import triton
import triton.language as tl

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, DeviceProperties
triton_helpers.set_driver_to_gpu()

@triton_heuristics.pointwise(
    size_hints={'x': 2048}, 
    filename=__file__,
    triton_meta={'signature': {'in_ptr0': '*bf16', 'in_ptr1': '*bf16', 'in_ptr2': '*bf16', 'out_ptr0': '*bf16', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=48, cc=121, major=12, regs_per_multiprocessor=65536, max_threads_per_multi_processor=1536, max_threads_per_block=1024, warp_size=32), 'constants': {}, 'native_matmul': False, 'enable_fp_fusion': True, 'launch_pdl': False, 'disable_ftz': False, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
    inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__scaled_dot_product_flash_attention__to_copy__unsafe_view_add_arange_cat_clone_index_mul_neg_split_transpose_view_2', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'atomic_add_found': False, 'num_load': 5, 'num_store': 1, 'num_reduction': 0, 'backend_hash': 'BE9F3F68E84A48F2C239366BB10BBF3F0E1498DBECF7F747524E7B7961E82579', 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False, 'deterministic': False, 'force_filter_reduction_configs': False, 'mix_order_reduction_allow_multi_stages': False, 'are_deterministic_algorithms_enabled': False, 'tiling_scores': {'x': 15360}},
    min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__scaled_dot_product_flash_attention__to_copy__unsafe_view_add_arange_cat_clone_index_mul_neg_split_transpose_view_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
    xnumel = 1280
    xoffset = tl.program_id(0) * XBLOCK
    xindex = xoffset + tl.arange(0, XBLOCK)[:]
    xmask = xindex < xnumel
    x3 = xindex
    x0 = (xindex % 128)
    x2 = xindex // 256
    x4 = xindex // 128
    x1 = ((xindex // 128) % 2)
    tmp0 = tl.load(in_ptr0 + (x3), xmask).to(tl.float32)
    tmp2 = tl.load(in_ptr1 + (x0 + 128*x2), xmask, eviction_policy='evict_last').to(tl.float32)
    tmp23 = tl.load(in_ptr2 + (x0 + 128*x2), xmask, eviction_policy='evict_last').to(tl.float32)
    tmp1 = tmp0.to(tl.float32)
    tmp3 = tmp2.to(tl.float32)
    tmp4 = tmp1 * tmp3
    tmp5 = x0
    tmp6 = tl.full([1], 0, tl.int64)
    tmp7 = tmp5 >= tmp6
    tmp8 = tl.full([1], 64, tl.int64)
    tmp9 = tmp5 < tmp8
    tmp10 = tl.load(in_ptr0 + (64 + 128*x4 + (x0)), tmp9 & xmask, eviction_policy='evict_last', other=0.0).to(tl.float32)
    tmp11 = tmp10.to(tl.float32)
    tmp12 = -tmp11
    tmp13 = tl.full(tmp12.shape, 0.0, tmp12.dtype)
    tmp14 = tl.where(tmp9, tmp12, tmp13)
    tmp15 = tmp5 >= tmp8
    tmp16 = tl.full([1], 128, tl.int64)
    tmp17 = tmp5 < tmp16
    tmp18 = tl.load(in_ptr0 + (128*x4 + ((-64) + x0)), tmp15 & xmask, eviction_policy='evict_last', other=0.0).to(tl.float32)
    tmp19 = tmp18.to(tl.float32)
    tmp20 = tl.full(tmp19.shape, 0.0, tmp19.dtype)
    tmp21 = tl.where(tmp15, tmp19, tmp20)
    tmp22 = tl.where(tmp9, tmp14, tmp21)
    tmp24 = tmp23.to(tl.float32)
    tmp25 = tmp22 * tmp24
    tmp26 = tmp4 + tmp25
    tmp27 = tmp26.to(tl.float32)
    tl.store(out_ptr0 + (x0 + 128*x2 + 640*x1), tmp27, xmask)