repro.py

gistfile1.txt

 
# AOT ID: ['0_inference']
from ctypes import c_void_p, c_long
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.codecache import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
alloc_from_pool = torch.ops.inductor._alloc_from_pool
reinterpret_tensor = torch.ops.inductor._reinterpret_tensor
async_compile = AsyncCompile()
# kernel path: /tmp/torchinductor_tarasg/rf/crfz34h2i76un6i6igtlap7xy6f3ukco3czymaj4tdhrdcoredec.py
# Source Nodes: [padded1], Original ATen: [test_inductor_ops.jagged_to_padded_dense]
# padded1 => jagged_to_padded_dense
triton_poi_fused_jagged_to_padded_dense_0 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[131072], 
    filename=__file__,
    triton_meta={'signature': {0: '*i32', 1: '*i64', 2: '*i64', 3: 'i32', 4: 'i32', 5: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 5), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_jagged_to_padded_dense_0', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, in_ptr1, out_ptr0, ks0, ks1, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x1 = (xindex // ks0)
    x0 = xindex % ks0
    x2 = xindex
    tmp0 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
    tmp4 = tl.load(in_ptr0 + (1 + x1), xmask, eviction_policy='evict_last')
    tmp1 = tmp0 + ks1 + 1
    tmp2 = tmp0 < 0
    tmp3 = tl.where(tmp2, tmp1, tmp0)
    tl.device_assert(((0 <= tmp3) & (tmp3 < 1 + ks1)) | ~xmask, "index out of bounds: 0 <= tmp3 < 1 + ks1")
    tmp5 = tmp3 + x0
    tmp6 = tmp5 < tmp4
    tmp7 = tl.load(in_ptr1 + (tmp3 + x0), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
    tmp8 = tl.full(tmp7.shape, 4611686018427387904, tmp7.dtype)
    tmp9 = tl.where(tmp6, tmp7, tmp8)
    tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
# kernel path: /tmp/torchinductor_tarasg/xb/cxboeyvzkjclnnuaecctdhoa7tuhbvk2j2wobcybczuumws5c5vh.py
# Source Nodes: [unique_0hot], Original ATen: [aten.cat]
# unique_0hot => cat
triton_poi_fused_cat_1 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[131072], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: 'i32', 3: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_1', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, out_ptr0, ks0, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex % ks0
    x2 = xindex
    tmp0 = x0
    tmp1 = tl.full([1], 0, tl.int64)
    tmp2 = tmp0 >= tmp1
    tmp3 = tl.full([1], 1, tl.int64)
    tmp4 = tmp0 < tmp3
    tmp5 = tl.full(tmp1.shape, 0.0, tmp1.dtype)
    tmp6 = tl.where(tmp4, tmp1, tmp5)
    tmp7 = tmp0 >= tmp3
    tmp8 = ks0
    tmp9 = tmp0 < tmp8
    tmp10 = tl.load(in_ptr0 + (x2), tmp7 & xmask, eviction_policy='evict_last', other=0.0)
    tmp11 = tl.load(in_ptr0 + ((-1) + x2), tmp7 & xmask, eviction_policy='evict_last', other=0.0)
    tmp12 = tmp10 - tmp11
    tmp13 = tmp12 > tmp1
    tmp14 = tmp13.to(tl.int8)
    tmp15 = tl.full([1], 1, tl.int8)
    tmp16 = tmp15 - tmp14
    tmp17 = tmp16.to(tl.int64)
    tmp18 = tl.full(tmp17.shape, 0.0, tmp17.dtype)
    tmp19 = tl.where(tmp7, tmp17, tmp18)
    tmp20 = tl.where(tmp4, tmp6, tmp19)
    tl.store(out_ptr0 + (x2), tmp20, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/kh/ckhdqw5p3h5ibw7ycw4thl5uf7m4rtyghbdkarofakutjtmfxj7v.py
# Source Nodes: [mask], Original ATen: [aten.zeros_like]
# mask => full_1
triton_poi_fused_zeros_like_2 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[131072], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_zeros_like_2', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(out_ptr0, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex
    tmp0 = tl.full([1], 0, tl.int64)
    tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/nl/cnlzhvarzel643wxe6ed6obbhv64tfuqtadbiq2iab6whtwb6nif.py
# Source Nodes: [adj_diff, diff, gt, mask, n_unique, setitem], Original ATen: [aten._to_copy, aten.gt, aten.index_put, aten.lift_fresh, aten.sub, aten.sum, aten.zeros_like]
# adj_diff => convert_element_type
# diff => sub_1
# gt => gt
# mask => full_1
# n_unique => sum_1
# setitem => full_default_1, index_put
triton_red_fused__to_copy_gt_index_put_lift_fresh_sub_sum_zeros_like_3 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.reduction(
    size_hints=[1024, 128],
    reduction_hint=ReductionHint.INNER,
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: 'i32', 3: 'i32', 4: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_gt_index_put_lift_fresh_sub_sum_zeros_like_3', 'mutated_arg_names': ['out_ptr1'], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, out_ptr1, ks0, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
    xnumel = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
    xmask = xindex < xnumel
    rbase = tl.arange(0, RBLOCK)[None, :]
    x0 = xindex
    _tmp8 = tl.full([XBLOCK, RBLOCK], 0, tl.int64)
    for roffset in range(0, rnumel, RBLOCK):
        rindex = roffset + rbase
        rmask = rindex < rnumel
        r1 = rindex
        tmp0 = tl.load(in_ptr0 + (1 + r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp1 = tl.load(in_ptr0 + (r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp2 = tmp0 - tmp1
        tmp3 = tl.full([1, 1], 0, tl.int64)
        tmp4 = tmp2 > tmp3
        tmp5 = tmp4.to(tl.int8)
        tmp6 = tmp5.to(tl.int64)
        tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
        tmp9 = _tmp8 + tmp7
        _tmp8 = tl.where(rmask & xmask, tmp9, _tmp8)
    tmp8 = tl.sum(_tmp8, 1)[:, None]
    tmp10 = tmp8 + ks0 + 1
    tmp11 = tmp8 < 0
    tmp12 = tl.where(tmp11, tmp10, tmp8)
    tl.device_assert(((0 <= tmp12) & (tmp12 < 1 + ks0)) | ~xmask, "index out of bounds: 0 <= tmp12 < 1 + ks0")
    tmp13 = tl.full([1, 1], 1, tl.int64)
    tl.store(out_ptr1 + (tmp12 + x0 + (ks0*x0)), tmp13, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/6m/c6m53wyb7krckmgwrwcecvcyjtuu664zyttgnscwbulfnkdmjjjb.py
# Source Nodes: [cumsum, gather_out, setitem_1], Original ATen: [aten.cumsum, aten.gather, aten.index_put, aten.lift_fresh]
# cumsum => cumsum
# gather_out => gather
# setitem_1 => full_default_2, index_put_1
triton_red_fused_cumsum_gather_index_put_lift_fresh_4 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton.jit
def _triton_helper_fn_add0(arg0_0, arg1_0):
    tmp0 = arg0_0 + arg1_0
    return tmp0
@triton_heuristics.reduction(
    size_hints=[1024, 128],
    reduction_hint=ReductionHint.INNER,
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: '*i64', 4: '*i64', 5: 'i32', 6: 'i32', 7: 'i32', 8: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 7), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused_cumsum_gather_index_put_lift_fresh_4', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, ks0, ks1, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
    xnumel = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
    xmask = xindex < xnumel
    rbase = tl.arange(0, RBLOCK)[None, :]
    x0 = xindex
    tmp3 = tl.full([XBLOCK, 1], -1, tl.int64)
    for roffset in range(0, rnumel, RBLOCK):
        rindex = roffset + rbase
        rmask = rindex < rnumel
        r1 = rindex
        tmp0 = tl.load(in_ptr0 + (r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp10 = tl.load(in_ptr1 + (r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp1 = tmp0.to(tl.int64)
        tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
        tmp4, = tl.associative_scan((tmp2,), 1, _triton_helper_fn_add0)
        tmp5, = tl.reduce((tmp2,), 1, _triton_helper_fn_add0)
        tmp6 = tmp3 + tmp5[:, None]
        tmp7 = tmp3 + tmp4
        tmp8 = tl.where(roffset > 0, tmp7, tmp4)
        tmp3 = tl.where(roffset > 0, tmp6, tmp5[:, None])
        tmp9 = (tmp8 != 0)
        tmp11 = tmp10 + ks0 + 1
        tmp12 = tmp10 < 0
        tmp13 = tl.where(tmp12, tmp11, tmp10)
        tl.device_assert(((0 <= tmp13) & (tmp13 < 1 + ks0)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp13 < 1 + ks0")
        tmp14 = tl.load(in_ptr2 + (tmp13 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp15 = tl.full([1, 1], 4611686018427387904, tl.int64)
        tmp16 = tl.where(tmp9, tmp15, tmp14)
        tl.store(out_ptr0 + (r1 + x0 + (ks0*x0)), tmp8, rmask & xmask)
        tl.store(out_ptr1 + (r1 + (2*x0) + (ks0*x0) + (ks1*x0)), tmp16, rmask & xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/gg/cggya3gpn3ejtsdq2ezgyaqi55k6325s5kndciveu5yjwvohcztz.py
# Source Nodes: [cumsum_1, gather_out_1, setitem_4], Original ATen: [aten.cumsum, aten.gather, aten.index_put, aten.lift_fresh]
# cumsum_1 => cumsum_1
# gather_out_1 => gather_1
# setitem_4 => full_default_6, index_put_4
triton_red_fused_cumsum_gather_index_put_lift_fresh_5 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton.jit
def _triton_helper_fn_add0(arg0_0, arg1_0):
    tmp0 = arg0_0 + arg1_0
    return tmp0
@triton_heuristics.reduction(
    size_hints=[1024, 128],
    reduction_hint=ReductionHint.INNER,
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: '*i64', 4: '*i64', 5: 'i32', 6: 'i32', 7: 'i32', 8: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 7), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused_cumsum_gather_index_put_lift_fresh_5', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, ks0, ks1, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
    xnumel = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
    xmask = xindex < xnumel
    rbase = tl.arange(0, RBLOCK)[None, :]
    x0 = xindex
    tmp3 = tl.full([XBLOCK, 1], -1, tl.int64)
    for roffset in range(0, rnumel, RBLOCK):
        rindex = roffset + rbase
        rmask = rindex < rnumel
        r1 = rindex
        tmp0 = tl.load(in_ptr0 + (r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp10 = tl.load(in_ptr1 + (r1 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp1 = tmp0.to(tl.int64)
        tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
        tmp4, = tl.associative_scan((tmp2,), 1, _triton_helper_fn_add0)
        tmp5, = tl.reduce((tmp2,), 1, _triton_helper_fn_add0)
        tmp6 = tmp3 + tmp5[:, None]
        tmp7 = tmp3 + tmp4
        tmp8 = tl.where(roffset > 0, tmp7, tmp4)
        tmp3 = tl.where(roffset > 0, tmp6, tmp5[:, None])
        tmp9 = (tmp8 != 0)
        tmp11 = tmp10 + ks0 + 1
        tmp12 = tmp10 < 0
        tmp13 = tl.where(tmp12, tmp11, tmp10)
        tl.device_assert(((0 <= tmp13) & (tmp13 < 1 + ks0)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp13 < 1 + ks0")
        tmp14 = tl.load(in_ptr2 + (tmp13 + x0 + (ks0*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp15 = tl.full([1, 1], 4611686018427387904, tl.int64)
        tmp16 = tl.where(tmp9, tmp15, tmp14)
        tl.store(out_ptr0 + (r1 + x0 + (ks0*x0)), tmp8, rmask & xmask)
        tl.store(out_ptr1 + (r1 + (2*x0) + (ks0*x0) + (ks1*x0)), tmp16, rmask & xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/et/cetmg7225yixhvwn5ik2slfjataizwcl2zguwayk7ysvmj6xdjdc.py
# Source Nodes: [setitem_2, unique_indices_1], Original ATen: [aten.index_put, aten.lift_fresh, aten.sub]
# setitem_2 => full_default_3, index_put_2
# unique_indices_1 => sub_4
triton_poi_fused_index_put_lift_fresh_sub_6 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[131072], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: 'i32', 4: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 4), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_index_put_lift_fresh_sub_6', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, in_ptr1, out_ptr0, ks0, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex % ks0
    x1 = (xindex // ks0)
    x2 = xindex
    tmp0 = tl.load(in_ptr0 + (x0 + x1 + (ks0*x1)), xmask, eviction_policy='evict_last')
    tmp2 = tl.load(in_ptr1 + (1 + x0 + x1 + (ks0*x1)), xmask, eviction_policy='evict_last')
    tmp3 = tl.load(in_ptr1 + (x0 + x1 + (ks0*x1)), xmask, eviction_policy='evict_last')
    tmp1 = (tmp0 != 0)
    tmp4 = tmp2 - tmp3
    tmp5 = tl.full([1], 0, tl.int64)
    tmp6 = tl.where(tmp1, tmp5, tmp4)
    tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/pf/cpfuizamjfs7clym7hsjrhozara6z4tylhydowigp4ttpalrmdz6.py
# Source Nodes: [joined_padded_counts], Original ATen: [aten.cat]
# joined_padded_counts => cat_5
triton_poi_fused_cat_7 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[262144], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: 'i32', 4: 'i32', 5: 'i32', 6: 'i32', 7: 'i32', 8: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 8), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_7', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, in_ptr1, out_ptr0, ks0, ks1, ks2, ks3, ks4, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex % ks0
    x1 = (xindex // ks0)
    x2 = xindex
    tmp0 = x0
    tmp1 = tl.full([1], 0, tl.int64)
    tmp2 = tmp0 >= tmp1
    tmp3 = ks1
    tmp4 = tmp0 < tmp3
    tmp5 = tl.broadcast_to(ks2, [XBLOCK])
    tmp6 = tmp0 < tmp5
    tmp7 = tmp6 & tmp4
    tmp8 = tl.load(in_ptr0 + (x0 + (ks2*x1)), tmp7 & xmask, eviction_policy='evict_last', other=0.0)
    tmp9 = tl.full(tmp8.shape, 0.0, tmp8.dtype)
    tmp10 = tl.where(tmp7, tmp8, tmp9)
    tmp11 = tmp0 >= tmp5
    tmp12 = tl.broadcast_to(ks1, [XBLOCK])
    tmp13 = tmp0 < tmp12
    tmp14 = tmp11 & tmp4
    tmp15 = tl.load(in_ptr0 + ((-1) + ks2 + (ks2*x1)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
    tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
    tmp17 = tl.where(tmp14, tmp15, tmp16)
    tmp18 = tl.where(tmp6, tmp10, tmp17)
    tmp19 = tl.full(tmp18.shape, 0.0, tmp18.dtype)
    tmp20 = tl.where(tmp4, tmp18, tmp19)
    tmp21 = tmp0 >= tmp3
    tmp22 = ks0
    tmp23 = tmp0 < tmp22
    tmp24 = (-1) + x0 + ((-1)*ks2)
    tmp25 = tmp24 >= tmp1
    tmp26 = tl.broadcast_to(ks3, [XBLOCK])
    tmp27 = tmp24 < tmp26
    tmp28 = tmp27 & tmp21
    tmp29 = tl.load(in_ptr1 + ((-1) + x0 + ((-1)*ks2) + (ks3*x1)), tmp28 & xmask, eviction_policy='evict_last', other=0.0)
    tmp30 = tl.full(tmp29.shape, 0.0, tmp29.dtype)
    tmp31 = tl.where(tmp28, tmp29, tmp30)
    tmp32 = tmp24 >= tmp26
    tmp33 = tl.broadcast_to(ks4, [XBLOCK])
    tmp34 = tmp24 < tmp33
    tmp35 = tmp32 & tmp21
    tmp36 = tl.load(in_ptr1 + ((-1) + ks3 + (ks3*x1)), tmp35 & xmask, eviction_policy='evict_last', other=0.0)
    tmp37 = tl.full(tmp36.shape, 0.0, tmp36.dtype)
    tmp38 = tl.where(tmp35, tmp36, tmp37)
    tmp39 = tl.where(tmp27, tmp31, tmp38)
    tmp40 = tl.full(tmp39.shape, 0.0, tmp39.dtype)
    tmp41 = tl.where(tmp21, tmp39, tmp40)
    tmp42 = tl.where(tmp4, tmp20, tmp41)
    tl.store(out_ptr0 + (x2), tmp42, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/ce/ccedtkpzbs66fozel5to6qjqjdjurk5kdgejw6n63xvxefut46ie.py
# Source Nodes: [min_1], Original ATen: [aten.min]
# min_1 => min_1
triton_poi_fused_min_8 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[262144], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: 'i32', 4: 'i32', 5: 'i32', 6: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 6), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_min_8', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, in_ptr1, out_ptr0, ks0, ks1, ks2, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x1 = (xindex // ks0)
    x2 = xindex
    x0 = xindex % ks0
    tmp0 = x1
    tmp1 = tl.full([1], 0, tl.int64)
    tmp2 = tmp0 >= tmp1
    tmp3 = tl.full([1], 1024, tl.int64)
    tmp4 = tmp0 < tmp3
    tmp5 = tl.load(in_ptr0 + (x2), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
    tmp6 = tmp5 + ks1 + ks2 + 2
    tmp7 = tmp5 < 0
    tmp8 = tl.where(tmp7, tmp6, tmp5)
    tl.device_assert(((0 <= tmp8) & (tmp8 < 2 + ks1 + ks2)) | ~(tmp4 & xmask), "index out of bounds: 0 <= tmp8 < 2 + ks1 + ks2")
    tmp9 = tl.load(in_ptr1 + (tmp8 + (2*x1) + (ks1*x1) + (ks2*x1)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
    tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
    tmp11 = tl.where(tmp4, tmp9, tmp10)
    tmp12 = tmp0 >= tmp3
    tmp13 = tl.full([1], 2048, tl.int64)
    tmp14 = tmp0 < tmp13
    tmp15 = tl.load(in_ptr0 + ((-2048) + ((-1024)*ks1) + ((-1024)*ks2) + (2*x1) + (ks1*x1) + (ks2*x1) + ((x0 + ((3 + ks1 + ks2) % ks0)) % ks0)), tmp12 & xmask, eviction_policy='evict_last', other=0.0)
    tmp16 = tmp15 + ks1 + ks2 + 2
    tmp17 = tmp15 < 0
    tmp18 = tl.where(tmp17, tmp16, tmp15)
    tl.device_assert(((0 <= tmp18) & (tmp18 < 2 + ks1 + ks2)) | ~(tmp12 & xmask), "index out of bounds: 0 <= tmp18 < 2 + ks1 + ks2")
    tmp19 = tl.load(in_ptr1 + ((-2048) + tmp18 + ((-1024)*ks1) + ((-1024)*ks2) + (2*x1) + (ks1*x1) + (ks2*x1)), tmp12 & xmask, eviction_policy='evict_last', other=0.0)
    tmp20 = tl.full(tmp19.shape, 0.0, tmp19.dtype)
    tmp21 = tl.where(tmp12, tmp19, tmp20)
    tmp22 = tl.where(tmp4, tmp11, tmp21)
    tmp23 = 1024 + x1
    tmp24 = tmp23 >= tmp1
    tmp25 = tmp23 < tmp3
    tmp26 = tl.load(in_ptr0 + (2048 + x2 + (1024*ks1) + (1024*ks2)), tmp25 & xmask, eviction_policy='evict_last', other=0.0)
    tmp27 = tmp26 + ks1 + ks2 + 2
    tmp28 = tmp26 < 0
    tmp29 = tl.where(tmp28, tmp27, tmp26)
    tl.device_assert(((0 <= tmp29) & (tmp29 < 2 + ks1 + ks2)) | ~(xmask & tmp25), "index out of bounds: 0 <= tmp29 < 2 + ks1 + ks2")
    tmp30 = tl.load(in_ptr1 + (2048 + tmp29 + (2*x1) + (1024*ks1) + (1024*ks2) + (ks1*x1) + (ks2*x1)), tmp25 & xmask, eviction_policy='evict_last', other=0.0)
    tmp31 = tl.full(tmp30.shape, 0.0, tmp30.dtype)
    tmp32 = tl.where(tmp25, tmp30, tmp31)
    tmp33 = tmp23 >= tmp3
    tmp34 = tmp23 < tmp13
    tmp35 = tl.load(in_ptr0 + ((2*x1) + (ks1*x1) + (ks2*x1) + ((x0 + ((3 + ks1 + ks2) % ks0)) % ks0)), tmp33 & xmask, eviction_policy='evict_last', other=0.0)
    tmp36 = tmp35 + ks1 + ks2 + 2
    tmp37 = tmp35 < 0
    tmp38 = tl.where(tmp37, tmp36, tmp35)
    tl.device_assert(((0 <= tmp38) & (tmp38 < 2 + ks1 + ks2)) | ~(xmask & tmp33), "index out of bounds: 0 <= tmp38 < 2 + ks1 + ks2")
    tmp39 = tl.load(in_ptr1 + (tmp38 + (2*x1) + (ks1*x1) + (ks2*x1)), tmp33 & xmask, eviction_policy='evict_last', other=0.0)
    tmp40 = tl.full(tmp39.shape, 0.0, tmp39.dtype)
    tmp41 = tl.where(tmp33, tmp39, tmp40)
    tmp42 = tl.where(tmp25, tmp32, tmp41)
    tmp43 = triton_helpers.minimum(tmp22, tmp42)
    tl.store(out_ptr0 + (x2), tmp43, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/j5/cj5tpoxiug34serot5wjm52idc2zg37fxasdgotwwdomnlu45xch.py
# Source Nodes: [diff_4, eq, joined_padded_lengths, padded_diff, padding_mask, stack_5, stack_7, to_4], Original ATen: [aten._to_copy, aten.eq, aten.gt, aten.mul, aten.stack, aten.sub, aten.sum]
# diff_4 => sub_11
# eq => eq
# joined_padded_lengths => sum_3
# padded_diff => mul
# padding_mask => gt_2
# stack_5 => cat_12
# stack_7 => cat_9
# to_4 => convert_element_type_4
triton_red_fused__to_copy_eq_gt_mul_stack_sub_sum_9 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.reduction(
    size_hints=[1024, 256],
    reduction_hint=ReductionHint.DEFAULT,
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: '*i8', 4: '*i64', 5: '*i64', 6: '*i64', 7: '*i64', 8: 'i32', 9: 'i32', 10: 'i32', 11: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 6, 10), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_eq_gt_mul_stack_sub_sum_9', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr2, out_ptr3, out_ptr4, out_ptr5, ks0, ks1, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
    xnumel = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
    xmask = xindex < xnumel
    rbase = tl.arange(0, RBLOCK)[None, :]
    x0 = xindex
    _tmp15 = tl.full([XBLOCK, RBLOCK], 0, tl.int64)
    for roffset in range(0, rnumel, RBLOCK):
        rindex = roffset + rbase
        rmask = rindex < rnumel
        r1 = rindex
        tmp0 = tl.load(in_ptr0 + (1 + r1 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp1 = tl.load(in_ptr0 + (r1 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp5 = tl.load(in_ptr1 + (r1 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp2 = tmp0 - tmp1
        tmp3 = tl.full([1, 1], 0, tl.int64)
        tmp4 = tmp2 == tmp3
        tmp6 = tmp5 + ks0 + ks1 + 2
        tmp7 = tmp5 < 0
        tmp8 = tl.where(tmp7, tmp6, tmp5)
        tl.device_assert(((0 <= tmp8) & (tmp8 < 2 + ks0 + ks1)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp8 < 2 + ks0 + ks1")
        tmp9 = tl.load(in_ptr2 + (tmp8 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp10 = tmp9 > tmp3
        tmp11 = tmp4 & tmp10
        tmp12 = tmp11.to(tl.int8)
        tmp13 = tmp11.to(tl.int64)
        tmp14 = tl.broadcast_to(tmp13, [XBLOCK, RBLOCK])
        tmp16 = _tmp15 + tmp14
        _tmp15 = tl.where(rmask & xmask, tmp16, _tmp15)
        tl.store(out_ptr0 + (r1 + x0 + (ks0*x0) + (ks1*x0)), tmp12, rmask & xmask)
    tmp15 = tl.sum(_tmp15, 1)[:, None]
    tmp17 = 1 + ks0 + ks1
    tmp18 = tmp17 - tmp15
    tl.store(out_ptr2 + (2*x0), tmp15, xmask)
    tl.store(out_ptr3 + (2*x0), tmp18, xmask)
    tl.store(out_ptr4 + (2*x0), tmp15, xmask)
    tl.store(out_ptr5 + (2*x0), tmp18, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/wx/cwxkhk4qmjok4eeufvw4k4qlxjfqtkmdeh4d6kpathqmv3phoj2y.py
# Source Nodes: [count_pairwise_min, count_pairwise_min_1, gather_3, joined_padded_2, padding_mask, rowwise_counts, sum_4], Original ATen: [aten._to_copy, aten.gather, aten.gt, aten.mul, aten.sum]
# count_pairwise_min => mul_1
# count_pairwise_min_1 => mul_2
# gather_3 => gather_3
# joined_padded_2 => gather_4
# padding_mask => gt_2
# rowwise_counts => convert_element_type_5
# sum_4 => sum_4
triton_red_fused__to_copy_gather_gt_mul_sum_10 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.reduction(
    size_hints=[1024, 256],
    reduction_hint=ReductionHint.DEFAULT,
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: '*i64', 3: '*i64', 4: '*i64', 5: '*i8', 6: '*i64', 7: '*i64', 8: '*i32', 9: 'i32', 10: 'i32', 11: 'i32', 12: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 11), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_gather_gt_mul_sum_10', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, out_ptr1, out_ptr3, ks0, ks1, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
    xnumel = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
    xmask = xindex < xnumel
    rbase = tl.arange(0, RBLOCK)[None, :]
    x0 = xindex
    _tmp21 = tl.full([XBLOCK, RBLOCK], 0, tl.int64)
    for roffset in range(0, rnumel, RBLOCK):
        rindex = roffset + rbase
        rmask = rindex < rnumel
        r1 = rindex
        tmp0 = tl.load(in_ptr0 + (r1 + x0 + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp17 = tl.load(in_ptr5 + (r1 + x0 + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_first', other=0.0)
        tmp1 = tmp0 + ks0 + ks1 + 2
        tmp2 = tmp0 < 0
        tmp3 = tl.where(tmp2, tmp1, tmp0)
        tl.device_assert(((0 <= tmp3) & (tmp3 < 2 + ks0 + ks1)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp3 < 2 + ks0 + ks1")
        tmp4 = tl.load(in_ptr1 + (tmp3 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp5 = tmp0 + ks0 + ks1 + 1
        tmp6 = tl.where(tmp2, tmp5, tmp0)
        tl.device_assert(((0 <= tmp6) & (tmp6 < 1 + ks0 + ks1)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp6 < 1 + ks0 + ks1")
        tmp7 = tl.load(in_ptr2 + (tmp6 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp8 = tl.load(in_ptr3 + (tmp6 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp9 = tmp8 + ks0 + ks1 + 2
        tmp10 = tmp8 < 0
        tmp11 = tl.where(tmp10, tmp9, tmp8)
        tl.device_assert(((0 <= tmp11) & (tmp11 < 2 + ks0 + ks1)) | ~(xmask & rmask), "index out of bounds: 0 <= tmp11 < 2 + ks0 + ks1")
        tmp12 = tl.load(in_ptr4 + (tmp11 + (2*x0) + (ks0*x0) + (ks1*x0)), rmask & xmask, eviction_policy='evict_last')
        tmp13 = tl.full([1, 1], 0, tl.int64)
        tmp14 = tmp12 > tmp13
        tmp15 = tmp14.to(tl.int64)
        tmp16 = tmp7 * tmp15
        tmp18 = tmp17.to(tl.int64)
        tmp19 = tmp16 * tmp18
        tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
        tmp22 = _tmp21 + tmp20
        _tmp21 = tl.where(rmask & xmask, tmp22, _tmp21)
        tl.store(out_ptr0 + (r1 + x0 + (ks0*x0) + (ks1*x0)), tmp4, rmask & xmask)
        tl.store(out_ptr1 + (r1 + x0 + (ks0*x0) + (ks1*x0)), tmp19, rmask & xmask)
    tmp21 = tl.sum(_tmp21, 1)[:, None]
    tmp23 = tmp21.to(tl.int32)
    tl.store(out_ptr3 + (x0), tmp23, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/3w/c3wcv6zxutzi3no2w2skx7bnsftvelmhnzxpaeviajxxgoohun2r.py
# Source Nodes: [unique_val], Original ATen: [aten.index]
# unique_val => index_2
triton_poi_fused_index_11 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[262144], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: 'i32', 3: 'i32', 4: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 4), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_index_11', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, out_ptr0, ks0, ks1, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex
    tmp0 = tl.load(in_ptr0 + ((ks0*(x0 // (1 + ks0 + ks1))) + (ks1*(x0 // (1 + ks0 + ks1))) + (x0 // (1 + ks0 + ks1)) + (x0 % (1 + ks0 + ks1))), xmask, eviction_policy='evict_last')
    tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/qd/cqd4ukqj4klbwsbc2vk5oauespotm4tvxd4yl6toq2ou76rbypvi.py
# Source Nodes: [mask_5], Original ATen: [aten.index_select]
# mask_5 => index_1
triton_poi_fused_index_select_12 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[262144], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i1', 2: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_index_select_12', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x0 = xindex
    tmp0 = tl.load(in_ptr0 + (x0), xmask)
    tmp1 = tmp0 + 2048
    tmp2 = tmp0 < 0
    tmp3 = tl.where(tmp2, tmp1, tmp0)
    tl.device_assert(((0 <= tmp3) & (tmp3 < 2048)) | ~xmask, "index out of bounds: 0 <= tmp3 < 2048")
    tmp4 = tmp3 % 2
    tmp5 = tl.full([1], 0, tl.int64)
    tmp6 = tmp4 >= tmp5
    tmp7 = tl.full([1], 1, tl.int64)
    tmp8 = tmp4 < tmp7
    tmp9 = tl.full([1], True, tl.int1)
    tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
    tmp11 = tl.where(tmp8, tmp9, tmp10)
    tmp12 = tmp4 >= tmp7
    tmp13 = tl.full([1], 2, tl.int64)
    tmp14 = tmp4 < tmp13
    tmp15 = tl.full([1], False, tl.int1)
    tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
    tmp17 = tl.where(tmp12, tmp15, tmp16)
    tmp18 = tl.where(tmp8, tmp11, tmp17)
    tl.store(out_ptr0 + (x0), tmp18, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/4y/c4y4u2zwj2a3ddvkk64qxn7k36hn3pklhua74i4cyqpafigzyyeb.py
# Source Nodes: [expanded_ids], Original ATen: [aten.index_select]
# expanded_ids => index_5
triton_poi_fused_index_select_13 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[8192], 
    filename=__file__,
    triton_meta={'signature': {0: '*i64', 1: '*i64', 2: 'i64', 3: 'i64'}, 'device': 0, 'device_type': 'cuda', 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_index_select_13', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(in_out_ptr0, in_ptr0, ks0, xnumel, XBLOCK : tl.constexpr):
    xoffset = tl.program_id(0).to(tl.int64) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:].to(tl.int64)
    xmask = xindex < xnumel
    x0 = xindex
    tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
    tmp1 = tmp0 + ks0
    tmp2 = tmp0 < 0
    tmp3 = tl.where(tmp2, tmp1, tmp0)
    tl.device_assert(((0 <= tmp3) & (tmp3 < ks0)) | ~xmask, "index out of bounds: 0 <= tmp3 < ks0")
    tmp4 = tl.load(in_ptr0 + (tmp3), xmask, eviction_policy='evict_last')
    tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/v7/cv7i6ynxmyxllr4jgh3p2vywmcplzsxqnecsn24dm5qavbghignv.py
# Source Nodes: [cumsum_4], Original ATen: [aten.cumsum]
# cumsum_4 => cumsum_4
triton_per_fused_cumsum_14 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton.jit
def _triton_helper_fn_add0(arg0_0, arg1_0):
    tmp0 = arg0_0 + arg1_0
    return tmp0
@triton_heuristics.persistent_reduction(
    size_hints=[1, 1024],
    reduction_hint=ReductionHint.INNER,
    filename=__file__,
    triton_meta={'signature': {0: '*i32', 1: '*i32', 2: 'i32', 3: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {2: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 3), equal_to_1=(2,))]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_cumsum_14', 'mutated_arg_names': [], 'no_x_dim': True, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True}
)
@triton.jit
def triton_(in_ptr0, out_ptr0, xnumel, rnumel):
    xnumel = 1
    XBLOCK: tl.constexpr = 1
    rnumel = 1024
    RBLOCK: tl.constexpr = 1024
    xoffset = tl.program_id(0) * XBLOCK
    xindex = tl.full([1], xoffset, tl.int32)
    xmask = xindex < xnumel
    rindex = tl.arange(0, RBLOCK)[:]
    roffset = 0
    rmask = rindex < rnumel
    r0 = rindex
    tmp0 = tl.load(in_ptr0 + (r0), rmask, other=0.0)
    tmp1 = tmp0.to(tl.int32)
    tmp2 = tl.broadcast_to(tmp1, [RBLOCK])
    tmp3, = tl.associative_scan((tmp2,), 0, _triton_helper_fn_add0)
    tl.store(out_ptr0 + (tl.broadcast_to(r0, [RBLOCK])), tmp3, rmask)
''', device_str='cuda')
# kernel path: /tmp/torchinductor_tarasg/dh/cdh7lau3uqef4ntphscaobz4ageuajvzw3ynqvdzf7gsl7bq7occ.py
# Source Nodes: [zeros_2], Original ATen: [aten.zeros]
# zeros_2 => full_default_14
triton_poi_fused_zeros_15 = async_compile.triton('triton_', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor
@triton_heuristics.pointwise(
    size_hints=[1], 
    filename=__file__,
    triton_meta={'signature': {0: '*i32', 1: 'i32'}, 'device': 0, 'device_type': 'cuda', 'constants': {1: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0,), equal_to_1=(1,))]},
    inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_zeros_15', 'mutated_arg_names': [], 'no_x_dim': False, 'backend_hash': '29c521dc5d9e4f14437c83e288b2df511c55eb789833b844b5bcc030e3e984a2', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'is_fbcode': True},
    min_elem_per_thread=0
)
@triton.jit
def triton_(out_ptr0, xnumel, XBLOCK : tl.constexpr):
    xnumel = 1
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    tmp0 = tl.full([1], 0, tl.int32)
    tl.store(out_ptr0 + (tl.full([XBLOCK], 0, tl.int32)), tmp0, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
    arg0_1, arg1_1, arg2_1, arg3_1, arg4_1, arg5_1, arg6_1, arg7_1, arg8_1, arg9_1 = args
    args.clear()
    s0 = arg0_1
    s1 = arg2_1
    s2 = arg5_1
    s3 = arg7_1
    assert_size_stride(arg1_1, (s0, ), (1, ))
    assert_size_stride(arg3_1, (s1, ), (1, ))
    assert_size_stride(arg4_1, (1025, ), (1, ))
    assert_size_stride(arg6_1, (s2, ), (1, ))
    assert_size_stride(arg8_1, (s3, ), (1, ))
    assert_size_stride(arg9_1, (1025, ), (1, ))
    if not s2 >= 2:
        raise RuntimeError('Runtime assertion failed for s2 >= 2')
    buf0 = None
    if not s2 <= 9223372036854775806:
        raise RuntimeError('Runtime assertion failed for s2 <= 9223372036854775806')
    buf1 = None
    if not s1 >= 2:
        raise RuntimeError('Runtime assertion failed for s1 >= 2')
    buf2 = None
    if not s1 <= 9223372036854775806:
        raise RuntimeError('Runtime assertion failed for s1 <= 9223372036854775806')
    buf3 = None
    with torch.cuda._DeviceGuard(0):
        torch.cuda.set_device(0)
        ps0 = 1 + s1
        buf4 = empty_strided_cuda((1024, 1 + s1), (1 + s1, 1), torch.int64)
        # Source Nodes: [padded1], Original ATen: [test_inductor_ops.jagged_to_padded_dense]
        triton_poi_fused_jagged_to_padded_dense_0_xnumel = 1024 + (1024*s1)
        stream0 = get_raw_stream(0)
        triton_poi_fused_jagged_to_padded_dense_0.run(arg4_1, arg1_1, buf4, ps0, s0, triton_poi_fused_jagged_to_padded_dense_0_xnumel, grid=grid(triton_poi_fused_jagged_to_padded_dense_0_xnumel), stream=stream0)
        del arg1_1
        del arg4_1
        # Source Nodes: [padded1, sort], Original ATen: [aten.sort, test_inductor_ops.jagged_to_padded_dense]
        buf5 = aten.sort.default(buf4, 1)
        buf6 = buf5[0]
        del buf5
        buf8 = buf4; del buf4  # reuse
        # Source Nodes: [unique_0hot], Original ATen: [aten.cat]
        triton_poi_fused_cat_1_xnumel = 1024 + (1024*s1)
        triton_poi_fused_cat_1.run(buf6, buf8, ps0, triton_poi_fused_cat_1_xnumel, grid=grid(triton_poi_fused_cat_1_xnumel), stream=stream0)
        # Source Nodes: [sort_1, unique_0hot], Original ATen: [aten.cat, aten.sort]
        buf9 = aten.sort.stable(buf8, stable=True, dim=1)
        buf11 = buf9[1]
        del buf9
        ps1 = 1 + s2
        buf12 = empty_strided_cuda((1024, 1 + s2), (1 + s2, 1), torch.int64)
        # Source Nodes: [padded2], Original ATen: [test_inductor_ops.jagged_to_padded_dense]
        triton_poi_fused_jagged_to_padded_dense_0_xnumel = 1024 + (1024*s2)
        triton_poi_fused_jagged_to_padded_dense_0.run(arg9_1, arg8_1, buf12, ps1, s3, triton_poi_fused_jagged_to_padded_dense_0_xnumel, grid=grid(triton_poi_fused_jagged_to_padded_dense_0_xnumel), stream=stream0)
        del arg8_1
        del arg9_1
        # Source Nodes: [padded2, sort_2], Original ATen: [aten.sort, test_inductor_ops.jagged_to_padded_dense]
        buf13 = aten.sort.default(buf12, 1)
        buf14 = buf13[0]
        del buf13
        buf16 = buf12; del buf12  # reuse
        # Source Nodes: [unique_0hot_1], Original ATen: [aten.cat]
        triton_poi_fused_cat_1_xnumel = 1024 + (1024*s2)
        triton_poi_fused_cat_1.run(buf14, buf16, ps1, triton_poi_fused_cat_1_xnumel, grid=grid(triton_poi_fused_cat_1_xnumel), stream=stream0)
        # Source Nodes: [sort_3, unique_0hot_1], Original ATen: [aten.cat, aten.sort]
        buf17 = aten.sort.stable(buf16, stable=True, dim=1)
        buf19 = buf17[1]
        del buf17
        buf21 = buf8; del buf8  # reuse
        # Source Nodes: [mask], Original ATen: [aten.zeros_like]
        triton_poi_fused_zeros_like_2_xnumel = 1024 + (1024*s1)
        triton_poi_fused_zeros_like_2.run(buf21, triton_poi_fused_zeros_like_2_xnumel, grid=grid(triton_poi_fused_zeros_like_2_xnumel), stream=stream0)
        # Source Nodes: [adj_diff, diff, gt, mask, n_unique, setitem], Original ATen: [aten._to_copy, aten.gt, aten.index_put, aten.lift_fresh, aten.sub, aten.sum, aten.zeros_like]
        triton_red_fused__to_copy_gt_index_put_lift_fresh_sub_sum_zeros_like_3.run(buf6, buf21, s1, 1024, s1, grid=grid(1024), stream=stream0)
        buf23 = empty_strided_cuda((1024, 1 + s1), (1 + s1, 1), torch.int64)
        buf30 = empty_strided_cuda((1024, 2 + s1 + s2), (2 + s1 + s2, 1), torch.int64)
        buf24 = reinterpret_tensor(buf30, (1024, 1 + s1), (2 + s1 + s2, 1), 0)  # alias
        # Source Nodes: [cumsum, gather_out, setitem_1], Original ATen: [aten.cumsum, aten.gather, aten.index_put, aten.lift_fresh]
        triton_red_fused_cumsum_gather_index_put_lift_fresh_4_rnumel = 1 + s1
        triton_red_fused_cumsum_gather_index_put_lift_fresh_4.run(buf21, buf11, buf6, buf23, buf24, s1, s2, 1024, triton_red_fused_cumsum_gather_index_put_lift_fresh_4_rnumel, grid=grid(1024), stream=stream0)
        del buf21
        del buf6
        buf26 = buf16; del buf16  # reuse
        # Source Nodes: [mask_2], Original ATen: [aten.zeros_like]
        triton_poi_fused_zeros_like_2_xnumel = 1024 + (1024*s2)
        triton_poi_fused_zeros_like_2.run(buf26, triton_poi_fused_zeros_like_2_xnumel, grid=grid(triton_poi_fused_zeros_like_2_xnumel), stream=stream0)
        # Source Nodes: [adj_diff_1, diff_2, gt_1, mask_2, n_unique_1, setitem_3], Original ATen: [aten._to_copy, aten.gt, aten.index_put, aten.lift_fresh, aten.sub, aten.sum, aten.zeros_like]
        triton_red_fused__to_copy_gt_index_put_lift_fresh_sub_sum_zeros_like_3.run(buf14, buf26, s2, 1024, s2, grid=grid(1024), stream=stream0)
        buf28 = empty_strided_cuda((1024, 1 + s2), (1 + s2, 1), torch.int64)
        buf29 = reinterpret_tensor(buf30, (1024, 1 + s2), (2 + s1 + s2, 1), 1 + s1)  # alias
        # Source Nodes: [cumsum_1, gather_out_1, setitem_4], Original ATen: [aten.cumsum, aten.gather, aten.index_put, aten.lift_fresh]
        triton_red_fused_cumsum_gather_index_put_lift_fresh_5_rnumel = 1 + s2
        triton_red_fused_cumsum_gather_index_put_lift_fresh_5.run(buf26, buf19, buf14, buf28, buf29, s2, s1, 1024, triton_red_fused_cumsum_gather_index_put_lift_fresh_5_rnumel, grid=grid(1024), stream=stream0)
        del buf14
        del buf26
        del buf24
        del buf29
        # Source Nodes: [sort_4], Original ATen: [aten.sort]
        buf31 = aten.sort.default(buf30, 1)
        buf32 = buf31[0]
        buf33 = buf31[1]
        del buf31
        buf34 = empty_strided_cuda((1024, s1), (s1, 1), torch.int64)
        # Source Nodes: [setitem_2, unique_indices_1], Original ATen: [aten.index_put, aten.lift_fresh, aten.sub]
        triton_poi_fused_index_put_lift_fresh_sub_6_xnumel = 1024*s1
        triton_poi_fused_index_put_lift_fresh_sub_6.run(buf23, buf11, buf34, s1, triton_poi_fused_index_put_lift_fresh_sub_6_xnumel, grid=grid(triton_poi_fused_index_put_lift_fresh_sub_6_xnumel), stream=stream0)
        del buf11
        del buf23
        buf35 = empty_strided_cuda((1024, s2), (s2, 1), torch.int64)
        # Source Nodes: [setitem_5, unique_indices_4], Original ATen: [aten.index_put, aten.lift_fresh, aten.sub]
        triton_poi_fused_index_put_lift_fresh_sub_6_xnumel = 1024*s2
        triton_poi_fused_index_put_lift_fresh_sub_6.run(buf28, buf19, buf35, s2, triton_poi_fused_index_put_lift_fresh_sub_6_xnumel, grid=grid(triton_poi_fused_index_put_lift_fresh_sub_6_xnumel), stream=stream0)
        del buf19
        del buf28
        ps2 = 2 + s1 + s2
        buf36 = buf30; del buf30  # reuse
        # Source Nodes: [joined_padded_counts], Original ATen: [aten.cat]
        triton_poi_fused_cat_7_xnumel = 2048 + (1024*s1) + (1024*s2)
        triton_poi_fused_cat_7.run(buf34, buf35, buf36, ps2, ps0, s1, s2, ps1, triton_poi_fused_cat_7_xnumel, grid=grid(triton_poi_fused_cat_7_xnumel), stream=stream0)
        del buf34
        del buf35
        buf37 = empty_strided_cuda((1024, 2 + s1 + s2), (2 + s1 + s2, 1), torch.int64)
        # Source Nodes: [min_1], Original ATen: [aten.min]
        triton_poi_fused_min_8_xnumel = 2048 + (1024*s1) + (1024*s2)
        triton_poi_fused_min_8.run(buf33, buf36, buf37, ps2, s1, s2, triton_poi_fused_min_8_xnumel, grid=grid(triton_poi_fused_min_8_xnumel), stream=stream0)
        buf38 = empty_strided_cuda((1024, 1 + s1 + s2), (1 + s1 + s2, 1), torch.int8)
        buf45 = empty_strided_cuda((1024, 2), (2, 1), torch.int64)
        buf43 = reinterpret_tensor(buf45, (1024, 1), (2, 1), 0)  # alias
        buf44 = reinterpret_tensor(buf45, (1024, 1), (2, 1), 1)  # alias
        buf56 = empty_strided_cuda((1024, 2), (2, 1), torch.int64)
        buf54 = reinterpret_tensor(buf56, (1024, 1), (2, 1), 0)  # alias
        buf55 = reinterpret_tensor(buf56, (1024, 1), (2, 1), 1)  # alias
        # Source Nodes: [diff_4, eq, joined_padded_lengths, padded_diff, padding_mask, stack_5, stack_7, to_4], Original ATen: [aten._to_copy, aten.eq, aten.gt, aten.mul, aten.stack, aten.sub, aten.sum]
        triton_red_fused__to_copy_eq_gt_mul_stack_sub_sum_9_rnumel = 1 + s1 + s2
        triton_red_fused__to_copy_eq_gt_mul_stack_sub_sum_9.run(buf32, buf33, buf36, buf38, buf43, buf44, buf54, buf55, s1, s2, 1024, triton_red_fused__to_copy_eq_gt_mul_stack_sub_sum_9_rnumel, grid=grid(1024), stream=stream0)
        # Source Nodes: [diff_4, eq, padded_diff, padding_mask, sort_5, to_4], Original ATen: [aten._to_copy, aten.eq, aten.gt, aten.mul, aten.sort, aten.sub]
        buf39 = aten.sort.stable(buf38, stable=True, dim=1, descending=True)
        del buf38
        buf40 = buf39[0]
        buf41 = buf39[1]
        del buf39
        del buf43
        del buf44
        # Source Nodes: [mask_5], Original ATen: [aten.repeat_interleave]
        buf46 = aten.repeat_interleave.Tensor(reinterpret_tensor(buf45, (2048, ), (1, ), 0))
        u0 = buf46.size(0)
        del buf45
        buf47 = buf46
        del buf46
        buf48 = empty_strided_cuda((1024, 1 + s1 + s2), (1 + s1 + s2, 1), torch.int64)
        buf53 = empty_strided_cuda((1024, 1 + s1 + s2), (1 + s1 + s2, 1), torch.int64)
        buf66 = empty_strided_cuda((1024, ), (1, ), torch.int32)
        # Source Nodes: [count_pairwise_min, count_pairwise_min_1, gather_3, joined_padded_2, padding_mask, rowwise_counts, sum_4], Original ATen: [aten._to_copy, aten.gather, aten.gt, aten.mul, aten.sum]
        triton_red_fused__to_copy_gather_gt_mul_sum_10_rnumel = 1 + s1 + s2
        triton_red_fused__to_copy_gather_gt_mul_sum_10.run(buf41, buf32, buf37, buf33, buf36, buf40, buf48, buf53, buf66, s1, s2, 1024, triton_red_fused__to_copy_gather_gt_mul_sum_10_rnumel, grid=grid(1024), stream=stream0)
        del buf32
        del buf33
        del buf36
        del buf37
        del buf40
        buf49 = reinterpret_tensor(buf41, (1024 + (1024*s1) + (1024*s2), ), (1, ), 0); del buf41  # reuse
        # Source Nodes: [unique_val], Original ATen: [aten.index]
        triton_poi_fused_index_11_xnumel = 1024 + (1024*s1) + (1024*s2)
        triton_poi_fused_index_11.run(buf48, buf49, s1, s2, triton_poi_fused_index_11_xnumel, grid=grid(triton_poi_fused_index_11_xnumel), stream=stream0)
        del buf48
        buf50 = empty_strided_cuda((1024 + (1024*s1) + (1024*s2), ), (1, ), torch.bool)
        # Source Nodes: [mask_5], Original ATen: [aten.index_select]
        triton_poi_fused_index_select_12_xnumel = 1024 + (1024*s1) + (1024*s2)
        triton_poi_fused_index_select_12.run(buf47, buf50, triton_poi_fused_index_select_12_xnumel, grid=grid(triton_poi_fused_index_select_12_xnumel), stream=stream0)
        del buf47
        # Source Nodes: [unique_val], Original ATen: [aten.index]
        buf51 = aten.index.Tensor(buf49, [buf50])
        u1 = buf51.size(0)
        del buf49
        buf52 = buf51
        del buf54
        del buf55
        # Source Nodes: [mask_7], Original ATen: [aten.repeat_interleave]
        buf57 = aten.repeat_interleave.Tensor(reinterpret_tensor(buf56, (2048, ), (1, ), 0))
        u11 = buf57.size(0)
        del buf56
        buf58 = buf57
        del buf57
        buf59 = buf50; del buf50  # reuse
        # Source Nodes: [mask_7], Original ATen: [aten.index_select]
        triton_poi_fused_index_select_12_xnumel = 1024 + (1024*s1) + (1024*s2)
        triton_poi_fused_index_select_12.run(buf58, buf59, triton_poi_fused_index_select_12_xnumel, grid=grid(triton_poi_fused_index_select_12_xnumel), stream=stream0)
        del buf58
        # Source Nodes: [counts], Original ATen: [aten.index]
        buf60 = aten.index.Tensor(reinterpret_tensor(buf53, (1024 + (1024*s1) + (1024*s2), ), (1, ), 0), [buf59])
        u2 = buf60.size(0)
        del buf53
        del buf59
        buf61 = buf60
        del buf60
        # Source Nodes: [expanded_ids], Original ATen: [aten.repeat_interleave]
        buf62 = aten.repeat_interleave.Tensor(buf61)
        u3 = buf62.size(0)
        del buf61
        buf63 = buf62
        buf64 = buf63; del buf63  # reuse
        # Source Nodes: [expanded_ids], Original ATen: [aten.index_select]
        triton_poi_fused_index_select_13.run(buf64, buf52, u1, u3, grid=grid(u3), stream=stream0)
        del buf52
        buf69 = empty_strided_cuda((1025, ), (1, ), torch.int32)
        buf67 = reinterpret_tensor(buf69, (1024, ), (1, ), 1)  # alias
        # Source Nodes: [cumsum_4], Original ATen: [aten.cumsum]
        triton_per_fused_cumsum_14.run(buf66, buf67, 1, 1024, grid=grid(1), stream=stream0)
        buf68 = reinterpret_tensor(buf69, (1, ), (1, ), 0)  # alias
        # Source Nodes: [zeros_2], Original ATen: [aten.zeros]
        triton_poi_fused_zeros_15.run(buf68, 1, grid=grid(1), stream=stream0)
    return (buf64, buf69, buf66, )
def benchmark_compiled_module(times=10, repeat=10):
    from torch._dynamo.testing import rand_strided
    from torch._inductor.utils import print_performance
    arg0_1 = 96282
    arg1_1 = rand_strided((96282, ), (1, ), device='cuda:0', dtype=torch.int64)
    arg2_1 = 98
    arg3_1 = rand_strided((98, ), (1, ), device='cuda:0', dtype=torch.float32)
    arg4_1 = rand_strided((1025, ), (1, ), device='cuda:0', dtype=torch.int32)
    arg5_1 = 95
    arg6_1 = rand_strided((95, ), (1, ), device='cuda:0', dtype=torch.float32)
    arg7_1 = 94741
    arg8_1 = rand_strided((94741, ), (1, ), device='cuda:0', dtype=torch.int64)
    arg9_1 = rand_strided((1025, ), (1, ), device='cuda:0', dtype=torch.int32)
    fn = lambda: call([arg0_1, arg1_1, arg2_1, arg3_1, arg4_1, arg5_1, arg6_1, arg7_1, arg8_1, arg9_1])
    return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
    from torch._inductor.wrapper_benchmark import compiled_module_main
    compiled_module_main('None', benchmark_compiled_module)