Contents Menu Expand Light mode Dark mode Auto light/dark, in light mode Auto light/dark, in dark mode Skip to content
Triton Ascend v3.2.2 文档
Triton Ascend v3.2.2 文档

快速开始

  • 快速入门
  • 安装指南

特性说明

  • 架构设计与核心特性

Triton-Ascend算子开发指南

  • Triton-Ascend算子开发

Triton-Ascend算子迁移指南

  • 昇腾与GPU的开发差异
  • GPU Triton算子迁移

典型算子样例

  • Tutorials
    • 向量相加 (Vector Addition)
    • 融合 Softmax (Fused Softmax)
    • 层标准化 (Layer Normalization)
    • 融合注意力(Fused Attention)
    • 矩阵乘法 (Matrix Multiplication)
    • 自动调优 (Autotune)
    • 精度比对 (Accuracy Comparison)
    • max_autotune 使用示例
    • Costmodel 端到端示例

Triton-Ascend算子调试与调优

  • Triton-Ascend调试指南
  • Triton-Ascend性能分析方法

Triton API 接口说明

  • triton
    • triton.jit
    • triton.autotune
    • triton.heuristics
    • triton.Config
  • triton.language
    • triton.language.tensor
    • triton.language.program_id
    • triton.language.num_programs
    • triton.language.arange
    • triton.language.cat
    • triton.language.full
    • triton.language.zeros
    • triton.language.zeros_like
    • triton.language.cast
    • triton.language.broadcast
    • triton.language.broadcast_to
    • triton.language.expand_dims
    • triton.language.interleave
    • triton.language.join
    • triton.language.permute
    • triton.language.ravel
    • triton.language.reshape
    • triton.language.split
    • triton.language.trans
    • triton.language.view
    • triton.language.dot
    • triton.language.dot_scaled
    • triton.language.load
    • triton.language.store
    • triton.language.make_tensor_descriptor
    • triton.language.load_tensor_descriptor
    • triton.language.store_tensor_descriptor
    • triton.language.make_block_ptr
    • triton.language.advance
    • triton.language.flip
    • triton.language.where
    • triton.language.swizzle2d
    • triton.language.gather
    • triton.language.abs
    • triton.language.cdiv
    • triton.language.ceil
    • triton.language.clamp
    • triton.language.cos
    • triton.language.div_rn
    • triton.language.erf
    • triton.language.exp
    • triton.language.exp2
    • triton.language.fdiv
    • triton.language.floor
    • triton.language.fma
    • triton.language.log
    • triton.language.log2
    • triton.language.maximum
    • triton.language.minimum
    • triton.language.rsqrt
    • triton.language.sigmoid
    • triton.language.sin
    • triton.language.softmax
    • triton.language.sqrt
    • triton.language.sqrt_rn
    • triton.language.umulhi
    • triton.language.argmax
    • triton.language.argmin
    • triton.language.max
    • triton.language.min
    • triton.language.reduce
    • triton.language.sum
    • triton.language.xor_sum
    • triton.language.associative_scan
    • triton.language.cumprod
    • triton.language.cumsum
    • triton.language.histogram
    • triton.language.sort
    • triton.language.topk
    • triton.language.atomic_add
    • triton.language.atomic_and
    • triton.language.atomic_cas
    • triton.language.atomic_max
    • triton.language.atomic_min
    • triton.language.atomic_or
    • triton.language.atomic_xchg
    • triton.language.atomic_xor
    • triton.language.randint4x
    • triton.language.randint
    • triton.language.rand
    • triton.language.randn
    • triton.language.range
    • triton.language.static_range
    • triton.language.assume
    • triton.language.debug_barrier
    • triton.language.max_constancy
    • triton.language.max_contiguous
    • triton.language.multiple_of
    • triton.language.static_print
    • triton.language.static_assert
    • triton.language.device_print
    • triton.language.device_assert
    • triton.language.inline_asm_elementwise
  • triton.testing
    • triton.testing.Benchmark
    • triton.testing.do_bench
    • triton.testing.do_bench_cudagraph
    • triton.testing.perf_report
    • triton.testing.assert_close

昇腾扩展 API 接口说明

  • triton.language.extra.cann.extension
    • triton.language.extra.cann.extension.scope
    • triton.language.extra.cann.extension.ascend_address_space
    • triton.language.extra.cann.extension.compile_hint
    • triton.language.extra.cann.extension.multibuffer
    • triton.language.extra.cann.extension.parallel
    • triton.language.extra.cann.extension.copy
    • triton.language.extra.cann.extension.copy_from_ub_to_l1
    • triton.language.extra.cann.extension.fixpipe
    • triton.language.extra.cann.extension.sync_block_all
    • triton.language.extra.cann.extension.sync_block_set
    • triton.language.extra.cann.extension.sync_block_wait
    • triton.language.extra.cann.extension.debug_barrier
    • triton.language.extra.cann.extension.sub_vec_id
    • triton.language.extra.cann.extension.sub_vec_num
    • triton.language.extra.cann.extension.PIPE
    • triton.language.extra.cann.extension.MODE
    • triton.language.extra.cann.extension.CORE
    • triton.language.extra.cann.extension.SYNC_IN_VF
    • triton.language.extra.cann.extension.IteratorType
    • triton.language.extra.cann.extension.FixpipeDMAMode
    • triton.language.extra.cann.extension.FixpipeDualDstMode
    • triton.language.extra.cann.extension.FixpipePreQuantMode
    • triton.language.extra.cann.extension.FixpipePreReluMode
    • triton.language.extra.cann.extension.extract_slice
    • triton.language.extra.cann.extension.insert_slice
    • triton.language.extra.cann.extension.get_element
    • triton.language.extra.cann.extension.sort
    • triton.language.extra.cann.extension.flip
    • triton.language.extra.cann.extension.cast
    • triton.language.extra.cann.extension.index_put
    • triton.language.extra.cann.extension.gather_out_to_ub
    • triton.language.extra.cann.extension.scatter_ub_to_out
    • triton.language.extra.cann.extension.index_select_simd
    • triton.language.extra.cann.extension.custom
    • triton.language.extra.cann.extension.custom_semantic
    • triton.language.extra.cann.extension.register_custom_op
    • triton.language.extra.cann.extension.affine_expr
    • triton.language.extra.cann.extension.affine_constant_expr
    • triton.language.extra.cann.extension.affine_dim_expr
    • triton.language.extra.cann.extension.affine_symbol_expr
    • triton.language.extra.cann.extension.affine_binary_op_expr
    • triton.language.extra.cann.extension.affine_map
  • triton.language.extra.cann.libdevice
    • triton.language.extra.cann.libdevice.acos
    • triton.language.extra.cann.libdevice.acosh
    • triton.language.extra.cann.libdevice.asin
    • triton.language.extra.cann.libdevice.asinh
    • triton.language.extra.cann.libdevice.atan
    • triton.language.extra.cann.libdevice.atan2
    • triton.language.extra.cann.libdevice.atanh
    • triton.language.extra.cann.libdevice.copysign
    • triton.language.extra.cann.libdevice.cosh
    • triton.language.extra.cann.libdevice.cyl_bessel_i0
    • triton.language.extra.cann.libdevice.div_rz
    • triton.language.extra.cann.libdevice.erfinv
    • triton.language.extra.cann.libdevice.expm1
    • triton.language.extra.cann.libdevice.fast_dividef
    • triton.language.extra.cann.libdevice.fast_expf
    • triton.language.extra.cann.libdevice.float_as_int
    • triton.language.extra.cann.libdevice.fmod
    • triton.language.extra.cann.libdevice.gamma
    • triton.language.extra.cann.libdevice.hypot
    • triton.language.extra.cann.libdevice.ilogb
    • triton.language.extra.cann.libdevice.isinf
    • triton.language.extra.cann.libdevice.isnan
    • triton.language.extra.cann.libdevice.ldexp
    • triton.language.extra.cann.libdevice.lgamma
    • triton.language.extra.cann.libdevice.log10
    • triton.language.extra.cann.libdevice.log1p
    • triton.language.extra.cann.libdevice.nearbyint
    • triton.language.extra.cann.libdevice.nextafter
    • triton.language.extra.cann.libdevice.pow
    • triton.language.extra.cann.libdevice.reciprocal
    • triton.language.extra.cann.libdevice.relu
    • triton.language.extra.cann.libdevice.rint
    • triton.language.extra.cann.libdevice.round
    • triton.language.extra.cann.libdevice.signbit
    • triton.language.extra.cann.libdevice.sinh
    • triton.language.extra.cann.libdevice.tan
    • triton.language.extra.cann.libdevice.tanh
    • triton.language.extra.cann.libdevice.trunc
  • triton.extension.buffer.language
    • triton.extension.buffer.language.buffer
    • triton.extension.buffer.language.address_space
    • triton.extension.buffer.language.alloc
    • triton.extension.buffer.language.to_buffer
    • triton.extension.buffer.language.to_tensor
    • triton.extension.buffer.language.subview

环境变量

  • 环境变量

常见问题

  • Triton-Ascend FAQ
Back to top
Copyright © 2026, Huawei
Last updated on 9月 02, 2026
GitHub