mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-14 02:02:46 +00:00
NHWC conv 2d: fwd bfp16/int8, Device level tuning and host API (#73)
* add fwd bf16 conv
* change tunning parametor
* add int8 for conv fwd
* remove comments
* change tunning parametor for int8
* change init int8 example
* add test for conv2d fwd
* change device operation file pos because merge develop
* fwd int8 use reference
* test_conv_fwd use reference
* add braket for if statement
* rename fwd example name
* remove StaticBufferOfVectorTypeV2
* tweak example
Co-authored-by: ltqin <letaoqin@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
[ROCm/composable_kernel commit: 880fbee957]
This commit is contained in:
ltqin
@@ -10,6 +10,12 @@ struct PassThrough
|
||||
__host__ __device__ void operator()(float& y, const float& x) const { y = x; }
|
||||
|
||||
__host__ __device__ void operator()(half_t& y, const half_t& x) const { y = x; }
|
||||
|
||||
__host__ __device__ void operator()(ushort& y, const ushort& x) const { y = x; }
|
||||
|
||||
__host__ __device__ void operator()(int32_t& y, const int32_t& x) const { y = x; }
|
||||
|
||||
__host__ __device__ void operator()(int8_t& y, const int8_t& x) const { y = x; }
|
||||
};
|
||||
|
||||
struct Add
|
||||
|
||||
@@ -30,8 +30,6 @@
|
||||
#include "amd_address_space.hpp"
|
||||
#include "amd_buffer_addressing.hpp"
|
||||
#include "static_buffer.hpp"
|
||||
// TODO remove this
|
||||
#include "static_buffer_of_vector_type_v2.hpp"
|
||||
#include "dynamic_buffer.hpp"
|
||||
#include "is_known_at_compile_time.hpp"
|
||||
#include "transpose_vectors.hpp"
|
||||
|
||||
@@ -171,6 +171,8 @@ struct DynamicBuffer
|
||||
is_same<remove_cvref_t<X>, int8x4_t>::value) ||
|
||||
(is_same<remove_cvref_t<T>, int8_t>::value &&
|
||||
is_same<remove_cvref_t<X>, int8x8_t>::value) ||
|
||||
(is_same<remove_cvref_t<T>, int8_t>::value &&
|
||||
is_same<remove_cvref_t<X>, int8x16_t>::value) ||
|
||||
(is_same<remove_cvref_t<T>, int8x4_t>::value &&
|
||||
is_same<remove_cvref_t<X>, int8x4_t>::value) ||
|
||||
(is_same<remove_cvref_t<T>, int8x8_t>::value &&
|
||||
@@ -212,6 +214,14 @@ struct DynamicBuffer
|
||||
*c_style_pointer_cast<int32x2_t*>(&p_data_[i]) =
|
||||
*c_style_pointer_cast<const int32x2_t*>(&x);
|
||||
}
|
||||
else if constexpr(is_same<remove_cvref_t<T>, int8_t>::value &&
|
||||
is_same<remove_cvref_t<X>, int8x16_t>::value)
|
||||
{
|
||||
// HACK: cast pointer of x is bad
|
||||
// TODO: remove this after compiler fix
|
||||
*c_style_pointer_cast<int32x4_t*>(&p_data_[i]) =
|
||||
*c_style_pointer_cast<const int32x4_t*>(&x);
|
||||
}
|
||||
else if constexpr(is_same<remove_cvref_t<T>, int8x4_t>::value &&
|
||||
is_same<remove_cvref_t<X>, int8x4_t>::value)
|
||||
{
|
||||
|
||||
@@ -1,118 +0,0 @@
|
||||
#ifndef CK_STATIC_BUFFER_OF_VECTOR_TYPE_V2_HPP
|
||||
#define CK_STATIC_BUFFER_OF_VECTOR_TYPE_V2_HPP
|
||||
|
||||
#include "statically_indexed_array.hpp"
|
||||
|
||||
namespace ck {
|
||||
template <AddressSpaceEnum_t BufferAddressSpace,
|
||||
typename T,
|
||||
index_t N,
|
||||
bool InvalidElementUseNumericalZeroValue>
|
||||
struct StaticBufferOfVectorTypeV2 : public StaticallyIndexedArray<T, N>
|
||||
{
|
||||
using type = T;
|
||||
using base = StaticallyIndexedArray<T, N>;
|
||||
|
||||
using VecBaseType = typename T::d1_t;
|
||||
|
||||
__host__ __device__ static constexpr index_t GetVectorSize()
|
||||
{
|
||||
return sizeof(typename T::type) / sizeof(VecBaseType);
|
||||
}
|
||||
|
||||
static constexpr index_t vector_size = GetVectorSize();
|
||||
|
||||
__host__ __device__ static constexpr index_t GetNumVectors() { return N; }
|
||||
|
||||
__host__ __device__ static constexpr index_t GetNumElements()
|
||||
{
|
||||
return GetVectorSize() * GetNumVectors();
|
||||
}
|
||||
|
||||
VecBaseType invalid_element_value_ = VecBaseType{0};
|
||||
|
||||
T invalid_vec_value_ = T{0};
|
||||
|
||||
__host__ __device__ constexpr StaticBufferOfVectorTypeV2() : base{} {}
|
||||
|
||||
__host__ __device__ constexpr StaticBufferOfVectorTypeV2(VecBaseType invalid_element_value)
|
||||
: base{},
|
||||
invalid_vec_value_{invalid_element_value},
|
||||
invalid_element_value_{invalid_element_value}
|
||||
{
|
||||
}
|
||||
|
||||
__host__ __device__ static constexpr AddressSpaceEnum_t GetAddressSpace()
|
||||
{
|
||||
return BufferAddressSpace;
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr auto& GetVector(Number<I> vec_id)
|
||||
{
|
||||
return this->At(vec_id);
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr const auto& GetVector(Number<I> vec_id) const
|
||||
{
|
||||
return this->At(vec_id);
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr auto& GetElement(Number<I> i, bool)
|
||||
{
|
||||
constexpr auto vec_id = Number<i / vector_size>{};
|
||||
constexpr auto vec_off = Number<i % vector_size>{};
|
||||
|
||||
return this->At(vec_id).template AsType<VecBaseType>()(vec_off);
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr auto GetElement(Number<I> i, bool is_valid_element) const
|
||||
{
|
||||
constexpr auto vec_id = Number<i / vector_size>{};
|
||||
constexpr auto vec_off = Number<i % vector_size>{};
|
||||
|
||||
if constexpr(InvalidElementUseNumericalZeroValue)
|
||||
{
|
||||
return is_valid_element ? this->At(vec_id).template AsType<VecBaseType>()[vec_off]
|
||||
: VecBaseType{0};
|
||||
}
|
||||
else
|
||||
{
|
||||
return is_valid_element ? this->At(vec_id).template AsType<VecBaseType>()[vec_off]
|
||||
: invalid_element_value_;
|
||||
}
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr auto operator[](Number<I> i) const
|
||||
{
|
||||
return GetElement(i, true);
|
||||
}
|
||||
|
||||
template <index_t I>
|
||||
__host__ __device__ constexpr auto& operator()(Number<I> i)
|
||||
{
|
||||
return GetElement(i, true);
|
||||
}
|
||||
|
||||
__host__ __device__ void Clear()
|
||||
{
|
||||
static_for<0, GetNumElements(), 1>{}(
|
||||
[&](auto i) { GetElement(i, true) = invalid_element_value_; });
|
||||
}
|
||||
|
||||
__host__ __device__ void Fill(VecBaseType v)
|
||||
{
|
||||
static_for<0, GetNumElements(), 1>{}([&](auto i) { GetElement(i, true) = v; });
|
||||
}
|
||||
|
||||
__host__ __device__ static constexpr bool IsStaticBuffer() { return true; }
|
||||
|
||||
__host__ __device__ static constexpr bool IsDynamicBuffer() { return false; }
|
||||
};
|
||||
|
||||
} // namespace ck
|
||||
#endif
|
||||
@@ -59,6 +59,8 @@ set(DEVICE_BATCHED_GEMM_INSTANCE_SOURCE
|
||||
set(DEVICE_CONV2D_FWD_INSTANCE_SOURCE
|
||||
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
|
||||
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
|
||||
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
|
||||
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp;
|
||||
${PROJECT_SOURCE_DIR}/device_operation/src/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp;
|
||||
)
|
||||
|
||||
|
||||
@@ -0,0 +1,109 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv2d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< ushort, ushort, ushort, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_bf16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,109 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv2d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_int8_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
57
example/9_conv2d_fwd_xdl_int8/README.md
Normal file
57
example/9_conv2d_fwd_xdl_int8/README.md
Normal file
@@ -0,0 +1,57 @@
|
||||
# Instructions for ```conv2d_fwd_xdl``` Example
|
||||
|
||||
## Docker script
|
||||
```bash
|
||||
docker run \
|
||||
-it \
|
||||
--rm \
|
||||
--privileged \
|
||||
--group-add sudo \
|
||||
-w /root/workspace \
|
||||
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
|
||||
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
|
||||
/bin/bash
|
||||
```
|
||||
|
||||
## Build ```conv2d_fwd_xdl```
|
||||
```bash
|
||||
mkdir build && cd build
|
||||
```
|
||||
|
||||
```bash
|
||||
# Need to specify target ID, example below is gfx908
|
||||
cmake \
|
||||
-D BUILD_DEV=OFF \
|
||||
-D CMAKE_BUILD_TYPE=Release \
|
||||
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
|
||||
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
|
||||
-D CMAKE_PREFIX_PATH=/opt/rocm \
|
||||
..
|
||||
```
|
||||
|
||||
```bash
|
||||
make -j conv2d_fwd_xdl
|
||||
```
|
||||
|
||||
## Run ```conv2d_fwd_xdl_int8```
|
||||
```bash
|
||||
#arg1: verification (0=no, 1=yes)
|
||||
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
|
||||
#arg3: run kernel # of times (>1)
|
||||
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
|
||||
./example/conv2d_fwd_xdl_int8 0 1 5
|
||||
```
|
||||
|
||||
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
|
||||
```
|
||||
in_n_c_hi_wi: dim 4, lengths {128, 192, 71, 71}, strides {967872, 1, 13632, 192}
|
||||
wei_k_c_y_x: dim 4, lengths {256, 192, 3, 3}, strides {1728, 1, 576, 192}
|
||||
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
|
||||
arg.a_grid_desc_k0_m_k1_{216, 165888, 8}
|
||||
arg.b_grid_desc_k0_n_k1_{216, 256, 8}
|
||||
arg.c_grid_desc_m_n_{ 165888, 256}
|
||||
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
|
||||
Warm up
|
||||
Start running 5 times...
|
||||
Perf: 1.43206 ms, 102.486 TFlops, 232.947 GB/s
|
||||
```
|
||||
270
example/9_conv2d_fwd_xdl_int8/conv2d_fwd_xdl_int8.cpp
Normal file
270
example/9_conv2d_fwd_xdl_int8/conv2d_fwd_xdl_int8.cpp
Normal file
@@ -0,0 +1,270 @@
|
||||
#include <iostream>
|
||||
#include <numeric>
|
||||
#include <initializer_list>
|
||||
#include <cstdlib>
|
||||
#include <stdlib.h>
|
||||
#include <half.hpp>
|
||||
#include "config.hpp"
|
||||
#include "print.hpp"
|
||||
#include "device.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_conv_fwd.hpp"
|
||||
|
||||
using InDataType = int8_t;
|
||||
using WeiDataType = int8_t;
|
||||
using OutDataType = int8_t;
|
||||
using AccDataType = int32_t;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using InLayout = ck::tensor_layout::convolution::NHWC;
|
||||
using WeiLayout = ck::tensor_layout::convolution::KYXC;
|
||||
using OutLayout = ck::tensor_layout::convolution::NHWK;
|
||||
|
||||
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
using DeviceConvFwdInstance = ck::tensor_operation::device::
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
|
||||
int8_t, // InDataType
|
||||
int8_t, // WeiDataType
|
||||
int8_t, // OutDataType
|
||||
int32_t, // AccDataType
|
||||
PassThrough, // InElementwiseOperation
|
||||
PassThrough, // WeiElementwiseOperation
|
||||
PassThrough, // OutElementwiseOperation
|
||||
ConvFwdDefault, // ConvForwardSpecialization
|
||||
256, // BlockSize
|
||||
128, // MPerBlock
|
||||
256, // NPerBlock
|
||||
4, // K0PerBlock
|
||||
16, // K1
|
||||
32, // MPerXdl
|
||||
32, // NPerXdl
|
||||
2, // MXdlPerWave
|
||||
4, // NXdlPerWave
|
||||
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
16, // ABlockTransferSrcScalarPerVector
|
||||
16, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
16, // BBlockTransferSrcScalarPerVector
|
||||
16, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockLdsAddExtraN
|
||||
7, // CThreadTransferSrcDstVectorDim
|
||||
1>; // CThreadTransferDstScalarPerVector
|
||||
|
||||
using ReferenceConvFwdInstance = ck::tensor_operation::host::
|
||||
ReferenceConvFwd<InDataType, WeiDataType, OutDataType, InElementOp, WeiElementOp, OutElementOp>;
|
||||
|
||||
int main(int argc, char* argv[])
|
||||
{
|
||||
bool do_verification = 0;
|
||||
int init_method = 0;
|
||||
int nrepeat = 5;
|
||||
|
||||
// Conv shape
|
||||
ck::index_t N = 128;
|
||||
ck::index_t K = 256;
|
||||
ck::index_t C = 192;
|
||||
ck::index_t Y = 3;
|
||||
ck::index_t X = 3;
|
||||
ck::index_t Hi = 71;
|
||||
ck::index_t Wi = 71;
|
||||
ck::index_t conv_stride_h = 2;
|
||||
ck::index_t conv_stride_w = 2;
|
||||
ck::index_t conv_dilation_h = 1;
|
||||
ck::index_t conv_dilation_w = 1;
|
||||
ck::index_t in_left_pad_h = 1;
|
||||
ck::index_t in_left_pad_w = 1;
|
||||
ck::index_t in_right_pad_h = 1;
|
||||
ck::index_t in_right_pad_w = 1;
|
||||
|
||||
if(argc == 4)
|
||||
{
|
||||
do_verification = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
nrepeat = std::stoi(argv[3]);
|
||||
}
|
||||
else if(argc == 19)
|
||||
{
|
||||
do_verification = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
nrepeat = std::stoi(argv[3]);
|
||||
|
||||
N = std::stoi(argv[4]);
|
||||
K = std::stoi(argv[5]);
|
||||
C = std::stoi(argv[6]);
|
||||
Y = std::stoi(argv[7]);
|
||||
X = std::stoi(argv[8]);
|
||||
Hi = std::stoi(argv[9]);
|
||||
Wi = std::stoi(argv[10]);
|
||||
conv_stride_h = std::stoi(argv[11]);
|
||||
conv_stride_w = std::stoi(argv[12]);
|
||||
conv_dilation_h = std::stoi(argv[13]);
|
||||
conv_dilation_w = std::stoi(argv[14]);
|
||||
in_left_pad_h = std::stoi(argv[15]);
|
||||
in_left_pad_w = std::stoi(argv[16]);
|
||||
in_right_pad_h = std::stoi(argv[17]);
|
||||
in_right_pad_w = std::stoi(argv[18]);
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("arg1: verification (0=no, 1=yes)\n");
|
||||
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
|
||||
printf("arg3: run kernel # of times (>1)\n");
|
||||
printf("arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
|
||||
"RightPx\n");
|
||||
exit(0);
|
||||
}
|
||||
|
||||
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
|
||||
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
|
||||
|
||||
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
|
||||
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
|
||||
|
||||
const std::vector<ck::index_t> conv_filter_strides{{conv_stride_h, conv_stride_w}};
|
||||
const std::vector<ck::index_t> conv_filter_dilations{{conv_dilation_h, conv_dilation_w}};
|
||||
const std::vector<ck::index_t> input_left_pads{{in_left_pad_h, in_left_pad_w}};
|
||||
const std::vector<ck::index_t> input_right_pads{{in_right_pad_h, in_right_pad_w}};
|
||||
|
||||
// tensor layout
|
||||
auto f_host_tensor_descriptor = [](std::size_t N_,
|
||||
std::size_t C_,
|
||||
std::size_t H,
|
||||
std::size_t W,
|
||||
auto layout) {
|
||||
if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value ||
|
||||
ck::is_same<decltype(layout), ck::tensor_layout::convolution::KCYX>::value ||
|
||||
ck::is_same<decltype(layout), ck::tensor_layout::convolution::NKHW>::value)
|
||||
{
|
||||
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
|
||||
std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
|
||||
}
|
||||
else if constexpr(ck::is_same<decltype(layout),
|
||||
ck::tensor_layout::convolution::NHWC>::value ||
|
||||
ck::is_same<decltype(layout),
|
||||
ck::tensor_layout::convolution::KYXC>::value ||
|
||||
ck::is_same<decltype(layout),
|
||||
ck::tensor_layout::convolution::NHWK>::value)
|
||||
{
|
||||
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
|
||||
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
|
||||
}
|
||||
};
|
||||
|
||||
Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi, InLayout{}));
|
||||
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X, WeiLayout{}));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_host_result(
|
||||
f_host_tensor_descriptor(N, K, Ho, Wo, OutLayout{}));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_device_result(
|
||||
f_host_tensor_descriptor(N, K, Ho, Wo, OutLayout{}));
|
||||
|
||||
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
|
||||
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
|
||||
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo_host_result.mDesc << std::endl;
|
||||
|
||||
switch(init_method)
|
||||
{
|
||||
case 0: break;
|
||||
case 1:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-1, 1});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-1, 1});
|
||||
break;
|
||||
default:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{0, 1});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1, 1});
|
||||
}
|
||||
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) *
|
||||
out_n_k_ho_wo_device_result.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
|
||||
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
|
||||
|
||||
// do GEMM
|
||||
auto conv = DeviceConvFwdInstance{};
|
||||
auto invoker = conv.MakeInvoker();
|
||||
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
N,
|
||||
K,
|
||||
C,
|
||||
std::vector<ck::index_t>{{Hi, Wi}},
|
||||
std::vector<ck::index_t>{{Y, X}},
|
||||
std::vector<ck::index_t>{{Ho, Wo}},
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
if(!conv.IsSupportedArgument(argument))
|
||||
{
|
||||
throw std::runtime_error(
|
||||
"wrong! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem");
|
||||
}
|
||||
|
||||
float ave_time = invoker.Run(argument, nrepeat);
|
||||
|
||||
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
|
||||
|
||||
std::size_t num_btype = sizeof(InDataType) * (N * C * Hi * Wi) +
|
||||
sizeof(WeiDataType) * (K * C * Y * X) +
|
||||
sizeof(OutDataType) * (N * K * Ho * Wo);
|
||||
|
||||
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
|
||||
|
||||
float gb_per_sec = num_btype / 1.E6 / ave_time;
|
||||
|
||||
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
|
||||
<< std::endl;
|
||||
|
||||
if(do_verification)
|
||||
{
|
||||
auto ref_conv = ReferenceConvFwdInstance{};
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
|
||||
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi,
|
||||
wei_k_c_y_x,
|
||||
out_n_k_ho_wo_host_result,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
|
||||
out_device_buf.FromDevice(out_n_k_ho_wo_device_result.mData.data());
|
||||
|
||||
check_error(out_n_k_ho_wo_host_result, out_n_k_ho_wo_device_result);
|
||||
}
|
||||
}
|
||||
@@ -20,6 +20,7 @@ set(CONV2D_FWD_XDL_BIAS_RELU_SOURCE 5_conv2d_fwd_xdl_bias_relu/conv2d_fwd_xdl_bi
|
||||
set(CONV2D_FWD_XDL_BIAS_RELU_ADD_SOURCE 6_conv2d_fwd_xdl_bias_relu_add/conv2d_fwd_xdl_bias_relu_add.cpp)
|
||||
set(CONV2D_FWD_XDL_BIAS_RELU_ATOMIC_ADD_SOURCE 7_conv2d_fwd_xdl_bias_relu_atomic_add/conv2d_fwd_xdl_bias_relu_atomic_add.cpp)
|
||||
set(GEMM_XDL_ALPHA_BETA_SOURCE 8_gemm_xdl_alpha_beta/gemm_xdl_alpha_beta.cpp)
|
||||
set(CONV2D_FWD_XDL_INT8_SOURCE 9_conv2d_fwd_xdl_int8/conv2d_fwd_xdl_int8.cpp)
|
||||
|
||||
add_executable(gemm_xdl ${GEMM_XDL_SOURCE})
|
||||
add_executable(gemm_xdl_bias_relu ${GEMM_XDL_BIAS_RELU_SOURCE})
|
||||
@@ -29,6 +30,7 @@ add_executable(conv2d_fwd_xdl_bias_relu ${CONV2D_FWD_XDL_BIAS_RELU_SOURCE})
|
||||
add_executable(conv2d_fwd_xdl_bias_relu_add ${CONV2D_FWD_XDL_BIAS_RELU_ADD_SOURCE})
|
||||
add_executable(conv2d_fwd_xdl_bias_relu_atomic_add ${CONV2D_FWD_XDL_BIAS_RELU_ATOMIC_ADD_SOURCE})
|
||||
add_executable(gemm_xdl_alpha_beta ${GEMM_XDL_ALPHA_BETA_SOURCE})
|
||||
add_executable(conv2d_fwd_xdl_int8 ${CONV2D_FWD_XDL_INT8_SOURCE})
|
||||
|
||||
target_link_libraries(gemm_xdl PRIVATE host_tensor)
|
||||
target_link_libraries(gemm_xdl_bias_relu PRIVATE host_tensor)
|
||||
@@ -38,3 +40,4 @@ target_link_libraries(conv2d_fwd_xdl_bias_relu PRIVATE host_tensor)
|
||||
target_link_libraries(conv2d_fwd_xdl_bias_relu_add PRIVATE host_tensor)
|
||||
target_link_libraries(conv2d_fwd_xdl_bias_relu_atomic_add PRIVATE host_tensor)
|
||||
target_link_libraries(gemm_xdl_alpha_beta PRIVATE host_tensor)
|
||||
target_link_libraries(conv2d_fwd_xdl_int8 PRIVATE host_tensor)
|
||||
|
||||
@@ -21,7 +21,7 @@ void host_conv_nchw_kcyx_nkhw(const Tensor<TIn>& in,
|
||||
constexpr auto I1 = ck::Number<1>{};
|
||||
|
||||
auto f_nchw = [&](auto n, auto k, auto ho, auto wo) {
|
||||
double v = 0;
|
||||
float v = 0;
|
||||
for(int c = 0; c < wei.mDesc.GetLengths()[1]; ++c)
|
||||
{
|
||||
for(int y = 0; y < wei.mDesc.GetLengths()[2]; ++y)
|
||||
@@ -33,13 +33,13 @@ void host_conv_nchw_kcyx_nkhw(const Tensor<TIn>& in,
|
||||
if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
|
||||
wi < in.mDesc.GetLengths()[3])
|
||||
{
|
||||
v += static_cast<const double>(in(n, c, hi, wi)) *
|
||||
static_cast<const double>(wei(k, c, y, x));
|
||||
v += ck::type_convert<float>(in(n, c, hi, wi)) *
|
||||
ck::type_convert<float>(wei(k, c, y, x));
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
out(n, k, ho, wo) = v;
|
||||
out(n, k, ho, wo) = ck::type_convert<TOut>(v);
|
||||
};
|
||||
|
||||
make_ParallelTensorFunctor(f_nchw,
|
||||
|
||||
@@ -25,6 +25,9 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(std::vector<DeviceCo
|
||||
void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
|
||||
std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
@@ -171,6 +174,20 @@ void profile_conv_fwd_impl(int do_verification,
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ushort> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ushort> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, ushort>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, int8_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
|
||||
}
|
||||
|
||||
if(conv_ptrs.size() <= 0)
|
||||
{
|
||||
|
||||
@@ -8,8 +8,10 @@
|
||||
|
||||
enum ConvDataType
|
||||
{
|
||||
F32_F32_F32, // 0
|
||||
F16_F16_F16, // 1
|
||||
F32_F32_F32, // 0
|
||||
F16_F16_F16, // 1
|
||||
BF16_BF16_BF16, // 2
|
||||
INT8_INT8_INT8, // 3
|
||||
};
|
||||
|
||||
enum ConvInputLayout
|
||||
@@ -130,6 +132,56 @@ int profile_conv_fwd(int argc, char* argv[])
|
||||
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
|
||||
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
|
||||
}
|
||||
else if(data_type == ConvDataType::BF16_BF16_BF16 && in_layout == ConvInputLayout::NHWC &&
|
||||
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
|
||||
{
|
||||
ck::profiler::profile_conv_fwd_impl<2,
|
||||
uint16_t,
|
||||
uint16_t,
|
||||
uint16_t,
|
||||
ck::tensor_layout::convolution::NHWC,
|
||||
ck::tensor_layout::convolution::KYXC,
|
||||
ck::tensor_layout::convolution::NHWK>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
N,
|
||||
K,
|
||||
C,
|
||||
std::vector<ck::index_t>{Hi, Wi},
|
||||
std::vector<ck::index_t>{Y, X},
|
||||
std::vector<ck::index_t>{Ho, Wo},
|
||||
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
|
||||
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
|
||||
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
|
||||
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
|
||||
}
|
||||
else if(data_type == ConvDataType::INT8_INT8_INT8 && in_layout == ConvInputLayout::NHWC &&
|
||||
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
|
||||
{
|
||||
ck::profiler::profile_conv_fwd_impl<2,
|
||||
int8_t,
|
||||
int8_t,
|
||||
int8_t,
|
||||
ck::tensor_layout::convolution::NHWC,
|
||||
ck::tensor_layout::convolution::KYXC,
|
||||
ck::tensor_layout::convolution::NHWK>(
|
||||
do_verification,
|
||||
init_method,
|
||||
do_log,
|
||||
nrepeat,
|
||||
N,
|
||||
K,
|
||||
C,
|
||||
std::vector<ck::index_t>{Hi, Wi},
|
||||
std::vector<ck::index_t>{Y, X},
|
||||
std::vector<ck::index_t>{Ho, Wo},
|
||||
std::vector<ck::index_t>{conv_stride_h, conv_stride_w},
|
||||
std::vector<ck::index_t>{conv_dilation_h, conv_dilation_w},
|
||||
std::vector<ck::index_t>{in_left_pad_h, in_left_pad_w},
|
||||
std::vector<ck::index_t>{in_right_pad_h, in_right_pad_w});
|
||||
}
|
||||
else
|
||||
{
|
||||
throw std::runtime_error("wrong! this Conv data_type & layout is not implemented");
|
||||
|
||||
@@ -86,10 +86,9 @@ struct ReferenceConvFwd : public device::BaseOperator
|
||||
float v_wei;
|
||||
|
||||
arg.in_element_op_(
|
||||
v_in,
|
||||
static_cast<const float>(arg.in_n_c_hi_wi_(n, c, hi, wi)));
|
||||
v_in, ck::type_convert<float>(arg.in_n_c_hi_wi_(n, c, hi, wi)));
|
||||
arg.wei_element_op_(
|
||||
v_wei, static_cast<const float>(arg.wei_k_c_y_x_(k, c, y, x)));
|
||||
v_wei, ck::type_convert<float>(arg.wei_k_c_y_x_(k, c, y, x)));
|
||||
|
||||
v_acc += v_in * v_wei;
|
||||
}
|
||||
@@ -101,7 +100,7 @@ struct ReferenceConvFwd : public device::BaseOperator
|
||||
|
||||
arg.out_element_op_(v_out, v_acc);
|
||||
|
||||
arg.out_n_k_ho_wo_(n, k, ho, wo) = v_out;
|
||||
arg.out_n_k_ho_wo_(n, k, ho, wo) = ck::type_convert<OutDataType>(v_out);
|
||||
};
|
||||
|
||||
make_ParallelTensorFunctor(f_nchw,
|
||||
|
||||
@@ -9,6 +9,7 @@ include_directories(BEFORE
|
||||
${PROJECT_SOURCE_DIR}/composable_kernel/include/tensor_operation
|
||||
${PROJECT_SOURCE_DIR}/composable_kernel/include/problem_transform
|
||||
${PROJECT_SOURCE_DIR}/external/rocm/include
|
||||
${PROJECT_SOURCE_DIR}/reference_operation/include
|
||||
)
|
||||
|
||||
# test_magic_number_division
|
||||
@@ -16,8 +17,16 @@ set(MAGIC_NUMBER_DIVISISON_SOURCE magic_number_division/main.cpp)
|
||||
add_executable(test_magic_number_division ${MAGIC_NUMBER_DIVISISON_SOURCE})
|
||||
target_link_libraries(test_magic_number_division PRIVATE host_tensor)
|
||||
|
||||
|
||||
set(CONV2D_FWD_SOURCE conv2d_fwd/main.cpp)
|
||||
|
||||
add_executable(test_conv2d_fwd ${CONV2D_FWD_SOURCE})
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE host_tensor)
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE device_conv2d_fwd_instance)
|
||||
|
||||
# test_split_k
|
||||
set(SPLIT_K_SOURCE split_k/main.cpp)
|
||||
add_executable(test_split_k ${SPLIT_K_SOURCE})
|
||||
target_link_libraries(test_split_k PRIVATE host_tensor)
|
||||
target_link_libraries(test_split_k PRIVATE device_gemm_instance)
|
||||
|
||||
|
||||
307
test/conv2d_fwd/main.cpp
Normal file
307
test/conv2d_fwd/main.cpp
Normal file
@@ -0,0 +1,307 @@
|
||||
#include "config.hpp"
|
||||
#include "device.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "host_conv.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_conv_fwd.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_conv_fwd.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv2d_fwd_instance {
|
||||
|
||||
using DeviceConvFwdNoOpPtr = DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
|
||||
std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
template <typename T>
|
||||
static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
|
||||
{
|
||||
float max_diff = 1e-6;
|
||||
|
||||
for(int i = 0; i < ref.mData.size(); ++i)
|
||||
{
|
||||
float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
|
||||
if(max_diff < diff)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
int main(int argc, char* argv[])
|
||||
{
|
||||
int data_type = 0;
|
||||
int init_method = 0;
|
||||
|
||||
// Conv shape
|
||||
ck::index_t N = 128;
|
||||
ck::index_t K = 256;
|
||||
ck::index_t C = 192;
|
||||
ck::index_t Y = 3;
|
||||
ck::index_t X = 3;
|
||||
ck::index_t Hi = 71;
|
||||
ck::index_t Wi = 71;
|
||||
ck::index_t conv_stride_h = 2;
|
||||
ck::index_t conv_stride_w = 2;
|
||||
ck::index_t conv_dilation_h = 1;
|
||||
ck::index_t conv_dilation_w = 1;
|
||||
ck::index_t in_left_pad_h = 1;
|
||||
ck::index_t in_left_pad_w = 1;
|
||||
ck::index_t in_right_pad_h = 1;
|
||||
ck::index_t in_right_pad_w = 1;
|
||||
|
||||
if(argc == 3)
|
||||
{
|
||||
data_type = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
}
|
||||
else if(argc == 18)
|
||||
{
|
||||
data_type = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
|
||||
N = std::stoi(argv[3]);
|
||||
K = std::stoi(argv[4]);
|
||||
C = std::stoi(argv[5]);
|
||||
Y = std::stoi(argv[6]);
|
||||
X = std::stoi(argv[7]);
|
||||
Hi = std::stoi(argv[8]);
|
||||
Wi = std::stoi(argv[9]);
|
||||
conv_stride_h = std::stoi(argv[10]);
|
||||
conv_stride_w = std::stoi(argv[11]);
|
||||
conv_dilation_h = std::stoi(argv[12]);
|
||||
conv_dilation_w = std::stoi(argv[13]);
|
||||
in_left_pad_h = std::stoi(argv[14]);
|
||||
in_left_pad_w = std::stoi(argv[15]);
|
||||
in_right_pad_h = std::stoi(argv[16]);
|
||||
in_right_pad_w = std::stoi(argv[17]);
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("arg1: verification (0=no, 1=yes)\n");
|
||||
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
|
||||
printf("arg3: run kernel # of times (>1)\n");
|
||||
printf("arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
|
||||
"RightPx\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
auto Run = [&](auto input_type, auto wei_type, auto out_type) {
|
||||
using InDataType = decltype(input_type);
|
||||
using WeiDataType = decltype(wei_type);
|
||||
using OutDataType = decltype(out_type);
|
||||
|
||||
using ReferenceConvFwdInstance = ck::tensor_operation::host::ReferenceConvFwd<InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp>;
|
||||
|
||||
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
|
||||
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
|
||||
|
||||
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
|
||||
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
|
||||
|
||||
const std::vector<ck::index_t> input_spatial_lengths{{Hi, Wi}};
|
||||
const std::vector<ck::index_t> filter_spatial_lengths{{Y, X}};
|
||||
const std::vector<ck::index_t> output_spatial_lengths{{Ho, Wo}};
|
||||
const std::vector<ck::index_t> conv_filter_strides{{conv_stride_h, conv_stride_w}};
|
||||
const std::vector<ck::index_t> conv_filter_dilations{{conv_dilation_h, conv_dilation_w}};
|
||||
const std::vector<ck::index_t> input_left_pads{{in_left_pad_h, in_left_pad_w}};
|
||||
const std::vector<ck::index_t> input_right_pads{{in_right_pad_h, in_right_pad_w}};
|
||||
|
||||
auto f_host_tensor_descriptor =
|
||||
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
|
||||
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
|
||||
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
|
||||
};
|
||||
|
||||
Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
|
||||
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_host_result(f_host_tensor_descriptor(N, K, Ho, Wo));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_device_result(f_host_tensor_descriptor(N, K, Ho, Wo));
|
||||
|
||||
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
|
||||
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
|
||||
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo_host_result.mDesc << std::endl;
|
||||
|
||||
switch(init_method)
|
||||
{
|
||||
case 0: break;
|
||||
case 1:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
|
||||
break;
|
||||
default:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{0, 1});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1, 1});
|
||||
}
|
||||
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) *
|
||||
out_n_k_ho_wo_device_result.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
|
||||
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>;
|
||||
|
||||
// add device Conv instances
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
|
||||
if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, float> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, float> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, float>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ck::half_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ck::half_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, ck::half_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ushort> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ushort> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, ushort>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, int8_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
|
||||
}
|
||||
|
||||
if(conv_ptrs.size() <= 0)
|
||||
{
|
||||
throw std::runtime_error("wrong! no device Conv instance found");
|
||||
}
|
||||
|
||||
auto ref_conv = ReferenceConvFwdInstance{};
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
|
||||
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi,
|
||||
wei_k_c_y_x,
|
||||
out_n_k_ho_wo_host_result,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
|
||||
// profile device Conv instances
|
||||
bool success = false;
|
||||
for(auto& conv_ptr : conv_ptrs)
|
||||
{
|
||||
auto argument_ptr = conv_ptr->MakeArgumentPointer(
|
||||
static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
N,
|
||||
K,
|
||||
C,
|
||||
input_spatial_lengths,
|
||||
filter_spatial_lengths,
|
||||
output_spatial_lengths,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
PassThrough{},
|
||||
PassThrough{},
|
||||
PassThrough{});
|
||||
|
||||
auto invoker_ptr = conv_ptr->MakeInvokerPointer();
|
||||
|
||||
if(conv_ptr->IsSupportedArgument(argument_ptr.get()))
|
||||
{
|
||||
invoker_ptr->Run(argument_ptr.get(), 0);
|
||||
|
||||
out_device_buf.FromDevice(out_n_k_ho_wo_device_result.mData.data());
|
||||
if(!check_out(out_n_k_ho_wo_host_result, out_n_k_ho_wo_device_result))
|
||||
{
|
||||
success = false;
|
||||
break;
|
||||
}
|
||||
success = true;
|
||||
}
|
||||
}
|
||||
|
||||
if(success)
|
||||
{
|
||||
std::cout << "test conv2d fwd : Pass" << std::endl;
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cout << "test conv2d fwd: Fail " << std::endl;
|
||||
}
|
||||
};
|
||||
|
||||
if(data_type == 0)
|
||||
{
|
||||
Run(float(), float(), float());
|
||||
}
|
||||
else if(data_type == 1)
|
||||
{
|
||||
Run(ck::half_t(), ck::half_t(), ck::half_t());
|
||||
}
|
||||
else if(data_type == 2)
|
||||
{
|
||||
Run(ushort(), ushort(), ushort());
|
||||
}
|
||||
else if(data_type == 3)
|
||||
{
|
||||
Run(int8_t(), int8_t(), int8_t());
|
||||
}
|
||||
else
|
||||
{
|
||||
return 1;
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
Reference in New Issue
Block a user