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

Improved formatting, clarity, and content of several documents. (#64)

Browse Source 
* Improved formatting, clarity, and content of several documents.
This commit is contained in:
Andrew Kerr
2019-11-20 10:42:15 -08:00
committed by GitHub
parent f4d9c8f755
commit 8aca98f9a7
8 changed files with 118 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

81
media/docs/code_organization.md
View File

@@ -82,35 +82,44 @@ scripts to span the design space.
```
tools/
library/ # static/dynamic library containing all kernel instantiations of interest
# (with some build-level filter switches to compile specific subsets, perhaps by architecture)
# (with some build-level filter switches to compile specific subsets)
include/
cutlass/
library/ # header files for CUTLASS Deliverables Library (in cutlass::library:: namespace)
library/ # header files for CUTLASS Deliverables Library (in cutlass::library:: namespace)
library.h # defines enums and structs to describe the tiled structure of operator instances
manifest.h # collection of all instances
library.h # defines enums and structs to describe the tiled structure of operator instances
manifest.h # collection of all instances
scripts/ # scripts to procedurally generate CUTLASS template instances
scripts/ # scripts to procedurally generate CUTLASS template instances
gemm_operations.py
library.py
generator.py # entry point of procedural generation scripts - invoked by cmake
generator.py # entry point of procedural generation scripts - invoked by cmake
manifest.py
src/
```
## Examples
To demonstrate CUTLASS components, several SDK examples are implemented in `examples/`.
When CMake is executed, the CUTLASS Instance Library generator scripts are executed to construct a set of
instantiations in `build/tools/library/generated/`.
The CUTLASS Profiler is designed to initialize the CUTLASS Instance Library and execute all operations contained therein.
### CUTLASS Profiler
The CUTLASS Profiler is designed to load the CUTLASS Instance Library and execute all operations contained therein.
This command-line driven application constructs an execution environment for evaluating functionality and performance.
It is implemented in
```
tools/
profiler/
```
and may be built as follows.
```
$ make cutlass_profiler -j
```
[Further details about the CUTLASS Profiler are described here.](/media/docs/profiler.md)
### CUTLASS Utilities
@@ -122,9 +131,13 @@ tools/
include/
cutlass/
util/ # CUTLASS Utility companion library
reference/ # reference implementation of CUTLASS operators - minimal consideration for performance
reference/ # functional reference implementation of CUTLASS operators
# (minimal consideration for performance)
detail/
*
device/ # device-side reference implementations of CUTLASS operators
thread/
kernel/
@@ -134,28 +147,36 @@ tools/
*
```
[More details about CUTLASS Utilities may be found here.](media/docs/utilities.md)
[More details about CUTLASS Utilities may be found here.](/media/docs/utilities.md)
### CUTLASS Profiler
This is application constructs an execution environment for evaluating the functionality and performance of
CUTLASS components. It is implemented in
```
tools/
profiler/
```
and may be built as follows.
```
$ make cutlass_profiler -j
```
[Further details about the CUTLASS Profiler are described here.](media/docs/profiler.md)
## Examples
To demonstrate CUTLASS components, several SDK examples are implemented in `examples/`.
CUTLASS SDK examples apply CUTLASS templates to implement basic computations.
```
examples/
00_basic_gemm/ # launches a basic GEMM with single precision inputs and outputs
01_cutlass_utilities/ # demonstrates CUTLASS Utilities for allocating and initializing tensors
02_dump_reg_smem/ # debugging utilities for printing register and shared memory contents
03_visualize_layout/ # utility for visualizing all layout functions in CUTLASS
04_tile_iterator/ # example demonstrating an iterator over tiles in memory
05_batched_gemm/ # example demonstrating CUTLASS's batched strided GEMM operation
06_splitK_gemm/ # exmaple demonstrating CUTLASS's Split-K parallel reduction kernel
07_volta_tensorop_gemm/ # example demonstrating mixed precision GEMM using Volta Tensor Cores
08_turing_tensorop_gemm/ # example demonstrating integer GEMM using Turing Tensor Cores
```
## Media
This directory contains documentation, images, and performance result data which accompanies the CUTLASS library and components.