Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
one
spconv
Commits
dc61940b
Commit
dc61940b
authored
Oct 20, 2021
by
yan.yan
Browse files
debug win CI: only compile one kernel to speed up
parent
520065c6
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
139 additions
and
137 deletions
+139
-137
pyproject.toml
pyproject.toml
+1
-1
setup.py
setup.py
+4
-2
spconv/core.py
spconv/core.py
+134
-134
No files found.
pyproject.toml
View file @
dc61940b
[build-system]
[build-system]
requires
=
[
"setuptools>=41.0"
,
"wheel"
,
"pccm>=0.2.11"
,
"cumm>=0.1.
6
"
]
requires
=
[
"setuptools>=41.0"
,
"wheel"
,
"pccm>=0.2.11"
,
"cumm>=0.1.
7
"
]
build-backend
=
"setuptools.build_meta"
build-backend
=
"setuptools.build_meta"
setup.py
View file @
dc61940b
...
@@ -143,7 +143,9 @@ if disable_jit is not None and disable_jit == "1":
...
@@ -143,7 +143,9 @@ if disable_jit is not None and disable_jit == "1":
'upload'
:
UploadCommand
,
'upload'
:
UploadCommand
,
'build_ext'
:
PCCMBuild
,
'build_ext'
:
PCCMBuild
,
}
}
from
cumm.gemm.main
import
GemmMainUnitTest
,
SHUFFLE_SIMT_PARAMS
,
SHUFFLE_VOLTA_PARAMS
,
SHUFFLE_TURING_PARAMS
from
cumm.gemm.main
import
GemmMainUnitTest
from
spconv.core
import
SHUFFLE_SIMT_PARAMS
,
SHUFFLE_VOLTA_PARAMS
,
SHUFFLE_TURING_PARAMS
from
spconv.csrc.sparse.all
import
SpconvOps
from
spconv.csrc.sparse.all
import
SpconvOps
cu
=
GemmMainUnitTest
(
SHUFFLE_SIMT_PARAMS
+
SHUFFLE_VOLTA_PARAMS
+
SHUFFLE_TURING_PARAMS
)
cu
=
GemmMainUnitTest
(
SHUFFLE_SIMT_PARAMS
+
SHUFFLE_VOLTA_PARAMS
+
SHUFFLE_TURING_PARAMS
)
...
@@ -154,7 +156,7 @@ if disable_jit is not None and disable_jit == "1":
...
@@ -154,7 +156,7 @@ if disable_jit is not None and disable_jit == "1":
else
:
else
:
std
=
"c++17"
std
=
"c++17"
ext_modules
:
List
[
Extension
]
=
[
ext_modules
:
List
[
Extension
]
=
[
PCCMExtension
([
cu
,
SpconvOps
()
],
PCCMExtension
([
cu
],
"spconv/core_cc"
,
"spconv/core_cc"
,
Path
(
__file__
).
resolve
().
parent
/
"spconv"
,
Path
(
__file__
).
resolve
().
parent
/
"spconv"
,
objects_folder
=
"objects"
,
objects_folder
=
"objects"
,
...
...
spconv/core.py
View file @
dc61940b
...
@@ -32,16 +32,16 @@ class AlgoHint(Enum):
...
@@ -32,16 +32,16 @@ class AlgoHint(Enum):
# TODO two step build: build gemm kernels first, then bind for every python
# TODO two step build: build gemm kernels first, then bind for every python
SHUFFLE_SIMT_PARAMS
:
List
[
GemmAlgoParams
]
=
[
SHUFFLE_SIMT_PARAMS
:
List
[
GemmAlgoParams
]
=
[
*
gen_shuffle_params
(
#
*gen_shuffle_params(
(
64
,
128
,
32
),
(
32
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
#
(64, 128, 32), (32, 64, 32), ["s8,s8,s32,s32,s32"],
2
,
kernel
.
GemmAlgo
.
SimtDP4A
,
None
),
#
2, kernel.GemmAlgo.SimtDP4A, None),
*
gen_shuffle_params
(
#
*gen_shuffle_params(
(
128
,
64
,
32
),
(
64
,
32
,
32
),
[
"s8,s8,s32,s32,s32"
],
#
(128, 64, 32), (64, 32, 32), ["s8,s8,s32,s32,s32"],
2
,
kernel
.
GemmAlgo
.
SimtDP4A
,
None
),
#
2, kernel.GemmAlgo.SimtDP4A, None),
*
gen_shuffle_params
(
#
*gen_shuffle_params(
(
128
,
128
,
32
),
#
(128, 128, 32),
(
32
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
#
(32, 64, 32), ["s8,s8,s32,s32,s32"], 2,
kernel
.
GemmAlgo
.
SimtDP4A
,
None
),
#
kernel.GemmAlgo.SimtDP4A, None),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 32),
# (128, 128, 32),
# (64, 32, 32), ["s8,s8,s8,s32,s32", "s8,s8,s32,s32,s32"], 2,
# (64, 32, 32), ["s8,s8,s8,s32,s32", "s8,s8,s32,s32,s32"], 2,
...
@@ -49,156 +49,156 @@ SHUFFLE_SIMT_PARAMS: List[GemmAlgoParams] = [
...
@@ -49,156 +49,156 @@ SHUFFLE_SIMT_PARAMS: List[GemmAlgoParams] = [
*
gen_shuffle_params
(
*
gen_shuffle_params
(
(
64
,
64
,
32
),
(
32
,
32
,
32
),
[
"s8,s8,s32,s32,s32"
],
(
64
,
64
,
32
),
(
32
,
32
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
kernel
.
GemmAlgo
.
SimtDP4A
,
None
),
2
,
kernel
.
GemmAlgo
.
SimtDP4A
,
None
),
*
gen_shuffle_params
(
(
64
,
256
,
8
),
(
32
,
64
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (64, 256, 8),
# (64, 256, 8),
# (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
# # *gen_shuffle_params(
(
32
,
128
,
16
),
# # (64, 256, 8),
(
32
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# # (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
32
,
512
,
8
),
# (32, 128, 16),
(
32
,
64
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# (32, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (32, 512, 8),
# (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# # *gen_shuffle_params(
# # (128, 128, 8),
# # (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 8),
# (128, 128, 8),
# (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
(
128
,
128
,
8
),
(
32
,
64
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
64
,
128
,
8
),
(
32
,
64
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (64, 128, 8),
# (64, 128, 8),
# (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# # *gen_shuffle_params(
# # (64, 128, 8),
# # (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# # *gen_shuffle_params(
# # (128, 64, 8),
# # (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 64, 8),
# (128, 64, 8),
# (32, 64, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (64, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
(
128
,
64
,
8
),
(
64
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
64
,
64
,
8
),
(
32
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
32
,
64
,
16
),
(
32
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
64
,
32
,
16
),
(
32
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
32
,
32
,
32
),
(
32
,
32
,
8
),
[
"f32,f32,f32,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# fall back kernels if mat is misaligned for half
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 8),
# (64, 64, 8),
# (32, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (32, 64, 16),
# (32, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (64, 32, 16),
# (32, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (32, 32, 32),
# (32, 32, 8), ["f32,f32,f32,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# # fall back kernels if mat is misaligned for half
# # *gen_shuffle_params(
# # (128, 128, 8),
# # (32, 64, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (32, 64, 32),
# (32, 32, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (32, 32, 32),
# (32, 32, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# # *gen_shuffle_params(
# # (64, 64, 16),
# # (32, 32, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# *gen_shuffle_params(
# (64, 128, 16),
# (32, 64, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (32, 64, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
(
32
,
64
,
32
),
(
32
,
32
,
8
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
32
,
32
,
32
),
(
32
,
32
,
8
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (64, 64,
16
),
# (64, 64,
8
),
# (32, 32, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
# (32, 32, 8), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2, kernel.GemmAlgo.Simt, None),
*
gen_shuffle_params
(
(
64
,
128
,
16
),
(
32
,
64
,
8
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
*
gen_shuffle_params
(
(
64
,
64
,
8
),
(
32
,
32
,
8
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Simt
,
None
),
]
]
SHUFFLE_VOLTA_PARAMS
:
List
[
GemmAlgoParams
]
=
[
SHUFFLE_VOLTA_PARAMS
:
List
[
GemmAlgoParams
]
=
[
*
gen_shuffle_params
(
(
64
,
64
,
32
),
(
32
,
32
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Volta
,
TensorOpParams
((
8
,
8
,
4
))),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 32),
# (64, 64, 32),
# (32, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
# # *gen_shuffle_params(
# # (128, 128, 32),
# # (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# # kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
# *gen_shuffle_params(
# (128, 256, 32),
# (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
128
,
256
,
32
),
# (256, 128, 32),
(
64
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
kernel
.
GemmAlgo
.
Volta
,
TensorOpParams
((
8
,
8
,
4
))),
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
256
,
128
,
32
),
# (128, 64, 32),
(
64
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# (64, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
kernel
.
GemmAlgo
.
Volta
,
TensorOpParams
((
8
,
8
,
4
))),
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
128
,
64
,
32
),
# (64, 128, 32),
(
64
,
32
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# (32, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
kernel
.
GemmAlgo
.
Volta
,
TensorOpParams
((
8
,
8
,
4
))),
# kernel.GemmAlgo.Volta, TensorOpParams((8, 8, 4))),
*
gen_shuffle_params
(
(
64
,
128
,
32
),
(
32
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Volta
,
TensorOpParams
((
8
,
8
,
4
))),
]
]
# SHUFFLE_VOLTA_PARAMS = []
# SHUFFLE_VOLTA_PARAMS = []
SHUFFLE_TURING_PARAMS
:
List
[
GemmAlgoParams
]
=
[
SHUFFLE_TURING_PARAMS
:
List
[
GemmAlgoParams
]
=
[
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
64
,
64
,
32
),
# (64, 64, 32),
(
32
,
32
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# (32, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
*
gen_shuffle_params
(
(
128
,
128
,
32
),
(
32
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 32),
# (128, 128, 32),
# (32, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# # *gen_shuffle_params(
# # (128, 128, 32),
# # (64, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# # kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# *gen_shuffle_params(
# (64, 64, 64),
# (32, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# *gen_shuffle_params(
# (64, 128, 64),
# (32, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# *gen_shuffle_params(
# (128, 256, 32),
# (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# *gen_shuffle_params(
# (256, 128, 32),
# (64, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# *gen_shuffle_params(
# (128, 64, 32),
# (64, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# (64, 32, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
64
,
64
,
64
),
# (64, 128, 32),
(
32
,
32
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# (32, 64, 32), ["f16,f16,f16,f16,f16", "f16,f16,f16,f32,f32"], 2,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
# kernel.GemmAlgo.Turing, TensorOpParams((16, 8, 8))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
64
,
128
,
64
),
# (64, 64, 32), (32, 32, 32), ["s8,s8,s32,s32,s32"],
(
32
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
# 2, kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
*
gen_shuffle_params
(
(
128
,
256
,
32
),
(
64
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
*
gen_shuffle_params
(
(
256
,
128
,
32
),
(
64
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
*
gen_shuffle_params
(
(
128
,
64
,
32
),
(
64
,
32
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
*
gen_shuffle_params
(
(
64
,
128
,
32
),
(
32
,
64
,
32
),
[
"f16,f16,f16,f16,f16"
,
"f16,f16,f16,f32,f32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
16
,
8
,
8
))),
*
gen_shuffle_params
(
(
64
,
64
,
32
),
(
32
,
32
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
*
gen_shuffle_params
(
(
128
,
128
,
32
),
(
32
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
# *gen_shuffle_params(
# *gen_shuffle_params(
# (128, 128, 32),
# (128, 128, 32),
# (
64, 32, 32), ["s8,s8,s8,s32,s32"
, "s8,s8,s32,s32,s32"], 2,
# (
32, 64, 32)
,
[
"s8,s8,s32,s32,s32"], 2,
# kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
# kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
*
gen_shuffle_params
(
# # *gen_shuffle_params(
(
128
,
256
,
32
),
# # (128, 128, 32),
(
64
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
# # (64, 32, 32), ["s8,s8,s8,s32,s32", "s8,s8,s32,s32,s32"], 2,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
# # kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
256
,
128
,
32
),
# (128, 256, 32),
(
64
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
2
,
# (64, 64, 32), ["s8,s8,s32,s32,s32"], 2,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
# kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
*
gen_shuffle_params
(
# *gen_shuffle_params(
(
128
,
64
,
32
),
(
64
,
32
,
32
),
[
"s8,s8,s32,s32,s32"
],
# (256, 128, 32),
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
# (64, 64, 32), ["s8,s8,s32,s32,s32"], 2,
*
gen_shuffle_params
(
# kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
(
64
,
128
,
32
),
(
32
,
64
,
32
),
[
"s8,s8,s32,s32,s32"
],
# *gen_shuffle_params(
2
,
kernel
.
GemmAlgo
.
Turing
,
TensorOpParams
((
8
,
8
,
16
))),
# (128, 64, 32), (64, 32, 32), ["s8,s8,s32,s32,s32"],
# 2, kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
# *gen_shuffle_params(
# (64, 128, 32), (32, 64, 32), ["s8,s8,s32,s32,s32"],
# 2, kernel.GemmAlgo.Turing, TensorOpParams((8, 8, 16))),
]
]
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
