ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 21:39:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Adding more instances of grouped convolution 3d forward for FP8 with ConvScale+Bias element-wise operation. (#1412)

Browse Source 
* Add CMakePresets configurations.

* Add binary elementwise ConvScaleAdd and an example.

* Numerical verification of results.

Observed significant irregularities in F8 to F32 type conversions:
```log
ConvScaleAdd: float=145.000000   f8_t=160.000000    e=144.000000
ConvScaleAdd: float=97.000000   f8_t=96.000000    e=104.000000
ConvScaleAdd: float=65.000000   f8_t=64.000000    e=72.000000
```

* Implemented ConvScaleAdd + Example.

* Add ConvScale+Bias Instances

* Add Client Example for ConvScale+Bias

* Fix number of bytes in an example..

* Cleanup.
This commit is contained in:
Andriy Roshchenko
2024-07-24 16:49:55 -04:00
committed by GitHub
parent ffabd70a15
commit 4a8a1befd5
13 changed files with 1151 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_convscale_add/CMakeLists.txt Normal file
View File

@@ -0,0 +1,5 @@
# ONLY XDL_KERNELS
set(GROUPED_CONV3D_FWD_CONVSCALE_ADD
xdl/device_grouped_conv3d_fwd_xdl_convscale_add_ndhwgc_gkzyxc_ndhwgk_f8_instance.cpp)
add_instance_library(device_grouped_conv3d_fwd_convscale_add_instance ${GROUPED_CONV3D_FWD_CONVSCALE_ADD})

63
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_convscale_add/xdl/device_grouped_conv3d_fwd_xdl_convscale_add_ndhwgc_gkzyxc_ndhwgk_f8_instance.cpp Normal file
View File

@@ -0,0 +1,63 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_binary_outelementop_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using ConvScaleAdd = ck::tensor_operation::element_wise::ConvScaleAdd;
void add_device_grouped_conv3d_fwd_xdl_convscale_add_ndhwgc_gkzyxc_ndhwgk_f8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
F8,
F8,
ck::Tuple<F32>,
F8,
PassThrough,
PassThrough,
ConvScaleAdd,
F8,
F8>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_binary_outelementop_f8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
ConvFwdDefault,
ConvScaleAdd>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_binary_outelementop_f8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1P0,
ConvScaleAdd>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_binary_outelementop_f8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<NDHWGK>,
NDHWGK,
ConvFwd1x1S1P0,
ConvScaleAdd>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck