#blocked = #triton_gpu.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [4], order = [0], CTAsPerCGA = [1], CTASplitNum = [1], CTAOrder = [0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0], CTAsPerCGA = [1], CTASplitNum = [1], CTAOrder = [0]}>
module attributes {"triton_gpu.compute-capability" = 89 : i32, "triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32, "triton_gpu.threads-per-warp" = 32 : i32} {
  tt.func public @triton__0d1d2de(%arg0: !tt.ptr<i64, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg2: i64 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
    %cst = arith.constant dense<50257> : tensor<512xi64, #blocked>
    %cst_0 = arith.constant dense<0> : tensor<512xi64, #blocked>
    %cst_1 = arith.constant dense<-1> : tensor<512xi64, #blocked>
    %cst_2 = arith.constant dense<-1.000000e+00> : tensor<512xf32, #blocked1>
    %c512_i64 = arith.constant 512 : i64
    %0 = tt.get_program_id x : i32
    %1 = arith.extsi %0 : i32 to i64
    %2 = arith.muli %1, %c512_i64 : i64
    %3 = tt.make_range {end = 512 : i32, start = 0 : i32} : tensor<512xi32, #blocked>
    %4 = arith.extsi %3 : tensor<512xi32, #blocked> to tensor<512xi64, #blocked>
    %5 = tt.splat %2 : (i64) -> tensor<512xi64, #blocked>
    %6 = arith.addi %5, %4 : tensor<512xi64, #blocked>
    %7 = tt.splat %arg0 : (!tt.ptr<i64, 1>) -> tensor<512x!tt.ptr<i64, 1>, #blocked>
    %8 = tt.addptr %7, %6 : tensor<512x!tt.ptr<i64, 1>, #blocked>, tensor<512xi64, #blocked>
    %9 = tt.load %8 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<512xi64, #blocked>
    %10 = arith.cmpi ne, %9, %cst_1 : tensor<512xi64, #blocked>
    %11 = arith.select %10, %9, %cst_0 : tensor<512xi1, #blocked>, tensor<512xi64, #blocked>
    %12 = arith.addi %11, %cst : tensor<512xi64, #blocked>
    %13 = arith.cmpi slt, %11, %cst_0 : tensor<512xi64, #blocked>
    %14 = arith.select %13, %12, %11 : tensor<512xi1, #blocked>, tensor<512xi64, #blocked>
    %15 = arith.cmpi sge, %14, %cst_0 : tensor<512xi64, #blocked>
    %16 = arith.cmpi slt, %14, %cst : tensor<512xi64, #blocked>
    %17 = arith.andi %15, %16 : tensor<512xi1, #blocked>
    %18 = triton_gpu.convert_layout %17 : (tensor<512xi1, #blocked>) -> tensor<512xi1, #blocked1>
    tt.assert %18, "index out of bounds: 0 <= tmp7 < 50257", "<frozen importlib._bootstrap_external>", "_call_with_frames_removed", 883 : tensor<512xi1, #blocked1>
    %19 = arith.muli %6, %cst : tensor<512xi64, #blocked>
    %20 = arith.addi %14, %19 : tensor<512xi64, #blocked>
    %21 = tt.splat %arg1 : (!tt.ptr<f32, 1>) -> tensor<512x!tt.ptr<f32, 1>, #blocked>
    %22 = tt.addptr %21, %20 : tensor<512x!tt.ptr<f32, 1>, #blocked>, tensor<512xi64, #blocked>
    %23 = triton_gpu.convert_layout %22 : (tensor<512x!tt.ptr<f32, 1>, #blocked>) -> tensor<512x!tt.ptr<f32, 1>, #blocked1>
    tt.store %23, %cst_2 {cache = 1 : i32, evict = 1 : i32} : tensor<512xf32, #blocked1>
    tt.return
  }
}