.*]] |
| ; IND: for.body: |
| ; IND-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[TMP11:%.*]] = trunc i64 [[I]] to i32 |
| ; IND-NEXT: [[TMP12:%.*]] = add i32 [[A]], [[TMP11]] |
| ; IND-NEXT: [[TMP13:%.*]] = trunc i32 [[TMP12]] to i16 |
| ; IND-NEXT: [[TMP14:%.*]] = getelementptr inbounds nuw [[PAIR_I16]], ptr [[P]], i64 [[I]], i32 1 |
| ; IND-NEXT: store i16 [[TMP13]], ptr [[TMP14]], align 2 |
| ; IND-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; IND-NEXT: [[TMP15:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; IND-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP15]] |
| ; IND-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] |
| ; IND: for.end: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @iv_vector_and_scalar_users( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588 |
| ; UNROLL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0 |
| ; UNROLL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; UNROLL-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; UNROLL-NEXT: [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; UNROLL-NEXT: [[TMP6:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[STEP_ADD]] |
| ; UNROLL-NEXT: [[TMP8:%.*]] = trunc <2 x i32> [[TMP6]] to <2 x i16> |
| ; UNROLL-NEXT: [[TMP9:%.*]] = trunc <2 x i32> [[TMP7]] to <2 x i16> |
| ; UNROLL-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; UNROLL-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP3]], i32 1 |
| ; UNROLL-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP4]], i32 1 |
| ; UNROLL-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP5]], i32 1 |
| ; UNROLL-NEXT: [[TMP14:%.*]] = extractelement <2 x i16> [[TMP8]], i64 0 |
| ; UNROLL-NEXT: store i16 [[TMP14]], ptr [[TMP10]], align 2 |
| ; UNROLL-NEXT: [[TMP15:%.*]] = extractelement <2 x i16> [[TMP8]], i64 1 |
| ; UNROLL-NEXT: store i16 [[TMP15]], ptr [[TMP11]], align 2 |
| ; UNROLL-NEXT: [[TMP16:%.*]] = extractelement <2 x i16> [[TMP9]], i64 0 |
| ; UNROLL-NEXT: store i16 [[TMP16]], ptr [[TMP12]], align 2 |
| ; UNROLL-NEXT: [[TMP17:%.*]] = extractelement <2 x i16> [[TMP9]], i64 1 |
| ; UNROLL-NEXT: store i16 [[TMP17]], ptr [[TMP13]], align 2 |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 4) |
| ; UNROLL-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL: for.body: |
| ; UNROLL-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[TMP19:%.*]] = trunc i64 [[I]] to i32 |
| ; UNROLL-NEXT: [[TMP20:%.*]] = add i32 [[A]], [[TMP19]] |
| ; UNROLL-NEXT: [[TMP21:%.*]] = trunc i32 [[TMP20]] to i16 |
| ; UNROLL-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[PAIR_I16]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NEXT: store i16 [[TMP21]], ptr [[TMP22]], align 2 |
| ; UNROLL-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NEXT: [[TMP23:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; UNROLL-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP23]] |
| ; UNROLL-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] |
| ; UNROLL: for.end: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @iv_vector_and_scalar_users( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0 |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 3 |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]] |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[STEP_ADD]] |
| ; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = trunc <2 x i32> [[TMP7]] to <2 x i16> |
| ; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = trunc <2 x i32> [[TMP8]] to <2 x i16> |
| ; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], ptr [[P:%.*]], i64 [[TMP3]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP4]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP5]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP6]], i32 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = extractelement <2 x i16> [[TMP9]], i32 0 |
| ; UNROLL-NO-IC-NEXT: store i16 [[TMP15]], ptr [[TMP11]], align 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = extractelement <2 x i16> [[TMP9]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i16 [[TMP16]], ptr [[TMP12]], align 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = extractelement <2 x i16> [[TMP10]], i32 0 |
| ; UNROLL-NO-IC-NEXT: store i16 [[TMP17]], ptr [[TMP13]], align 2 |
| ; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = extractelement <2 x i16> [[TMP10]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i16 [[TMP18]], ptr [[TMP14]], align 2 |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], splat (i32 2) |
| ; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]] |
| ; UNROLL-NO-IC: for.body: |
| ; UNROLL-NO-IC-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = trunc i64 [[I]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = add i32 [[A]], [[TMP20]] |
| ; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = trunc i32 [[TMP21]] to i16 |
| ; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[I]], i32 1 |
| ; UNROLL-NO-IC-NEXT: store i16 [[TMP22]], ptr [[TMP23]], align 2 |
| ; UNROLL-NO-IC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[COND:%.*]] = icmp eq i32 [[TMP24]], [[N]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] |
| ; UNROLL-NO-IC: for.end: |
| ; UNROLL-NO-IC-NEXT: ret void |
| ; |
| ; INTERLEAVE-LABEL: @iv_vector_and_scalar_users( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[N:%.*]], -1 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934584 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[A:%.*]], i64 0 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], splat (i32 4) |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 1 |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 2 |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = or disjoint i64 [[INDEX]], 3 |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = or disjoint i64 [[INDEX]], 4 |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = or disjoint i64 [[INDEX]], 5 |
| ; INTERLEAVE-NEXT: [[TMP8:%.*]] = or disjoint i64 [[INDEX]], 6 |
| ; INTERLEAVE-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 7 |
| ; INTERLEAVE-NEXT: [[TMP10:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]] |
| ; INTERLEAVE-NEXT: [[TMP11:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], [[STEP_ADD]] |
| ; INTERLEAVE-NEXT: [[TMP12:%.*]] = trunc <4 x i32> [[TMP10]] to <4 x i16> |
| ; INTERLEAVE-NEXT: [[TMP13:%.*]] = trunc <4 x i32> [[TMP11]] to <4 x i16> |
| ; INTERLEAVE-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP3]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP4]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP5]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP6]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP7]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP20:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP8]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[PAIR_I16]], ptr [[P]], i64 [[TMP9]], i32 1 |
| ; INTERLEAVE-NEXT: [[TMP22:%.*]] = extractelement <4 x i16> [[TMP12]], i64 0 |
| ; INTERLEAVE-NEXT: store i16 [[TMP22]], ptr [[TMP14]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP23:%.*]] = extractelement <4 x i16> [[TMP12]], i64 1 |
| ; INTERLEAVE-NEXT: store i16 [[TMP23]], ptr [[TMP15]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP24:%.*]] = extractelement <4 x i16> [[TMP12]], i64 2 |
| ; INTERLEAVE-NEXT: store i16 [[TMP24]], ptr [[TMP16]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP25:%.*]] = extractelement <4 x i16> [[TMP12]], i64 3 |
| ; INTERLEAVE-NEXT: store i16 [[TMP25]], ptr [[TMP17]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP26:%.*]] = extractelement <4 x i16> [[TMP13]], i64 0 |
| ; INTERLEAVE-NEXT: store i16 [[TMP26]], ptr [[TMP18]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP27:%.*]] = extractelement <4 x i16> [[TMP13]], i64 1 |
| ; INTERLEAVE-NEXT: store i16 [[TMP27]], ptr [[TMP19]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP28:%.*]] = extractelement <4 x i16> [[TMP13]], i64 2 |
| ; INTERLEAVE-NEXT: store i16 [[TMP28]], ptr [[TMP20]], align 2 |
| ; INTERLEAVE-NEXT: [[TMP29:%.*]] = extractelement <4 x i16> [[TMP13]], i64 3 |
| ; INTERLEAVE-NEXT: store i16 [[TMP29]], ptr [[TMP21]], align 2 |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 8) |
| ; INTERLEAVE-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: br label [[FOR_BODY:%.*]] |
| ; INTERLEAVE: for.body: |
| ; INTERLEAVE-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; INTERLEAVE-NEXT: [[TMP31:%.*]] = trunc i64 [[I]] to i32 |
| ; INTERLEAVE-NEXT: [[TMP32:%.*]] = add i32 [[A]], [[TMP31]] |
| ; INTERLEAVE-NEXT: [[TMP33:%.*]] = trunc i32 [[TMP32]] to i16 |
| ; INTERLEAVE-NEXT: [[TMP34:%.*]] = getelementptr inbounds nuw [[PAIR_I16]], ptr [[P]], i64 [[I]], i32 1 |
| ; INTERLEAVE-NEXT: store i16 [[TMP33]], ptr [[TMP34]], align 2 |
| ; INTERLEAVE-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1 |
| ; INTERLEAVE-NEXT: [[TMP35:%.*]] = trunc i64 [[I_NEXT]] to i32 |
| ; INTERLEAVE-NEXT: [[COND:%.*]] = icmp eq i32 [[N]], [[TMP35]] |
| ; INTERLEAVE-NEXT: br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]] |
| ; INTERLEAVE: for.end: |
| ; INTERLEAVE-NEXT: ret void |
| ; |
| entry: |
| br label %for.body |
| |
| for.body: |
| %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] |
| %0 = trunc i64 %i to i32 |
| %1 = add i32 %a, %0 |
| %2 = trunc i32 %1 to i16 |
| %3 = getelementptr inbounds %pair.i16, ptr %p, i64 %i, i32 1 |
| store i16 %2, ptr %3, align 2 |
| %i.next = add nuw nsw i64 %i, 1 |
| %4 = trunc i64 %i.next to i32 |
| %cond = icmp eq i32 %4, %n |
| br i1 %cond, label %for.end, label %for.body |
| |
| for.end: |
| ret void |
| } |
| |
| ; Make sure that the loop exit count computation does not overflow for i8 and |
| ; i16. The exit count of these loops is i8/i16 max + 1. If we don't cast the |
| ; induction variable to a bigger type the exit count computation will overflow |
| ; to 0. |
| ; PR17532 |
| |
| define i32 @i8_loop() nounwind readnone ssp uwtable { |
| ; CHECK-LABEL: @i8_loop( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0]] = and <2 x i32> [[VEC_PHI]], splat (i32 4) |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256 |
| ; CHECK-NEXT: br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP0]]) |
| ; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP2]], [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[LOOP:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[B_0:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; CHECK-NEXT: [[B_NEXT]] = add i8 [[B_0]], -1 |
| ; CHECK-NEXT: [[EC:%.*]] = icmp eq i8 [[B_NEXT]], 0 |
| ; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; IND-LABEL: @i8_loop( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256 |
| ; IND-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: br label [[LOOP:%.*]] |
| ; IND: loop: |
| ; IND-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]] |
| ; IND: exit: |
| ; IND-NEXT: ret i32 0 |
| ; |
| ; UNROLL-LABEL: @i8_loop( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256 |
| ; UNROLL-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL: loop: |
| ; UNROLL-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]] |
| ; UNROLL: exit: |
| ; UNROLL-NEXT: ret i32 0 |
| ; |
| ; UNROLL-NO-IC-LABEL: @i8_loop( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ splat (i32 -1), [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP0]] = and <2 x i32> [[VEC_PHI]], splat (i32 4) |
| ; UNROLL-NO-IC-NEXT: [[TMP1]] = and <2 x i32> [[VEC_PHI1]], splat (i32 4) |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = and <2 x i32> [[TMP1]], [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL-NO-IC: loop: |
| ; UNROLL-NO-IC-NEXT: [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[B_0:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; UNROLL-NO-IC-NEXT: [[B_NEXT]] = add i8 [[B_0]], -1 |
| ; UNROLL-NO-IC-NEXT: [[EC:%.*]] = icmp eq i8 [[B_NEXT]], 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]] |
| ; UNROLL-NO-IC: exit: |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; INTERLEAVE-LABEL: @i8_loop( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256 |
| ; INTERLEAVE-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: br label [[LOOP:%.*]] |
| ; INTERLEAVE: loop: |
| ; INTERLEAVE-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]] |
| ; INTERLEAVE: exit: |
| ; INTERLEAVE-NEXT: ret i32 0 |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ] |
| %b.0 = phi i8 [ 0, %entry ], [ %b.next, %loop ] |
| %a.0.and = and i32 %a.0, 4 |
| %b.next = add i8 %b.0, -1 |
| %ec = icmp eq i8 %b.next, 0 |
| br i1 %ec, label %exit, label %loop |
| |
| exit: |
| ret i32 %a.0.and |
| } |
| |
| |
| define i32 @i16_loop() nounwind readnone ssp uwtable { |
| ; CHECK-LABEL: @i16_loop( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP0]] = and <2 x i32> [[VEC_PHI]], splat (i32 4) |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536 |
| ; CHECK-NEXT: br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP0]]) |
| ; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP2]], [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[LOOP:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[B_0:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_0_NEXT:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; CHECK-NEXT: [[B_0_NEXT]] = add i16 [[B_0]], -1 |
| ; CHECK-NEXT: [[EC:%.*]] = icmp eq i16 [[B_0_NEXT]], 0 |
| ; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; IND-LABEL: @i16_loop( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536 |
| ; IND-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: br label [[LOOP:%.*]] |
| ; IND: loop: |
| ; IND-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]] |
| ; IND: exit: |
| ; IND-NEXT: ret i32 0 |
| ; |
| ; UNROLL-LABEL: @i16_loop( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536 |
| ; UNROLL-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL: loop: |
| ; UNROLL-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]] |
| ; UNROLL: exit: |
| ; UNROLL-NEXT: ret i32 0 |
| ; |
| ; UNROLL-NO-IC-LABEL: @i16_loop( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ splat (i32 -1), [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP0]] = and <2 x i32> [[VEC_PHI]], splat (i32 4) |
| ; UNROLL-NO-IC-NEXT: [[TMP1]] = and <2 x i32> [[VEC_PHI1]], splat (i32 4) |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = and <2 x i32> [[TMP1]], [[TMP0]] |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL-NO-IC: loop: |
| ; UNROLL-NO-IC-NEXT: [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[B_0:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_0_NEXT:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; UNROLL-NO-IC-NEXT: [[B_0_NEXT]] = add i16 [[B_0]], -1 |
| ; UNROLL-NO-IC-NEXT: [[EC:%.*]] = icmp eq i16 [[B_0_NEXT]], 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]] |
| ; UNROLL-NO-IC: exit: |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; INTERLEAVE-LABEL: @i16_loop( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536 |
| ; INTERLEAVE-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: br label [[LOOP:%.*]] |
| ; INTERLEAVE: loop: |
| ; INTERLEAVE-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]] |
| ; INTERLEAVE: exit: |
| ; INTERLEAVE-NEXT: ret i32 0 |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ] |
| %b.0 = phi i16 [ 0, %entry ], [ %b.0.next, %loop ] |
| %a.0.and = and i32 %a.0, 4 |
| %b.0.next = add i16 %b.0, -1 |
| %ec = icmp eq i16 %b.0.next, 0 |
| br i1 %ec, label %exit, label %loop |
| |
| exit: |
| ret i32 %a.0.and |
| } |
| |
| ; This loop has a backedge taken count of i32_max. We need to check for this |
| ; condition and can skip vectorizing. |
| |
| define i32 @max_i32_backedgetaken() { |
| ; CHECK-LABEL: @max_i32_backedgetaken( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: br label [[LOOP:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[A_0:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[B_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; CHECK-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; CHECK-NEXT: [[B_NEXT]] = add i32 [[B_0]], -1 |
| ; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0 |
| ; CHECK-NEXT: br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ] |
| ; CHECK-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; IND-LABEL: @max_i32_backedgetaken( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: br label [[LOOP:%.*]] |
| ; IND: loop: |
| ; IND-NEXT: [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH:%.*]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; IND-NEXT: [[B_NEXT]] = add i32 [[B_0]], -1 |
| ; IND-NEXT: [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0 |
| ; IND-NEXT: br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]] |
| ; IND: exit: |
| ; IND-NEXT: ret i32 0 |
| ; |
| ; UNROLL-LABEL: @max_i32_backedgetaken( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL: loop: |
| ; UNROLL-NEXT: [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH:%.*]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; UNROLL-NEXT: [[B_NEXT]] = add i32 [[B_0]], -1 |
| ; UNROLL-NEXT: [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0 |
| ; UNROLL-NEXT: br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]] |
| ; UNROLL: exit: |
| ; UNROLL-NEXT: ret i32 0 |
| ; |
| ; UNROLL-NO-IC-LABEL: @max_i32_backedgetaken( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL-NO-IC: loop: |
| ; UNROLL-NO-IC-NEXT: [[A_0:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ [[A_0_AND:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[B_0:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND]] = and i32 [[A_0]], 4 |
| ; UNROLL-NO-IC-NEXT: [[B_NEXT]] = add i32 [[B_0]], -1 |
| ; UNROLL-NO-IC-NEXT: [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]] |
| ; UNROLL-NO-IC: exit: |
| ; UNROLL-NO-IC-NEXT: [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ] |
| ; UNROLL-NO-IC-NEXT: ret i32 [[A_0_AND_LCSSA]] |
| ; |
| ; INTERLEAVE-LABEL: @max_i32_backedgetaken( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: br label [[LOOP:%.*]] |
| ; INTERLEAVE: loop: |
| ; INTERLEAVE-NEXT: [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH:%.*]] ], [ [[B_NEXT:%.*]], [[LOOP]] ] |
| ; INTERLEAVE-NEXT: [[B_NEXT]] = add i32 [[B_0]], -1 |
| ; INTERLEAVE-NEXT: [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0 |
| ; INTERLEAVE-NEXT: br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP]] |
| ; INTERLEAVE: exit: |
| ; INTERLEAVE-NEXT: ret i32 0 |
| ; |
| entry: |
| br label %loop |
| |
| loop: |
| %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ] |
| %b.0 = phi i32 [ 0, %entry ], [ %b.next, %loop ] |
| %a.0.and = and i32 %a.0, 4 |
| %b.next = add i32 %b.0, -1 |
| %ec = icmp eq i32 %b.next, 0 |
| br i1 %ec, label %exit, label %loop |
| |
| exit: |
| ret i32 %a.0.and |
| } |
| |
| ; When generating the overflow check we must sure that the induction start value |
| ; is defined before the branch to the scalar preheader. |
| |
| |
| |
| @e = global i8 1, align 1 |
| @d = common global i32 0, align 4 |
| @c = common global i32 0, align 4 |
| define i32 @testoverflowcheck() { |
| ; CHECK-LABEL: @testoverflowcheck( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[DOTPR_I:%.*]] = load i8, ptr @e, align 1 |
| ; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4 |
| ; CHECK-NEXT: [[C_PROMOTED_I:%.*]] = load i32, ptr @c, align 4 |
| ; CHECK-NEXT: [[TMP1:%.*]] = sub i8 -1, [[DOTPR_I]] |
| ; CHECK-NEXT: [[TMP2:%.*]] = zext i8 [[TMP1]] to i32 |
| ; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP3]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP3]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; CHECK-NEXT: [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[DOTCAST]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> splat (i32 -1), i32 [[C_PROMOTED_I]], i32 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0 |
| ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[TMP5]] = and <2 x i32> [[BROADCAST_SPLAT]], [[VEC_PHI]] |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP5]]) |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ] |
| ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ] |
| ; CHECK-NEXT: br label [[COND_END_I:%.*]] |
| ; CHECK: cond.end.i: |
| ; CHECK-NEXT: [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ] |
| ; CHECK-NEXT: [[AND3_I:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[AND_I:%.*]], [[COND_END_I]] ] |
| ; CHECK-NEXT: [[AND_I]] = and i32 [[TMP0]], [[AND3_I]] |
| ; CHECK-NEXT: [[INC_I]] = add i8 [[INC4_I]], 1 |
| ; CHECK-NEXT: [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0 |
| ; CHECK-NEXT: br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]] |
| ; CHECK: loopexit: |
| ; CHECK-NEXT: [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ] |
| ; CHECK-NEXT: ret i32 [[AND_I_LCSSA]] |
| ; |
| ; IND-LABEL: @testoverflowcheck( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[DOTPR_I:%.*]] = load i8, ptr @e, align 1 |
| ; IND-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4 |
| ; IND-NEXT: [[C_PROMOTED_I:%.*]] = load i32, ptr @c, align 4 |
| ; IND-NEXT: [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1 |
| ; IND-NEXT: [[TMP2:%.*]] = zext i8 [[TMP1]] to i32 |
| ; IND-NEXT: [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i8 [[DOTPR_I]], -1 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i32 [[TMP3]], 510 |
| ; IND-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; IND-NEXT: [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[DOTCAST]] |
| ; IND-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> <i32 poison, i32 -1>, i32 [[C_PROMOTED_I]], i64 0 |
| ; IND-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0 |
| ; IND-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; IND-NEXT: [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[TMP6:%.*]] = and <2 x i32> [[TMP4]], [[BROADCAST_SPLAT]] |
| ; IND-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP6]]) |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ] |
| ; IND-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ] |
| ; IND-NEXT: br label [[COND_END_I:%.*]] |
| ; IND: cond.end.i: |
| ; IND-NEXT: [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ] |
| ; IND-NEXT: [[AND_I:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]] |
| ; IND-NEXT: [[INC_I]] = add i8 [[INC4_I]], 1 |
| ; IND-NEXT: [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0 |
| ; IND-NEXT: br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]] |
| ; IND: loopexit: |
| ; IND-NEXT: [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ] |
| ; IND-NEXT: ret i32 [[AND_I_LCSSA]] |
| ; |
| ; UNROLL-LABEL: @testoverflowcheck( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[DOTPR_I:%.*]] = load i8, ptr @e, align 1 |
| ; UNROLL-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4 |
| ; UNROLL-NEXT: [[C_PROMOTED_I:%.*]] = load i32, ptr @c, align 4 |
| ; UNROLL-NEXT: [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = zext i8 [[TMP1]] to i32 |
| ; UNROLL-NEXT: [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i8 [[DOTPR_I]], -4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i32 [[TMP3]], 508 |
| ; UNROLL-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; UNROLL-NEXT: [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[DOTCAST]] |
| ; UNROLL-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> <i32 poison, i32 -1>, i32 [[C_PROMOTED_I]], i64 0 |
| ; UNROLL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0 |
| ; UNROLL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[TMP6:%.*]] = and <2 x i32> [[TMP4]], [[BROADCAST_SPLAT]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP6]]) |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ] |
| ; UNROLL-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ] |
| ; UNROLL-NEXT: br label [[COND_END_I:%.*]] |
| ; UNROLL: cond.end.i: |
| ; UNROLL-NEXT: [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ] |
| ; UNROLL-NEXT: [[AND_I:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]] |
| ; UNROLL-NEXT: [[INC_I]] = add i8 [[INC4_I]], 1 |
| ; UNROLL-NEXT: [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0 |
| ; UNROLL-NEXT: br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]] |
| ; UNROLL: loopexit: |
| ; UNROLL-NEXT: [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NEXT: ret i32 [[AND_I_LCSSA]] |
| ; |
| ; UNROLL-NO-IC-LABEL: @testoverflowcheck( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[DOTPR_I:%.*]] = load i8, ptr @e, align 1 |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4 |
| ; UNROLL-NO-IC-NEXT: [[C_PROMOTED_I:%.*]] = load i32, ptr @c, align 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = sub i8 -1, [[DOTPR_I]] |
| ; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = zext i8 [[TMP1]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; UNROLL-NO-IC: vector.ph: |
| ; UNROLL-NO-IC-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP3]], 4 |
| ; UNROLL-NO-IC-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP3]], [[N_MOD_VF]] |
| ; UNROLL-NO-IC-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; UNROLL-NO-IC-NEXT: [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[DOTCAST]] |
| ; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> splat (i32 -1), i32 [[C_PROMOTED_I]], i32 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0 |
| ; UNROLL-NO-IC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL-NO-IC: vector.body: |
| ; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ splat (i32 -1), [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NO-IC-NEXT: [[TMP5]] = and <2 x i32> [[BROADCAST_SPLAT]], [[VEC_PHI]] |
| ; UNROLL-NO-IC-NEXT: [[TMP6]] = and <2 x i32> [[BROADCAST_SPLAT]], [[VEC_PHI1]] |
| ; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]] |
| ; UNROLL-NO-IC: middle.block: |
| ; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = and <2 x i32> [[TMP6]], [[TMP5]] |
| ; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]]) |
| ; UNROLL-NO-IC-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]] |
| ; UNROLL-NO-IC-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL-NO-IC: scalar.ph: |
| ; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ] |
| ; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP8]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ] |
| ; UNROLL-NO-IC-NEXT: br label [[COND_END_I:%.*]] |
| ; UNROLL-NO-IC: cond.end.i: |
| ; UNROLL-NO-IC-NEXT: [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ] |
| ; UNROLL-NO-IC-NEXT: [[AND3_I:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[AND_I:%.*]], [[COND_END_I]] ] |
| ; UNROLL-NO-IC-NEXT: [[AND_I]] = and i32 [[TMP0]], [[AND3_I]] |
| ; UNROLL-NO-IC-NEXT: [[INC_I]] = add i8 [[INC4_I]], 1 |
| ; UNROLL-NO-IC-NEXT: [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0 |
| ; UNROLL-NO-IC-NEXT: br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]] |
| ; UNROLL-NO-IC: loopexit: |
| ; UNROLL-NO-IC-NEXT: [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ] |
| ; UNROLL-NO-IC-NEXT: ret i32 [[AND_I_LCSSA]] |
| ; |
| ; INTERLEAVE-LABEL: @testoverflowcheck( |
| ; INTERLEAVE-NEXT: entry: |
| ; INTERLEAVE-NEXT: [[DOTPR_I:%.*]] = load i8, ptr @e, align 1 |
| ; INTERLEAVE-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4 |
| ; INTERLEAVE-NEXT: [[C_PROMOTED_I:%.*]] = load i32, ptr @c, align 4 |
| ; INTERLEAVE-NEXT: [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1 |
| ; INTERLEAVE-NEXT: [[TMP2:%.*]] = zext i8 [[TMP1]] to i32 |
| ; INTERLEAVE-NEXT: [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1 |
| ; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i8 [[DOTPR_I]], -8 |
| ; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| ; INTERLEAVE: vector.ph: |
| ; INTERLEAVE-NEXT: [[N_VEC:%.*]] = and i32 [[TMP3]], 504 |
| ; INTERLEAVE-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; INTERLEAVE-NEXT: [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[DOTCAST]] |
| ; INTERLEAVE-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> <i32 poison, i32 -1, i32 -1, i32 -1>, i32 [[C_PROMOTED_I]], i64 0 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0 |
| ; INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer |
| ; INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; INTERLEAVE: vector.body: |
| ; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8 |
| ; INTERLEAVE-NEXT: [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]] |
| ; INTERLEAVE: middle.block: |
| ; INTERLEAVE-NEXT: [[TMP6:%.*]] = and <4 x i32> [[TMP4]], [[BROADCAST_SPLAT]] |
| ; INTERLEAVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP6]]) |
| ; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]] |
| ; INTERLEAVE-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; INTERLEAVE: scalar.ph: |
| ; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ] |
| ; INTERLEAVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ] |
| ; INTERLEAVE-NEXT: br label [[COND_END_I:%.*]] |
| ; INTERLEAVE: cond.end.i: |
| ; INTERLEAVE-NEXT: [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ] |
| ; INTERLEAVE-NEXT: [[AND_I:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]] |
| ; INTERLEAVE-NEXT: [[INC_I]] = add i8 [[INC4_I]], 1 |
| ; INTERLEAVE-NEXT: [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0 |
| ; INTERLEAVE-NEXT: br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]] |
| ; INTERLEAVE: loopexit: |
| ; INTERLEAVE-NEXT: [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ] |
| ; INTERLEAVE-NEXT: ret i32 [[AND_I_LCSSA]] |
| ; |
| entry: |
| %.pr.i = load i8, ptr @e, align 1 |
| %0 = load i32, ptr @d, align 4 |
| %c.promoted.i = load i32, ptr @c, align 4 |
| br label %cond.end.i |
| |
| cond.end.i: |
| %inc4.i = phi i8 [ %.pr.i, %entry ], [ %inc.i, %cond.end.i ] |
| %and3.i = phi i32 [ %c.promoted.i, %entry ], [ %and.i, %cond.end.i ] |
| %and.i = and i32 %0, %and3.i |
| %inc.i = add i8 %inc4.i, 1 |
| %tobool.i = icmp eq i8 %inc.i, 0 |
| br i1 %tobool.i, label %loopexit, label %cond.end.i |
| |
| loopexit: |
| ret i32 %and.i |
| } |
| |
| ; The SCEV expression of %sphi is (zext i8 {%t,+,1}<%loop> to i32) |
| ; In order to recognize %sphi as an induction PHI and vectorize this loop, |
| ; we need to convert the SCEV expression into an AddRecExpr. |
| ; The expression gets converted to {zext i8 %t to i32,+,1}. |
| |
| define void @wrappingindvars1(i8 %t, i32 %len, ptr %A) { |
| ; CHECK-LABEL: @wrappingindvars1( |
| ; CHECK-NEXT: entry: |
| ; CHECK-NEXT: [[ST:%.*]] = zext i8 [[T:%.*]] to i16 |
| ; CHECK-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i32 |
| ; CHECK-NEXT: [[ECMP:%.*]] = icmp ult i16 [[ST]], 42 |
| ; CHECK-NEXT: br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]] |
| ; CHECK: loop.preheader: |
| ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1 |
| ; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2 |
| ; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] |
| ; CHECK: vector.scevcheck: |
| ; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[LEN]] to i8 |
| ; CHECK-NEXT: [[TMP2:%.*]] = add i8 [[T]], [[TMP1]] |
| ; CHECK-NEXT: [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[T]] |
| ; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt i32 [[LEN]], 255 |
| ; CHECK-NEXT: [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]] |
| ; CHECK-NEXT: [[TMP6:%.*]] = trunc i32 [[LEN]] to i8 |
| ; CHECK-NEXT: [[TMP7:%.*]] = add i8 [[T]], [[TMP6]] |
| ; CHECK-NEXT: [[TMP8:%.*]] = icmp slt i8 [[TMP7]], [[T]] |
| ; CHECK-NEXT: [[TMP9:%.*]] = icmp ugt i32 [[LEN]], 255 |
| ; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]] |
| ; CHECK-NEXT: [[TMP11:%.*]] = or i1 [[TMP5]], [[TMP10]] |
| ; CHECK-NEXT: br i1 [[TMP11]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; CHECK: vector.ph: |
| ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 2 |
| ; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]] |
| ; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; CHECK-NEXT: [[IND_END:%.*]] = add i8 [[T]], [[DOTCAST]] |
| ; CHECK-NEXT: [[IND_END2:%.*]] = add i32 [[EXT]], [[N_VEC]] |
| ; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i64 0 |
| ; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; CHECK-NEXT: [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1> |
| ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; CHECK: vector.body: |
| ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; CHECK-NEXT: [[DOTCAST4:%.*]] = trunc i32 [[INDEX]] to i8 |
| ; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i8 [[T]], [[DOTCAST4]] |
| ; CHECK-NEXT: [[TMP12:%.*]] = add i8 [[OFFSET_IDX]], 0 |
| ; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i8 [[TMP12]] |
| ; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[TMP13]], i32 0 |
| ; CHECK-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP14]], align 4 |
| ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]] |
| ; CHECK: middle.block: |
| ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]] |
| ; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; CHECK: scalar.ph: |
| ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ], [ [[T]], [[LOOP_PREHEADER]] ] |
| ; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[LOOP_PREHEADER]] ] |
| ; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ] |
| ; CHECK-NEXT: br label [[LOOP:%.*]] |
| ; CHECK: loop: |
| ; CHECK-NEXT: [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ] |
| ; CHECK-NEXT: [[PTR:%.*]] = getelementptr inbounds i32, ptr [[A]], i8 [[IDX]] |
| ; CHECK-NEXT: store i32 [[SPHI]], ptr [[PTR]], align 4 |
| ; CHECK-NEXT: [[IDX_INC]] = add i8 [[IDX]], 1 |
| ; CHECK-NEXT: [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32 |
| ; CHECK-NEXT: [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1 |
| ; CHECK-NEXT: [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]] |
| ; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]] |
| ; CHECK: exit.loopexit: |
| ; CHECK-NEXT: br label [[EXIT]] |
| ; CHECK: exit: |
| ; CHECK-NEXT: ret void |
| ; |
| ; IND-LABEL: @wrappingindvars1( |
| ; IND-NEXT: entry: |
| ; IND-NEXT: [[EXT:%.*]] = zext i8 [[T:%.*]] to i32 |
| ; IND-NEXT: [[ECMP:%.*]] = icmp ult i8 [[T]], 42 |
| ; IND-NEXT: br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]] |
| ; IND: loop.preheader: |
| ; IND-NEXT: [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1 |
| ; IND-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2 |
| ; IND-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] |
| ; IND: vector.scevcheck: |
| ; IND-NEXT: [[TMP1:%.*]] = trunc i32 [[LEN]] to i8 |
| ; IND-NEXT: [[TMP2:%.*]] = xor i8 [[T]], -1 |
| ; IND-NEXT: [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]] |
| ; IND-NEXT: [[TMP4:%.*]] = trunc i32 [[LEN]] to i8 |
| ; IND-NEXT: [[TMP5:%.*]] = add i8 [[T]], [[TMP4]] |
| ; IND-NEXT: [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]] |
| ; IND-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255 |
| ; IND-NEXT: [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]] |
| ; IND-NEXT: [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]] |
| ; IND-NEXT: br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; IND: vector.ph: |
| ; IND-NEXT: [[N_VEC:%.*]] = and i32 [[TMP0]], 510 |
| ; IND-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; IND-NEXT: [[IND_END:%.*]] = add i8 [[T]], [[DOTCAST]] |
| ; IND-NEXT: [[IND_END2:%.*]] = add nuw nsw i32 [[N_VEC]], [[EXT]] |
| ; IND-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i64 0 |
| ; IND-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; IND-NEXT: [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1> |
| ; IND-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; IND: vector.body: |
| ; IND-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; IND-NEXT: [[DOTCAST4:%.*]] = trunc i32 [[INDEX]] to i8 |
| ; IND-NEXT: [[OFFSET_IDX:%.*]] = add i8 [[T]], [[DOTCAST4]] |
| ; IND-NEXT: [[TMP10:%.*]] = sext i8 [[OFFSET_IDX]] to i64 |
| ; IND-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP10]] |
| ; IND-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP11]], align 4 |
| ; IND-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 |
| ; IND-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; IND-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]] |
| ; IND: middle.block: |
| ; IND-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]] |
| ; IND-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; IND: scalar.ph: |
| ; IND-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ], [ [[T]], [[LOOP_PREHEADER]] ] |
| ; IND-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[LOOP_PREHEADER]] ] |
| ; IND-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ] |
| ; IND-NEXT: br label [[LOOP:%.*]] |
| ; IND: loop: |
| ; IND-NEXT: [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ] |
| ; IND-NEXT: [[TMP13:%.*]] = sext i8 [[IDX]] to i64 |
| ; IND-NEXT: [[PTR:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP13]] |
| ; IND-NEXT: store i32 [[SPHI]], ptr [[PTR]], align 4 |
| ; IND-NEXT: [[IDX_INC]] = add i8 [[IDX]], 1 |
| ; IND-NEXT: [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32 |
| ; IND-NEXT: [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1 |
| ; IND-NEXT: [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]] |
| ; IND-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]] |
| ; IND: exit.loopexit: |
| ; IND-NEXT: br label [[EXIT]] |
| ; IND: exit: |
| ; IND-NEXT: ret void |
| ; |
| ; UNROLL-LABEL: @wrappingindvars1( |
| ; UNROLL-NEXT: entry: |
| ; UNROLL-NEXT: [[EXT:%.*]] = zext i8 [[T:%.*]] to i32 |
| ; UNROLL-NEXT: [[ECMP:%.*]] = icmp ult i8 [[T]], 42 |
| ; UNROLL-NEXT: br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]] |
| ; UNROLL: loop.preheader: |
| ; UNROLL-NEXT: [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1 |
| ; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4 |
| ; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] |
| ; UNROLL: vector.scevcheck: |
| ; UNROLL-NEXT: [[TMP1:%.*]] = trunc i32 [[LEN]] to i8 |
| ; UNROLL-NEXT: [[TMP2:%.*]] = xor i8 [[T]], -1 |
| ; UNROLL-NEXT: [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]] |
| ; UNROLL-NEXT: [[TMP4:%.*]] = trunc i32 [[LEN]] to i8 |
| ; UNROLL-NEXT: [[TMP5:%.*]] = add i8 [[T]], [[TMP4]] |
| ; UNROLL-NEXT: [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]] |
| ; UNROLL-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255 |
| ; UNROLL-NEXT: [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]] |
| ; UNROLL-NEXT: [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]] |
| ; UNROLL-NEXT: br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] |
| ; UNROLL: vector.ph: |
| ; UNROLL-NEXT: [[N_VEC:%.*]] = and i32 [[TMP0]], 508 |
| ; UNROLL-NEXT: [[DOTCAST:%.*]] = trunc i32 [[N_VEC]] to i8 |
| ; UNROLL-NEXT: [[IND_END:%.*]] = add i8 [[T]], [[DOTCAST]] |
| ; UNROLL-NEXT: [[IND_END2:%.*]] = add nuw nsw i32 [[N_VEC]], [[EXT]] |
| ; UNROLL-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i64 0 |
| ; UNROLL-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer |
| ; UNROLL-NEXT: [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1> |
| ; UNROLL-NEXT: br label [[VECTOR_BODY:%.*]] |
| ; UNROLL: vector.body: |
| ; UNROLL-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| ; UNROLL-NEXT: [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) |
| ; UNROLL-NEXT: [[DOTCAST4:%.*]] = trunc i32 [[INDEX]] to i8 |
| ; UNROLL-NEXT: [[OFFSET_IDX:%.*]] = add i8 [[T]], [[DOTCAST4]] |
| ; UNROLL-NEXT: [[TMP10:%.*]] = sext i8 [[OFFSET_IDX]] to i64 |
| ; UNROLL-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[TMP10]] |
| ; UNROLL-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP11]], i64 8 |
| ; UNROLL-NEXT: store <2 x i32> [[VEC_IND]], ptr [[TMP11]], align 4 |
| ; UNROLL-NEXT: store <2 x i32> [[STEP_ADD]], ptr [[TMP12]], align 4 |
| ; UNROLL-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| ; UNROLL-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 4) |
| ; UNROLL-NEXT: [[TMP13:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]] |
| ; UNROLL: middle.block: |
| ; UNROLL-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]] |
| ; UNROLL-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] |
| ; UNROLL: scalar.ph: |
| ; UNROLL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ], [ [[T]], [[LOOP_PREHEADER]] ] |
| ; UNROLL-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[LOOP_PREHEADER]] ] |
| ; UNROLL-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ] |
| ; UNROLL-NEXT: br label [[LOOP:%.*]] |
| ; UNROLL: loop: |
| ; UNROLL-NEXT: [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ] |
| ; UNROLL-NEXT: [[TMP14:%.*]] = sext i8 [[IDX]] to i64 |
| ; UNROLL-NEXT: [[PTR:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP14]] |
| ; UNROLL-NEXT: store i32 [[SPHI]], ptr [[PTR]], align 4 |
| ; UNROLL-NEXT: [[IDX_INC]] = add i8 [[IDX]], 1 |
| ; UNROLL-NEXT: [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32 |
| ; UNROLL-NEXT: [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1 |
| ; UNROLL-NEXT: [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]] |
| ; UNROLL-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]] |
| ; UNROLL: exit.loopexit: |
| ; UNROLL-NEXT: br label [[EXIT]] |
| ; UNROLL: exit: |
| ; UNROLL-NEXT: ret void |
| ; |
| ; UNROLL-NO-IC-LABEL: @wrappingindvars1( |
| ; UNROLL-NO-IC-NEXT: entry: |
| ; UNROLL-NO-IC-NEXT: [[ST:%.*]] = zext i8 [[T:%.*]] to i16 |
| ; UNROLL-NO-IC-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i32 |
| ; UNROLL-NO-IC-NEXT: [[ECMP:%.*]] = icmp ult i16 [[ST]], 42 |
| ; UNROLL-NO-IC-NEXT: br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]] |
| ; UNROLL-NO-IC: loop.preheader: |
| ; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1 |
| ; UNROLL-NO-IC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4 |
| ; UNROLL-NO-IC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] |
| ; UNROLL-NO-IC: vector.scevcheck: |
| ; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = trunc i32 [[LEN]] to i8 |
|
