ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 14:59:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Ck_tile] smoothquant (#1617)

Browse Source 
* fix compile error

* fix typo of padding

* Add smoothquant op

* Add smoothquant instance library

* refine type

* add test script

* Re-generate smoothquant.hpp

* Always use 'current year' in copyright

* use Generic2dBlockShape instead

* Add vector = 8 instance back

* Find exe path automatically

* Simplify the api condition

* Remove debugging code

* update year

* Add blank line between function declaration

* explicitly cast return value to dim3

* refine return value

* Fix default warmup and repeat value

* Add comment

* refactor sommthquant cmake

* Add README

* Fix typo

---------

Co-authored-by: Po Yen, Chen <PoYen.Chen@amd.com>
This commit is contained in:
rocking
2024-11-01 13:51:56 +08:00
committed by GitHub
parent 550248deec
commit fbd654545a
62 changed files with 1758 additions and 219 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/10_rmsnorm2d/example_rmsnorm2d_fwd.cpp
View File

@@ -69,7 +69,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
using WarpTile = ck_tile::sequence<1, 64>;
using Vector = ck_tile::sequence<1, 1>;
using Shape = ck_tile::Rmsnorm2dShape<BlockTile, BlockWarps, WarpTile, Vector>;
using Shape = ck_tile::Generic2dBlockShape<BlockTile, BlockWarps, WarpTile, Vector>;
using Problem = ck_tile::Rmsnorm2dFwdPipelineProblem<XDataType,
GammaDataType,
ComputeDataType,

9
example/ck_tile/10_rmsnorm2d/instances/rmsnorm2d_fwd_api.cpp
View File

@@ -28,7 +28,6 @@ float rmsnorm2d_fwd_b16_(rmsnorm2d_fwd_traits /*t*/,
rmsnorm2d_fwd_args a,
const ck_tile::stream_config& s)
{
#if 1
float r = -1;
// clang-format off
// rm rn tm tn vn pd rms 2p
@@ -128,16 +127,12 @@ float rmsnorm2d_fwd_b16_(rmsnorm2d_fwd_traits /*t*/,
r = rmsnorm2d_fwd_<trait_<data_type, 1, 4, 1, 1024, 1, true, false, true>>(s, a);
}
return r;
#else
return rmsnorm2d_fwd_<trait_<data_type, 1, 1, 1, 256, 4, true, false, false>>(s, a);
#endif
// clang-format on
}
float rmsnorm2d_fwd(rmsnorm2d_fwd_traits t, rmsnorm2d_fwd_args a, const ck_tile::stream_config& s)
{
float r = -1;
if(t.data_type.compare("fp16") == 0)
{
return rmsnorm2d_fwd_b16_<ck_tile::fp16_t>(t, a, s);
@@ -146,8 +141,6 @@ float rmsnorm2d_fwd(rmsnorm2d_fwd_traits t, rmsnorm2d_fwd_args a, const ck_tile:
{
return rmsnorm2d_fwd_b16_<ck_tile::bf16_t>(t, a, s);
}
if(r < 0)
else
throw std::runtime_error("Without supported instances!");
return r;
}

2
example/ck_tile/10_rmsnorm2d/rmsnorm2d_fwd.hpp
View File

@@ -97,7 +97,7 @@ struct rmsnorm2d_fwd_traits_
using WarpTile = ck_tile::sequence<Warp_M, Warp_N>;
using Vector = ck_tile::sequence<1, Vector_N_>;
using Shape = ck_tile::Rmsnorm2dShape<BlockTile, BlockWarps, WarpTile, Vector>;
using Shape = ck_tile::Generic2dBlockShape<BlockTile, BlockWarps, WarpTile, Vector>;
static constexpr bool kPadN = kPadN_;
static constexpr bool kSaveInvRms = kSaveInvRms_;

5
example/ck_tile/10_rmsnorm2d/script/perf_test.sh
View File

@@ -1,6 +1,5 @@
# run from top of ck folder
EXE=build/bin/tile_rmsnorm2d_fwd
#!/bin/sh
EXE="$(find . -name tile_rmsnorm2d_fwd -type f | head -n 1)"
$EXE -m=1 -n=1 -e=1e-12 -v=1 -prec=bf16 -repeat=1000
$EXE -m=700 -n=80 -e=1e-12 -v=1 -prec=bf16 -repeat=1000

3
example/ck_tile/10_rmsnorm2d/script/smoke_test.sh
View File

@@ -1,6 +1,5 @@
#!/bin/sh
# call from top of CK folder
EXE=./build/bin/tile_rmsnorm2d_fwd
EXE="$(find . -name tile_rmsnorm2d_fwd -type f | head -n 1)"
for pr_i in "fp16" "bf16" ; do
$EXE -prec=$pr_i -m=99 -n=13