Code
Hub
Workspaces
Following
Trending
Connect
MCP
copy
Create free account
hub
/
github.com/tile-ai/tilelang
/ types & classes
Types & classes
525 in github.com/tile-ai/tilelang
⨍
Functions
7,254
◇
Types & classes
525
↳
Endpoints
224
↓ 345 callers
Class
int
tilelang/language/dtypes.py:353
↓ 181 callers
Class
dtype
tilelang/language/dtypes.py:14
↓ 152 callers
Class
float
tilelang/language/dtypes.py:357
↓ 47 callers
Class
Layout
tilelang/layout/layout.py:12
↓ 26 callers
Class
bool
tilelang/language/dtypes.py:351
↓ 21 callers
Class
EnvVar
Descriptor for managing access to a single environment variable. Purpose ------- In many projects, access to environment variables i
tilelang/env.py:203
↓ 20 callers
Class
Fragment
A Fragment layout object that encapsulates iteration variables (forward_vars), thread iteration variables (forward_thread), and index transfo
tilelang/layout/fragment.py:12
↓ 11 callers
Class
BitLinear
examples/bitnet-1.58b/utils_quant.py:213
↓ 10 callers
Class
Analyzer
A class to analyze the performance of a TVM IR module. It calculates metrics such as FLOPs, memory bandwidth, and estimated execution time.
tilelang/tools/Analyzer.py:34
↓ 10 callers
Class
CachedTextSource
tilelang/jit/adapter/base.py:16
↓ 9 callers
Class
JITNoBuilderError
Exception raised when JIT-related operations require a Builder but none exists. In eager mode, TileLang constructs AST directly without an e
tilelang/jit/exceptions.py:4
↓ 8 callers
Class
KernelCache
Caches compiled kernels using a class and database persistence to avoid redundant compilation. Cache files: kernel.cu: The compiled k
tilelang/cache/kernel_cache.py:32
↓ 8 callers
Class
Linear
Custom linear layer with support for quantized weights and optional bias. Args: in_features (int): Number of input features.
examples/deepseek_v32/inference/model.py:171
↓ 7 callers
Class
BitnetRMSNorm
examples/bitnet-1.58b/modeling_bitnet.py:88
↓ 7 callers
Class
CDNA
tilelang/carver/arch/cdna.py:24
↓ 7 callers
Class
CUDA
tilelang/carver/arch/cuda.py:124
↓ 7 callers
Class
IntSetChecker
testing/python/arith/test_arith_intset.py:24
↓ 5 callers
Class
ColumnParallelLinear
Linear layer with column parallelism, splitting output features across distributed processes. Args: in_features (int): Number of inp
examples/deepseek_v32/inference/model.py:215
↓ 5 callers
Class
GeneralRemove
tilelang/autodd.py:101
↓ 5 callers
Class
JITKernel
A wrapper class for compiling and invoking TileLang (TVM TIR) functions as PyTorch-compatible functions. Attributes ---------- artif
tilelang/jit/kernel.py:36
↓ 5 callers
Class
PassPipeline
Lowering pass pipeline for a specific backend. Each backend should register its own Pipeline so that the compiler can resolve the correct pas
tilelang/backend/pass_pipeline/pipeline.py:11
↓ 5 callers
Class
PrimFuncNode
tilelang/carver/roller/node.py:190
↓ 5 callers
Class
RMSNorm
Root Mean Square Layer Normalization (RMSNorm). Args: dim (int): Dimension of the input tensor. eps (float): Epsilon value f
examples/deepseek_v32/inference/model.py:286
↓ 5 callers
Class
Stride
Manages stride information for a given axis of a tensor.
tilelang/carver/roller/hint.py:29
↓ 5 callers
Class
TensorCoreIntrinEmitterSM75
SM75/Turing-specific MMA shape policy.
tilelang/cuda/intrinsics/macro/mma_sm75_macro_generator.py:9
↓ 5 callers
Class
dim3
Three-dimensional CUDA index tuple.
tilelang/contrib/cutedsl/threadblock_swizzle.py:20
↓ 4 callers
Class
AutotuneResult
Results from auto-tuning process. Attributes: latency: Best achieved execution latency. config: Configuration that produced the b
tilelang/autotuner/param.py:144
↓ 4 callers
Class
Edge
tilelang/carver/roller/node.py:88
↓ 4 callers
Class
Gemm_SM80
tilelang/contrib/cutedsl/gemm_v1.py:231
↓ 4 callers
Class
MatmulTemplate
A template for matrix multiplication (MatMul). This class defines the computation for a matrix-matrix multiplication with configurable p
tilelang/carver/template/matmul.py:10
↓ 4 callers
Class
NoRasterization
tilelang/carver/roller/rasterization.py:19
↓ 4 callers
Class
OutputNode
tilelang/carver/roller/node.py:501
↓ 4 callers
Class
TensorCoreIntrinEmitter
To eliminate Python syntax within TIR Macro.
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:36
↓ 4 callers
Class
TensorInstruction
tilelang/carver/arch/cuda.py:113
↓ 4 callers
Class
WGSparseTensorCoreIntrinEmitter
tilelang/cuda/intrinsics/macro/wgmma_sp_macro_generator.py:31
↓ 3 callers
Class
BitnetModel
Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`BitnetDecoderLayer`] Args: config: BitnetConf
examples/bitnet-1.58b/modeling_bitnet.py:956
↓ 3 callers
Class
Block
tilelang/carver/roller/bestfit.py:4
↓ 3 callers
Class
Builder
tilelang/language/eager/builder.py:201
↓ 3 callers
Class
CompiledArtifact
Represents a compiled kernel artifact containing both host and device code. Stores all necessary components for kernel execution in the TVM r
tilelang/engine/param.py:154
↓ 3 callers
Class
IntrinInfo
The information of tensorcore intrinsic related information
tilelang/carver/roller/hint.py:107
↓ 3 callers
Class
MatrixCoreIntrinEmitter
To eliminate Python syntax within TIR Macro.
tilelang/rocm/intrinsics/mfma_macro_generator.py:44
↓ 3 callers
Class
OutTensor
tilelang/language/eager/builder.py:143
↓ 3 callers
Class
PostProcCounter
Track postproc callback invocations with a simple counter.
testing/python/cache/test_tilelang_kernel_cache.py:57
↓ 3 callers
Class
TCGEN05DescriptorParams
Pre-computed parameters for TCGEN05 descriptor initialization and atom offset computation. Returned by ``compute_tcgen05_*_desc_params()`` and co
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:26
↓ 3 callers
Class
Task
tilelang/autodd.py:544
↓ 3 callers
Class
TensorCoreIntrinEmitter
Intrinsic emitter for Blackwell (SM100) TCGEN5MMA instructions. Generates TIR macros that lower to ``tcgen05.mma`` PTX instructions for both
tilelang/cuda/intrinsics/macro/tcgen05_macro_generator.py:105
↓ 3 callers
Class
_BenchmarkWorkerState
tilelang/autotuner/tuner.py:218
↓ 2 callers
Class
BinOpFwdArg
tilelang/autodd.py:185
↓ 2 callers
Class
BitnetRotaryEmbedding
examples/bitnet-1.58b/modeling_bitnet.py:108
↓ 2 callers
Class
BoolOpFrame
tilelang/language/eager/builder.py:125
↓ 2 callers
Class
CompileArgs
Compile arguments for the auto-tuner. Detailed description can be found in `tilelang.jit.compile`. Attributes: out_idx: List of output ten
tilelang/autotuner/param.py:44
↓ 2 callers
Class
CuTeDSLLibraryGenerator
tilelang/jit/adapter/cutedsl/libgen.py:19
↓ 2 callers
Class
Expert
examples/fusedmoe/example_fusedmoe_tilelang.py:228
↓ 2 callers
Class
ExpertTorch
examples/fusedmoe/example_fusedmoe_torch.py:8
↓ 2 callers
Class
Gemm_SM90
tilelang/contrib/cutedsl/gemm_v1.py:390
↓ 2 callers
Class
Hint
Central configuration class for managing various parameters of computational tasks.
tilelang/carver/roller/hint.py:154
↓ 2 callers
Class
KernelResourceUsage
Resource counts as reported by clang's kernel-resource-usage pass. Field names mirror the remark labels (lower-cased, normalized) so we can e
tilelang/contrib/hip_resource_info.py:35
↓ 2 callers
Class
LibraryGenerator
tilelang/jit/adapter/libgen.py:30
↓ 2 callers
Class
MLP
Multi-Layer Perceptron (MLP) used as a feed-forward layer. Attributes: w1 (nn.Module): Linear layer for input-to-hidden transformati
examples/deepseek_v32/inference/model.py:647
↓ 2 callers
Class
MatmulConfig
Configuration class - increases code complexity but is actually useless
examples/autodd/tilelang_buggy.py:55
↓ 2 callers
Class
MatrixCorePreshuffleIntrinEmitter
tilelang/rocm/intrinsics/mfma_macro_generator.py:796
↓ 2 callers
Class
MoE
examples/fusedmoe/example_fusedmoe_tilelang.py:264
↓ 2 callers
Class
ModelArgs
Data class for defining model arguments and hyperparameters. Attributes: max_batch_size (int): Maximum batch size. max_seq_l
examples/deepseek_v32/inference/model.py:19
↓ 2 callers
Class
NVRTCKernelCache
tilelang/jit/adapter/nvrtc/kernel_cache.py:7
↓ 2 callers
Class
NVRTCLibraryGenerator
Runtime compiler and loader for NVRTC-compiled CUDA kernels. Lifecycle: 1. compile_lib(): CUDA source → cubin + Python launcher 2
tilelang/jit/adapter/nvrtc/libgen.py:42
↓ 2 callers
Class
ProfileArgs
Profile arguments for the auto-tuner. Attributes: warmup: Number of warmup iterations. rep: Number of repetitions for timing.
tilelang/autotuner/param.py:90
↓ 2 callers
Class
RowParallelLinear
Linear layer with row parallelism, splitting input features across distributed processes. Args: in_features (int): Total number of i
examples/deepseek_v32/inference/model.py:245
↓ 2 callers
Class
SparseFlashAttn
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_indice.py:159
↓ 2 callers
Class
SparseFlashAttn
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_paged.py:170
↓ 2 callers
Class
SparseFlashAttn
examples/blocksparse_attention/example_tilelang_sparse_gqa_decode_varlen_mask.py:153
↓ 2 callers
Class
SparseTensorCoreIntrinEmitter
To eliminate Python syntax within TIR Macro.
tilelang/cuda/intrinsics/macro/mma_sp_macro_generator.py:43
↓ 2 callers
Class
TLCuTeDSLSourceWrapper
Wrapper class for TileLang CuTe DSL backend with C++ launcher. Generates optimized C++ launcher code that: - Loads cubin via CUDA Driver API
tilelang/jit/adapter/cutedsl/wrapper.py:710
↓ 2 callers
Class
TLPyWrapper
tilelang/jit/adapter/wrapper.py:989
↓ 2 callers
Class
TVMFFIKernelAdapter
Adapter that runs a TVM runtime.Executable with Torch tensors. Notes - We capture the "current" PyTorch CUDA stream/device as thunks (callabl
tilelang/jit/adapter/tvm_ffi.py:39
↓ 2 callers
Class
TensorCoreIntrinEmitter
To eliminate Python syntax within TIR Macro.
tilelang/cuda/intrinsics/macro/mma_sm70_macro_generator.py:26
↓ 2 callers
Class
TensorCoreIntrinEmitter
To eliminate Python syntax within TIR Macro.
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:93
↓ 2 callers
Class
TensorCoreIntrinEmitterWithLadderTransform
To eliminate Python syntax within TIR Macro. With Ladder Transform Plugin.
tilelang/cuda/intrinsics/macro/mma_macro_generator.py:897
↓ 2 callers
Class
TimeoutException
tilelang/autotuner/tuner.py:43
↓ 2 callers
Class
TransformKind
tilelang/common/transform_kind.py:5
↓ 2 callers
Class
Transformer
Transformer model with positional embeddings, multiple layers, and output projection. Attributes: max_seq_len (int): Maximum sequenc
examples/deepseek_v32/inference/model.py:901
↓ 2 callers
Class
VllmRunner
examples/bitnet-1.58b/vllm_workspace/conftest.py:411
↓ 2 callers
Class
WGMMADescriptorParams
Pre-computed parameters for WGMMA descriptor initialization and atom offset computation. Returned by ``compute_wgmma_*_desc_params()`` and consum
tilelang/cuda/intrinsics/macro/wgmma_macro_generator.py:29
↓ 2 callers
Class
_FakeKernel
testing/python/cache/test_tilelang_kernel_cache_atomic_save.py:23
↓ 2 callers
Class
_RDNATuningConfig
Internal Roller tuning heuristics for RDNA WMMA targets. Keep this deliberately small and table-driven so exact mcpu overrides stay localized
tilelang/carver/arch/rdna.py:17
↓ 2 callers
Class
_TestBuilder
testing/python/language/test_tilelang_language_eager_ast_signature.py:4
↓ 1 callers
Class
AccessReplacer
Mutator to replace memory BufferStores/BufferLoads with cast buffer accesses. Matches by both buffer and indices (structural equality) so that ac
tilelang/transform/decouple_type_cast.py:556
↓ 1 callers
Class
AllReduceInstance
tilelang/contrib/cutedsl/reduce.py:540
↓ 1 callers
Class
AnalysisResult
A data class to store the results of the analysis. Attributes: total_flops: Total floating-point operations. total_global_byt
tilelang/tools/Analyzer.py:16
↓ 1 callers
Class
Args
tilelang/autodd.py:1248
↓ 1 callers
Class
AsyncPythonRunner
tilelang/autodd.py:771
↓ 1 callers
Class
AttachFullFuncArgs
tilelang/autodd.py:143
↓ 1 callers
Class
AutoTuneImpl
tilelang/autotuner/tuner.py:1212
↓ 1 callers
Class
AutoTuner
Auto-tuner for tilelang programs. This class handles the auto-tuning process by testing different configurations and finding the optimal para
tilelang/autotuner/tuner.py:224
↓ 1 callers
Class
AutotuneInputsCapture
tilelang/autotuner/capture.py:87
↓ 1 callers
Class
BestFit
tilelang/carver/roller/bestfit.py:22
↓ 1 callers
Class
BitnetDecoderLayer
examples/bitnet-1.58b/modeling_bitnet.py:740
↓ 1 callers
Class
BitnetMLP
examples/bitnet-1.58b/modeling_bitnet.py:194
↓ 1 callers
Class
Block
Transformer block combining attention and feed-forward layers. Attributes: attn (nn.Module): Attention layer (MLA). ffn (nn.
examples/deepseek_v32/inference/model.py:852
↓ 1 callers
Class
BlockAnalyzer
tilelang/carver/roller/node.py:35
↓ 1 callers
Class
BlockInfo
Information about a TIR block.
tilelang/carver/analysis.py:47
next →
1–100 of 525, ranked by callers