Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
gaoqiong
composable_kernel_ROCM
Commits
b321bd86
Commit
b321bd86
authored
Nov 29, 2024
by
rocking
Browse files
Support pure quant in instance library
parent
26f221eb
Changes
24
Show whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
185 additions
and
110 deletions
+185
-110
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n1024_instance.cpp
...smoothquant/instances/smoothquant_bf16_n1024_instance.cpp
+9
-13
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n1536_instance.cpp
...smoothquant/instances/smoothquant_bf16_n1536_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n2048_instance.cpp
...smoothquant/instances/smoothquant_bf16_n2048_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n256_instance.cpp
..._smoothquant/instances/smoothquant_bf16_n256_instance.cpp
+8
-4
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n3072_instance.cpp
...smoothquant/instances/smoothquant_bf16_n3072_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n4096_instance.cpp
...smoothquant/instances/smoothquant_bf16_n4096_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n4096_tp_instance.cpp
...othquant/instances/smoothquant_bf16_n4096_tp_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n512_instance.cpp
..._smoothquant/instances/smoothquant_bf16_n512_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n64_n128_instance.cpp
...othquant/instances/smoothquant_bf16_n64_n128_instance.cpp
+8
-4
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n768_instance.cpp
..._smoothquant/instances/smoothquant_bf16_n768_instance.cpp
+8
-4
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n1024_instance.cpp
...smoothquant/instances/smoothquant_fp16_n1024_instance.cpp
+9
-13
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n1536_instance.cpp
...smoothquant/instances/smoothquant_fp16_n1536_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n2048_instance.cpp
...smoothquant/instances/smoothquant_fp16_n2048_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n256_instance.cpp
..._smoothquant/instances/smoothquant_fp16_n256_instance.cpp
+8
-4
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n3072_instance.cpp
...smoothquant/instances/smoothquant_fp16_n3072_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n4096_instance.cpp
...smoothquant/instances/smoothquant_fp16_n4096_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n4096_tp_instance.cpp
...othquant/instances/smoothquant_fp16_n4096_tp_instance.cpp
+9
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n512_instance.cpp
..._smoothquant/instances/smoothquant_fp16_n512_instance.cpp
+10
-5
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n64_n128_instance.cpp
...othquant/instances/smoothquant_fp16_n64_n128_instance.cpp
+8
-4
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n768_instance.cpp
..._smoothquant/instances/smoothquant_fp16_n768_instance.cpp
+8
-4
No files found.
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n1024_instance.cpp
View file @
b321bd86
...
@@ -5,18 +5,14 @@
...
@@ -5,18 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
#if 0
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 4, 64, 8, true, false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 4, 64, 4, true, false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::bf16_t, 1, 8, 4, 64, 2, true, false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::bf16_t, 1, 16, 4, 64, 1, true, false>>(const S&, A);
template float smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 1, 256, 4, true, false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
#endif
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n1536_instance.cpp
View file @
b321bd86
...
@@ -5,9 +5,14 @@
...
@@ -5,9 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n2048_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n256_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n3072_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n4096_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n4096_tp_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n512_instance.cpp
View file @
b321bd86
...
@@ -5,9 +5,14 @@
...
@@ -5,9 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n64_n128_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_bf16_n768_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
bf16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n1024_instance.cpp
View file @
b321bd86
...
@@ -5,18 +5,14 @@
...
@@ -5,18 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
#if 0
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::fp16_t, 1, 2, 4, 64, 8, true ,false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::fp16_t, 1, 4, 4, 64, 4, true ,false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::fp16_t, 1, 8, 4, 64, 2, true ,false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template float smoothquant_<trait_<ck_tile::fp16_t, 1, 16, 4, 64, 1, true ,false>>(const S&, A);
template float smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 1, 256, 4, true ,false>>(const S&, A);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
#endif
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
2
,
128
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
2
,
128
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
2
,
128
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n1536_instance.cpp
View file @
b321bd86
...
@@ -5,9 +5,14 @@
...
@@ -5,9 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
2
,
128
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n2048_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
1
,
256
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
1
,
256
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n256_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n3072_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
128
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
1
,
256
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
1
,
1024
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n4096_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,15 @@
...
@@ -5,10 +5,15 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n4096_tp_instance.cpp
View file @
b321bd86
...
@@ -5,10 +5,14 @@
...
@@ -5,10 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p
sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
256
,
8
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
256
,
4
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
1
,
1024
,
2
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
1
,
1024
,
1
,
true
,
true
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n512_instance.cpp
View file @
b321bd86
...
@@ -5,9 +5,14 @@
...
@@ -5,9 +5,14 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
8
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
4
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
8
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n64_n128_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
1
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
2
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
example/ck_tile/12_smoothquant/instances/smoothquant_fp16_n768_instance.cpp
View file @
b321bd86
...
@@ -5,8 +5,12 @@
...
@@ -5,8 +5,12 @@
#include "smoothquant_instance_common.hpp"
#include "smoothquant_instance_common.hpp"
// clang-format off
// clang-format off
// rm rn tm tn vn pd 2p
// rm rn tm tn vn pd 2p sx
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
,
false
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
3
,
4
,
64
,
4
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
6
,
4
,
64
,
2
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
template
float
smoothquant_
<
trait_
<
ck_tile
::
fp16_t
,
1
,
12
,
4
,
64
,
1
,
true
,
false
,
true
>
>
(
const
S
&
,
A
);
// clang-format on
// clang-format on
Prev
1
2
Next
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