leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

v4.1 release

Browse Source
This commit is contained in:
Junkai-Wu
2025-07-03 20:07:53 +08:00
committed by GitHub
parent b995f93317
commit a1aaf2300a
155 changed files with 18407 additions and 6068 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/python/CuTeDSL/notebooks/cute_layout_algebra.ipynb
View File

@@ -297,7 +297,7 @@
" assert depth <= 1, f\"Depth of coalesced layout should be <= 1, got {depth}\"\n",
"\n",
" print(\">>> 3. Checking layout functionality remains the same after the coalesce operation:\")\n",
" for i in range(original_size):\n",
" for i in cutlass.range_constexpr(original_size):\n",
" original_value = layout(i)\n",
" coalesced_value = result(i)\n",
" print(f\"Index {i}: original {original_value}, coalesced {coalesced_value}\")\n",

43
examples/python/CuTeDSL/notebooks/data_types.ipynb
View File

@@ -60,48 +60,7 @@
"@cute.jit\n",
"def foo(a: cutlass.Int32): # annotate `a` as 32-bit integer passed to jit function via ABI\n",
" ...\n",
"```\n",
"To differentiate between compile-time and runtime values, CuTe DSL introduces primitive types that \n",
"represent dynamic values in JIT-compiled code.\n",
"\n",
"CuTe DSL provides a comprehensive set of primitive numeric types for representing dynamic values at \n",
"runtime. These types are formally defined within the CuTe DSL typing system:\n",
"\n",
"### Integer Types\n",
"- `Int8` - 8-bit signed integer\n",
"- `Int16` - 16-bit signed integer \n",
"- `Int32` - 32-bit signed integer\n",
"- `Int64` - 64-bit signed integer\n",
"- `Int128` - 128-bit signed integer\n",
"- `Uint8` - 8-bit unsigned integer\n",
"- `Uint16` - 16-bit unsigned integer\n",
"- `Uint32` - 32-bit unsigned integer\n",
"- `Uint64` - 64-bit unsigned integer\n",
"- `Uint128` - 128-bit unsigned integer\n",
"\n",
"### Floating Point Types\n",
"- `Float16` - 16-bit floating point\n",
"- `Float32` - 32-bit floating point \n",
"- `Float64` - 64-bit floating point\n",
"- `BFloat16` - Brain Floating Point format (16-bit)\n",
"- `TFloat32` - Tensor Float32 format (reduced precision format used in tensor operations)\n",
"- `Float8E4M3` - 8-bit floating point with 4-bit exponent and 3-bit mantissa\n",
"- `Float8E5M2` - 8-bit floating point with 5-bit exponent and 2-bit mantissa\n",
"\n",
"These specialized types are designed to represent dynamic values in CuTe DSL code that will be \n",
"evaluated at runtime, in contrast to Python's built-in numeric types which are evaluated during \n",
"compilation.\n",
"\n",
"### Example usage:\n",
"\n",
"```python\n",
"x = cutlass.Int32(5) # Creates a 32-bit integer\n",
"y = cutlass.Float32(3.14) # Creates a 32-bit float\n",
"\n",
"@cute.jit\n",
"def foo(a: cutlass.Int32): # annotate `a` as 32-bit integer passed to jit function via ABI\n",
" ...\n",
"```"
"```\n"
]
},
{

2
examples/python/CuTeDSL/notebooks/tensorssa.ipynb
View File

@@ -120,7 +120,7 @@
" src_vec = src.load()\n",
" dst_vec = src_vec[indices]\n",
" print(f\"{src_vec} -> {dst_vec}\")\n",
" if isinstance(dst_vec, cute.TensorSSA):\n",
" if cutlass.const_expr(isinstance(dst_vec, cute.TensorSSA)):\n",
" dst.store(dst_vec)\n",
" cute.print_tensor(dst)\n",
" else:\n",