NVIDIA/cutlass
1
0
Fork 0
mirror of https://github.com/NVIDIA/cutlass.git synced 2026-05-11 17:00:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

releaase 2.11 (#703)

Browse Source
This commit is contained in:
Aditya Atluri
2022-11-19 06:02:15 -08:00
committed by GitHub
parent 3c90f6aea6
commit c975e2ccbb
329 changed files with 47332 additions and 10607 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/03_visualize_layout/CMakeLists.txt
View File

@@ -29,7 +29,6 @@
set(TEST_COMMAND_00 RowMajor --extent=16,16)
set(TEST_COMMAND_01 \"ColumnMajorInterleaved<4>\" --extent=32,8 --output-shape=16 --vectorize=4)
cutlass_example_add_executable(
03_visualize_layout
@@ -37,6 +36,5 @@ cutlass_example_add_executable(
register_layout.cu
TEST_COMMAND_OPTIONS
TEST_COMMAND_00
TEST_COMMAND_01
)

10
examples/03_visualize_layout/register_layout.cu
View File

@@ -64,15 +64,15 @@ void RegisterLayouts(std::map<std::string, std::unique_ptr<VisualizeLayoutBase>
// All Ampere/Turing H/Integer matrix multiply tensor core kernels uses the same swizzling
// layout implementation with different templates.
//
// BMMA 88128 Interleaved-256
// BMMA 168256 Interleaved-256
// mma.sync.aligned.m8n8k128.s32.b1.b1.s32 Interleaved-256
// mma.sync.aligned.m16n8k256.s32.b1.b1.s32 Interleaved-256
{"TensorOpMultiplicand<1,256>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<1, 256>>},
// BMMA 88128 TN kblock512
// BMMA 168256 TN kblock512
// mma.sync.aligned.m8n8k128.s32.b1.b1.s32 TN kblock512
// mma.sync.aligned.m16n8k256.s32.b1.b1.s32 TN kblock512
{"TensorOpMultiplicand<1,512>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<1, 512>>},
// BMMA 168256 TN kblock1024
// mma.sync.aligned.m16n8k256.s32.b1.b1.s32 TN kblock1024
{"TensorOpMultiplicand<1,1024>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<1, 1024>>},
// Integer matrix multiply.int4 8832 Interleaved-64