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
925e6ac3
Commit
925e6ac3
authored
Jun 11, 2024
by
root
Browse files
clean
parent
8f5690c4
Changes
2
Expand all
Show whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
236 additions
and
685 deletions
+236
-685
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
...gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
+93
-306
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_abd.hpp
...tion/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_abd.hpp
+143
-379
No files found.
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
View file @
925e6ac3
...
...
@@ -88,7 +88,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
using
BLayout
=
remove_cvref_t
<
tuple_element_t
<
0
,
BsLayout
>>
;
// GridwiseGemm
using
GridwiseGemm
=
GridwiseGemm_xdl_cshuffle_v3
<
using
GridwiseGemm
=
GridwiseGemm
MultiABD
_xdl_cshuffle_v3
<
ALayout
,
BLayout
,
CLayout
,
...
...
@@ -165,12 +165,6 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
const
bool
has_main_k_block_loop
=
GridwiseGemm
::
CalculateHasMainKBlockLoop
(
K_split
);
const
auto
Run
=
[
&
](
const
auto
&
kernel
)
{
if
(
arg
.
KBatch
>
1
)
hipGetErrorString
(
hipMemsetAsync
(
arg
.
p_c_grid
,
0
,
arg
.
M
*
arg
.
N
*
sizeof
(
CDataType
),
stream_config
.
stream_id_
));
ave_time
=
launch_and_time_kernel
(
stream_config
,
kernel
,
dim3
(
gdx
,
gdy
,
gdz
),
dim3
(
BlockSize
),
0
,
arg
);
};
...
...
@@ -184,146 +178,16 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
if
constexpr
(
BlkGemmPipelineVer
==
BlockGemmPipelineVersion
::
v1
||
BlkGemmPipelineVer
==
BlockGemmPipelineVersion
::
v3
)
{
#if 0
if(arg.KBatch > 1)
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy>;
Run(kernel);
}
else
#endif
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
true
,
InMemoryDataOperationEnum
::
Set
,
minimum_occupancy
>
;
Run
(
kernel
);
}
}
// Tail number could be One to Seven
else
if
constexpr
(
BlkGemmPipelineVer
==
BlockGemmPipelineVersion
::
v2
)
{
#if 0
if(arg.KBatch > 1)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::One)
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::One>;
Run(kernel);
}
else if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
TailNumber::Full)
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Full>;
Run(kernel);
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 2)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Two)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Two>;
Run(kernel);
}
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 3)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
TailNumber::Three)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Three>;
Run(kernel);
}
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 4)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
TailNumber::Four)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Four>;
Run(kernel);
}
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 5)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
TailNumber::Five)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Five>;
Run(kernel);
}
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 6)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Six)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Six>;
Run(kernel);
}
}
if constexpr(GridwiseGemm::BlockwiseGemmPipe::PrefetchStages > 7)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
TailNumber::Seven)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Seven>;
Run(kernel);
}
}
}
else
#endif
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
One
)
{
const
auto
kernel
=
...
...
@@ -334,8 +198,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
TailNumber
::
One
>
;
Run
(
kernel
);
}
else
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Full
)
else
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Full
)
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
...
...
@@ -362,8 +225,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
if
constexpr
(
GridwiseGemm
::
BlockwiseGemmPipe
::
PrefetchStages
>
3
)
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Three
)
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Three
)
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
...
...
@@ -377,8 +239,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
if
constexpr
(
GridwiseGemm
::
BlockwiseGemmPipe
::
PrefetchStages
>
4
)
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Four
)
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Four
)
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
...
...
@@ -392,8 +253,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
if
constexpr
(
GridwiseGemm
::
BlockwiseGemmPipe
::
PrefetchStages
>
5
)
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Five
)
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Five
)
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
...
...
@@ -421,8 +281,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
if
constexpr
(
GridwiseGemm
::
BlockwiseGemmPipe
::
PrefetchStages
>
7
)
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Seven
)
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Seven
)
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
...
...
@@ -434,37 +293,9 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
}
}
}
}
// Tail number could be Odd or Even
else
if
constexpr
(
BlkGemmPipelineVer
==
BlockGemmPipelineVersion
::
v4
)
{
#if 0
if(arg.KBatch > 1)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3_2lds<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Odd>;
Run(kernel);
}
else
{
const auto kernel = kernel_gemm_xdl_cshuffle_v3_2lds<
GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Even>;
Run(kernel);
}
}
else
#endif
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Odd
)
{
const
auto
kernel
=
...
...
@@ -486,35 +317,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
Run
(
kernel
);
}
}
}
else
{
#if 0
if(arg.KBatch > 1)
{
if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Odd>;
Run(kernel);
}
else
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
true,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy,
TailNumber::Even>;
Run(kernel);
}
}
else
#endif
{
if
(
GridwiseGemm
::
CalculateKBlockLoopTailNum
(
K_split
)
==
TailNumber
::
Odd
)
{
...
...
@@ -538,34 +341,18 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
}
}
}
}
else
{
// Tail number always 1
if
constexpr
(
BlkGemmPipelineVer
==
BlockGemmPipelineVersion
::
v1
)
{
#if 0
if(arg.KBatch > 1)
{
const auto kernel =
kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
false,
InMemoryDataOperationEnum::AtomicAdd,
minimum_occupancy>;
Run(kernel);
}
else
#endif
{
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
const
auto
kernel
=
kernel_gemm_xdl_cshuffle_v3
<
GridwiseGemm
,
false
,
InMemoryDataOperationEnum
::
Set
,
minimum_occupancy
>
;
Run
(
kernel
);
}
}
}
return
ave_time
;
}
...
...
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_abd.hpp
View file @
925e6ac3
This diff is collapsed.
Click to expand it.
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