leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

CUTLASS 3.6.0 (#1850)

Browse Source 
* v3.6

* update changelog

* update readme

* fix typo

* fixing typos

* hopper gemm with weight prefetch

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2024-10-09 15:33:27 -04:00
committed by GitHub
co-authored by yuzhai Haicheng Wu
parent 0837a2a00a
commit cc3c29a81a
354 changed files with 105937 additions and 8197 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/gemm/warp/gemm_sm80.cu
-209
View File
@@ -1357,215 +1357,6 @@ TEST(SM80_warp_gemm_tensor_op_crosswise_i4, 128x128x256_16x16x256_16x8x64) {
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x512_64x64x512_16x8x256) {
using Shape = cutlass::gemm::GemmShape<64, 64, 512>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 512> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x512_64x32x512_16x8x256) {
using Shape = cutlass::gemm::GemmShape<64, 32, 512>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 512> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x512_32x32x512_16x8x256) {
using Shape = cutlass::gemm::GemmShape<32, 32, 512>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 512> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x512_32x16x512_16x8x256) {
using Shape = cutlass::gemm::GemmShape<32, 16, 512>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 512> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x512_16x16x512_16x8x256) {
using Shape = cutlass::gemm::GemmShape<16, 16, 512>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 512>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 512> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x1024_64x64x1024_16x8x256) {
using Shape = cutlass::gemm::GemmShape<64, 64, 1024>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 1024> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x1024_64x32x1024_16x8x256) {
using Shape = cutlass::gemm::GemmShape<64, 32, 1024>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 1024> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x1024_32x32x1024_16x8x256) {
using Shape = cutlass::gemm::GemmShape<32, 32, 1024>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 1024> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x1024_32x16x1024_16x8x256) {
using Shape = cutlass::gemm::GemmShape<32, 16, 1024>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 1024> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_crosswise_b1, 128x128x1024_16x16x1024_16x8x256) {
using Shape = cutlass::gemm::GemmShape<16, 16, 1024>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 256>;
using Element = cutlass::uint1b_t;
using ElementC = int;
using LayoutA = cutlass::layout::RowMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using LayoutB = cutlass::layout::ColumnMajorTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<Element>::value, 1024>;
using MmaTensorOp = typename cutlass::gemm::warp::DefaultMmaTensorOp<
Shape, InstructionShape, Element, LayoutA, Element, LayoutB, ElementC,
cutlass::layout::RowMajor, cutlass::arch::OpMultiplyAdd>::Type;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<128, 128, 1024> >()
.run();
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_tensor_op_congruous_f64, 16x16x4_16x16x4_8x8x4) {
using Shape = cutlass::gemm::GemmShape<16, 16, 4>;